Decision-making technology based on knowledge engineering and experiment on the intelligent water-fertilizer irrigation system

Abstract

Although the irrigation technologies based on the Decision-making System (DMS) began in the late 1990s, while being merely embryonic from laboratory research into application in the agricultural irrigation areas, DMS based on intelligent algorithms have drawn much attention from the academia over the recent years. In this study, we have provided an overview of the decision-making technology based on knowledge engineering for intelligent irrigation system referred to as Knowledge-based Engineering (KBE). As the modern technical research and scientific theory on agricultural water saving is further developed, the water-fertilizer irrigation is becoming increasingly intelligent. We have put forward the concept of KBE intelligent irrigation system and its support to decision-making in the study, while adopting the techniques and methods of knowledge engineering. In addition, we have combined our research findings with the expert knowledge on the water-fertilizer irrigation in a system integrated with computer network, intelligent reasoning and artificial intelligence (AI), among other modern high-techs. We have set up the decision-making models and analytical methods of irrigation and fertilization for KBE by referring to the expert experience and data of fertilization. Moreover, we have taken into account the web crawler technology in irrigation and fertilization, and we have put forward novel methods of knowledge acquisition based on the web crawler. Correspondingly, we have established the knowledge base for the decision-making support system tailored to irrigation and fertilization.

The experiment result shows that the recommended irrigation quota is compared with local cultivation technology experience to obtain a decision accuracy of 81.7%. And the water and fertilizer management plan obtained by the intelligent decision-making system has a thicker stem and higher plant height during the growth period than the crops obtained by local cultivation experience. The output of the decision-making system is 620 kg, which a relative increase of 5.08% is compared with the 590 kg obtained from local cultivation experience.

Keywords

Knowledge-based engineering decision-making support intelligent irrigation artificial intelligence

1. Introduction

Over the recent years, China has attached great importance to developing the technology of integrating water with fertilizer. Given that this technology serves as the Top 1 technology of modern agriculture, the promotion target in 2020 is 150 million mu [1]. From the aspect of developing modern agriculture and applying the technologies of water and fertilizer while integrating the technology they would hold key to reducing the application of fertilizer and enhancing the level of efficiency, thus providing an effective way to the achieve sustainable socio-economic development. In terms of the core technologies of irrigation decision-making for the fertilizer integration system, except for several highly structured issues easy to be expressed by mathematical models, we are unable to express the system quantitatively. Furthermore, there are models which feature high dimensionality and large amount of calculation [2]. The research constitutes an emerging field of research integrated with decision-making on crop water demand, so as to develop a support system for decision-making. Through the mechanism of self-learning and self-reasoning, we are able to ensure that the system could cope with intricate and highly nonlinear issues arisen from the irrigation decision-making, while enhancing its reliability. In recent years, researchers have conducted related studies in the aspect. For instance, George et al. [3] has developed a support system for decision-making by referring to the crop evapotranspiration prediction. This system consists of a variety of prediction models, enabling users to select the most suitable one based on the type of data input. By adopting the calculation method, we are able to forecast the evapotranspiration of crops and enhance the accuracy of prediction, indicating the significance of this method for determining water requirements for crops. Giusti et al. has set up the fuzzy decision support system by introducing the fuzzy theory to enhance the scientific nature of irrigation water [4]. As a result, the decision-making on irrigation water would feature the functions of both qualitative knowledge reasoning and the quantitative knowledge calculation functions, thereby enabling decision makers to address the agricultural semi-structured and unstructured issues. Proper methods of irrigation could enhance the accuracy of crop irrigation. In 2017, Knapp and Huang [5]have studied numerous factors of climatic conditions and the impact imposed by extreme climatic conditions as well as irrigation decision-making. Moreover, they have elaborated on the role of climate factors in influencing producers’ decision-making on irrigation, and identified the average climatic conditions and extreme climatic events. It is found that the producers’ decision-making on irrigation has a stronger predictive capacity than the coefficient of variation, providing guidance for our analysis on the forecast of the irrigation decision-making system.

In general, the intelligent technologies of irrigation decision-making have been rapidly applied in the irrigation sector. Both domestic and international scholars have conducted extensive research on intelligent technologies of support to irrigation decision-making, and have achieved fruitful and in-depth results. However, although the intelligent systems for irrigation decision-making have been designed and developed, the majority of such systems are still undergoing the research and experimental stage, where most of the work is to research the irrigation control system and the decision system separately. Furthermore, we need to step up efforts on the collaboration of decision-making and control. Researchers have carried out relevant studies on the application of the models of crop growth, neural networks, fuzzy clustering and knowledge-based engineering (KBE) in the sector of water-saving irrigation and fertilization. Through the studies, researchers intend to address the issues of decision-making on irrigation and fertilization schemes during the irrigation management from varying aspects, so as to tackle the existing fitting issues such as poor level of accuracy, semi-structured system, model portability and lack of consideration for the comprehensive economic benefits. Therefore, by conducting the research on the KBE intelligent technologies of support to irrigation decision-making, we intend to ensure self-learning functions, real-time update and reproduction of expert information related to decision-making as well as fine irrigation of crops based on water and fertilizer.

2. Design principles based on Knowledge-based Engineering

The Knowledge-based Engineering (KBE) involves the use of principles and methods of artificial intelligence (AI) to provide solutions to application issues that require expert knowledge [6]. It is a vital technical issue to properly leverage the composition and interpretation of the acquisition, expression, and reasoning procedures of expert knowledge to design a knowledge system. The KBE approach, which is based on knowledge, mainly focuses on ways of illustrating professional domain knowledge by computer and automatically coping with issues through knowledge reasoning. Major technologies related to KBE is shown in Fig. 1.

Figure 1.

Knowledge engineering.

2.1 Knowledge acquisition

Knowledge acquisition refers to the procedure of summarizing and abstracting the expert knowledge used to tackle issues arisen from the agricultural and water engineering sector from a certain knowledge source (including information such as irrigation database, expert opinions and irrigation text), while converting such information into knowledge in a computer system of knowledge base [7]. Moreover, we may divide the methods into manual, semi-automatic and automatic approaches to knowledge acquisition. Specifically, we have mainly adopted the manual and semi-automatic approaches to knowledge acquisition, so as to obtain massive amount of professional knowledge from visiting experts in the field. In general, the conventional expert systems rely on the method of obtaining expert knowledge, though its efficiency is rather limited [8]. As the computer and AI-related technologies are further developed, an automatic scientific modeling procedure has gradually been formed for the knowledge acquisition. Common approaches to knowledge acquisition include but not limited to machine learning, data mining and neural networks.

In general, the knowledge acquisition can be divided into four steps [7], namely, identification of problems and extraction of features, acquisition of concepts and relationships, structured representation of knowledge, and formation of knowledge base, as illustrated in Fig. 2.

Figure 2.

Steps of knowledge acquisition.

Web pages and texts have gradually become a critical treasure house of resources in terms of knowledge acquisition in the agricultural sector. In the agricultural sector, the acquisition of web content is embodied in the extraction of effective information online. The information available on the web pages is semi-structured in essence. The web crawler runs through the designated sites in search of relevant information. Moreover, with the use of web semantic technology, the web crawler is able to extract knowledge from the originally unstructured web texts in the form of HTML or XML, so as to establish a domain knowledge base. The specific results of knowledge acquisition are illustrated in Fig. 3.

Figure 3.

Knowledge acquisition.

2.2 Knowledge representation

Knowledge representation refers to the use of symbols and methods that computers are able to recognize, accept, and process, thereby expressing the knowledge acquired by humankind in the objective world [9]. By leveraging the knowledge representation, we mainly aim to identify the mapping relationship between knowledge and representation. Common approaches primarily include but not limited to methods of knowledge representation based on production rules, case-based methods of knowledge representation as well as object-oriented methods of knowledge representation [10].

The production rules were initially put forward by the American mathematician E.Post in 1943, and were further developed by A. Newell and H.A.Simon. These production rules are most suitable for representing the causal relationships among knowledge, and thus they have found a wide range of applications. Furthermore, they are more suitable for empirical sectors such as the agricultural industry [11]. These rules are typically of an IF-THEN structure with conditional sentences to express the results of a certain action under specific circumstances. The general formula is specified as follows:

IF condition (1) & condition (2) & $\ldots$ & condition (i) THEN conclusion (1) || conclusion (2) || $\ldots$ conclusion (i), where condition (i) refers to the precondition, and conclusion (i) represents the specific conclusion.

Rule 1 in the knowledge base for the corn irrigation system can be expressed as follows:

IF crops $=$ “spring corn” and type $=$ “ground irrigation” and year $=$ “dry year” and period $=$ “joining” THEN irrigation $=$ “50 m ${}^{3}$ /acreage”.

Rule 2 can be expressed as follows:

IF crops $=$ “summer corn” and type $=$ “ground irrigation” and year $=$ “general year” and period $=$ “grouting” THEN irrigation $=$ “50m ${}^{3}$ /acreage”.

2.3 Knowledge reasoning

Based on the known facts, the knowledge reasoning refers to the process of using the knowledge already mastered to identify the facts contained therein or to infer new facts [12]. In general, the reasoning process includes two sorts of judgments: The first type is defined as the known judgments; he other one is categorized as the new judgments and conclusions inferred from existing knowledge. The process of applying the varying rules of inference rules in the decision-making support system is illustrated in Fig. 4 as follows.

Figure 4.

Knowledge inference.

The rule-based reasoning is the primary method adopted in this study. This method refers to the problem-based reasoning approach based on the irrigation knowledge derived from production rules, and deductive reasoning technologies hold key to the approach [13]. The process of reasoning is specified as follows: In consistence with the prerequisites and based on a certain approach to searching, we have gradually searched the matching rules until we have met all relevant conditions. Subsequently, we have obtained the solution to the specific issue by drawing insights from the conclusions reached by satisfying each rule configured in the knowledge base. Take the selection of the method of irrigation as an example. In the study, we have taken the flat terrain as the topography and landform, clay loam as the soil texture, and well water as the water resources, while assuming the wind speed as 2 m/s, among other conditions. Correspondingly, we have gradually searched for the rules capable of meeting all conditions, and we have adopted the irrigation method of sprinkler irrigation. The confidence level amounts to 0.85 [14].

2.4 Irrigation methods for knowledge management

Knowledge management constitutes a vital part of the intelligent system of support to decision-making. For an intelligent system of support to irrigation decision-making, we must enhance its efficiency by ensuring efficient integration of knowledge and establishment of management mechanisms. The general framework of knowledge management is illustrated in Fig. 5. It can be noted that the management and application of knowledge involves the reasoning, description and processing of knowledge [15]. Knowledge management has organically integrated both explicit and implicit knowledge of irrigation, facilitating better use of the two sorts of knowledge and effectively supporting the problem-solving process related to irrigation.

Figure 5.

Diagram on the knowledge management framework.

2.5 Visual analysis on the data of irrigation decision-making

Knowledge visualization refers to the approach of applying vision and image to represent complex and abstract knowledge. Factors of presentation include designers, learners, knowledge content and visual representation [16]. The relationship among these elements is illustrated in Fig. 6, which determines the structure of data and form of knowledge visualization. Furthermore, through mechanisms of feedback and correction, machines are able to attain and create knowledge. This is one of the orientations of applying knowledge engineering in the irrigation sector.

Figure 6.

Relationship among elements of visual representation for knowledge visualization.

3. Irrigation decision system based on the KBE approach

In the 1970s, the Decision Support System was initially put forward by Michael S. Scott Morton and Peter of the Massachusetts Institute of Technology (MIT) in the U.S. [17]. Later in 1980, Sprague [18] proposed a framework of decision support system based on two databases. This system is deemed to be an integrated structure composed of three components, namely, user interface, database management system (DBMS), and model base management system (MBMS).

In this study, we have proposed the system of decision-making on irrigation based on the KBE approach, which is a methodology of design oriented towards the entire process of technological development. In addition, this system is based on such sectors as agriculture, water conservancy, astronomy, and AI. In the bottom layer, we have used a data management system and a platform of the Internet of Things (IoT), so as to express and build knowledge. During all stages of the technological development for water and fertilizer irrigation engineering, we have utilized tool sets such as modeling, mining, propagation, reasoning, integration and management, enabling the decision-making system to meet the needs of precise water and fertilizer irrigation and crop growth on demand. The scheme of the system design is illustrated in Fig. 7.

Figure 7.

General design of system.

3.1 Database design

The database constitutes one of the fundamental components of the intelligent decision support system while featuring storage of all the basic data and facts related to the system. The database mainly plays the role of managing relevant data, providing data for both the knowledge base and the model base, and conducting model calculations. Given that the database functions by providing or storing the original data, intermediate results as well as final results of the calculation, the database is not only composed of the data files necessary for the model, but also the files related to the results of model running. In addition, the database also enables query and update of the data, thereby providing statistics for the decision-making in a timely and holistic manner. Moreover, it is also necessary to set up the decision support system for irrigation based on the KBE approach, so as to include the varying data in the database for collection, quality analysis and decision analysis of data. In addition, the design of the database module is rather independent, and is capable of conducting operations such as query, addition, deletion and sorting [19].

We have opted for SQLite as the tool of designing and developing database for the system. Given that the SQlite database features high velocity, small size, low cost, and optimal portability, it is adopted as the option of developing database for the system. The module of database mainly stores data of varying sorts necessary for the intelligent decision-making on water and fertilizer, including but not limited to information of meteorology, crops, geography, and irrigation.

Specifically, the contents of each database are set out as follows:

(1)
Weather database. This database primarily comprises meteorological data of numerous regions, including the maximum, minimum and average temperature, relative humidity, average wind speed and sunshine hours.
(2)
Crop database. This database mainly serves the purpose of preserving the growth period of crops, crop coefficients, key points of cultivation, quality characteristics and drought resistance, etc.
(3)
Geographic database. This database primarily helps to save the latitude and longitude of an area, its altitude, topography, soil characteristics and water source information, etc.
(4)
Irrigation database. This database mainly comprises the temperature and humidity of soil, rainfall, the amount, time and method of irrigation, the ratio of concentration of water and fertilizer, etc. in the irrigation area.

3.2 Model base design

Model base refers to the computer system that functions by enabling storage of model and mode of representation. In essence, the model is equivalent to the abstraction and simulation of the objective laws in a nutshell, which are summarized according to a huge quantity of professional studies on the described objects and processes. It serves as a link between the application system and the specific professional sector, and is regarded as a tool for holistic analysis, processing as well as utilization of data [20]. In addition, the model base helps users gain access to the characteristic information about attributes of the model, which mainly include the variable data on the model, description of both the input and output, functions available in the model, mathematical equation of the model and interpretation of the solutions. Such information enables users to gain insights into information related to the model as well as its input and output parameters. By properly adopting the model, we are able to draw accurate conclusions on the results calculated through the model.

The model base is designed according to the key components of the KBE system of support to intelligent irrigation decision-making. There are two major categories: One category comprises the classic empirical models, which are primarily represented by empirical formulas and mathematical models developed based on agricultural experiments, production and applications; he other category consists of models of mechanisms, which primarily commence from the relationship among crops, fertilization, environment and measures. For this type of models, researchers have comprehensively taken into account such factors as crop water and fertilizer requirements, rainfall and evapotranspiration, so as, to facilitate the establishment of models [21]. The model base mainly consists of the ET0 model, the ET model, the rainfall model, the water balance model as well as the water and fertilizer coupling model. The main structure of the model base is illustrated in Fig. 8.

Figure 8.

Structure of model base.

3.2.1 ET0 model

The ET0 model base is composed of the Penman-Monteith formula, the Hargreaves formula, the Priestley-Taylor formula, the support vector machine model and the random forest model. As for the Penman-Monteith formula, the Hargreaves formula, the Priestley-Taylor formula, their details are specified as the following Eqs (1)–(3), respectively.

$\displaystyle\textit{ET}_{0}=\frac{0.408\times\Delta\times(R_{n}-G)+\gamma% \times\frac{900}{273+T}\times U{}_{2}\times(e_{a}-e_{d})}{\Delta+\gamma\times(% 1+0.34\times U_{2})}$ (1) $\displaystyle\textit{ET}_{0}=0.0023\frac{\textit{Ra}}{\lambda}(T+17.8)\sqrt{T_% {\max}-T_{\min}}$ (2) $\displaystyle\textit{ET}_{0}=\alpha\times\frac{\Delta}{\Delta+\gamma}\times(R_% {n}-G)$ (3)

Where: $\textit{ET}_{0}$ refers to the evapotranspiration of reference crops, measured in mm/d; $R_{n}$ refers to Net radiation on the surface of the crop, measured in MJ/m ${}^{2}$ .d; $G$ refers to Soil heat flux, measured in MJ/m ${}^{2}$ .d; $T$ refers to Average daily humidity at a height of 2 meters, measured in ${{}^{\circ}}$ C; $U_{2}$ refers to Wind speed at a height of 2 meters, measured in m/s; $e_{a}$ refers to saturated vapor pressure, measured in kPa; $e_{d}$ refers to actual water vapor pressure, measured in kPa; $e_{a}-e_{d}$ refers to saturated vapor pressure difference, measured in kPa; $\vartriangle$ refers to Inclination of saturated vapor pressure curve; $\gamma$ refers to Hygrometer constant, measured in kPa/ ${{}^{\circ}}$ C; $\lambda$ refers to Latent heat of vaporization, measured in MJ/kg.

All parameters in the aforementioned formulas can be determined by the following eight parameters (namely, average temperature, daily minimum temperature, daily maximum temperature, latitude, humidity, wind speed, sunshine hours and altitude). Data related to the parameters are extracted from China Meteorological Station in the system, whereas the training data ranges between 1990 and 2017. The website link is: http://openweather.weather.com.cn/.

Eight basic meteorological parameters are used both in the support vector machines and the random forest models. These parameters include: maximum temperature Tmax, average temperature $T_{m}$ , minimum temperature Tmin, actual sunshine time $n$ , wind speed $U_{h}$ , average relative humidity RHm, latitude w, and altitude $z$ , which serve as the input of the $\textit{ET}_{0}$ model Parameter. We have taken the $\textit{ET}_{0}$ result calculated through the FAO-56 PM (Penman-Montieth) formula as the expected value of output for the establishment of the model. During the modeling, we have integrated varying approaches to meteorological parameter input and leveraged the data mining technologies. In addition, we have adopted the support vector machine algorithm as well as the random forest algorithm to set up the model and to forecast the $\textit{ET}_{0}$ .

The random forest model is illustrated in Fig. 9. In the model, we have taken the meteorological data and the results of calculation of the FAO56 Penman-Monteith formula as the training set. It is trained for numerous times so that we can continuously adjust the parameters, so as to obtain the optimal model. Subsequently, we may make forecasts on the crop evapotranspiration based on the inputed 8 basic meteorological parameters.

Figure 9.

Random forest model.

3.2.2 ET model

There are two methods for calculating the demands of crops for water in the system. The first method is the direct approach to calculating the water demand:

(1) The method of calculating water demands is the a-value method with water surface evaporation taken as a parameter, and the formula is specified as follows:

$\displaystyle\textit{ET}=\alpha\times E_{0}$ (4)

Where: ET refers to the demands of crops for water throughout a certain period of time, measured in mm as the depth of water layer; $E_{0}$ refers to the water surface evaporation with ET at the same period of time, measured as the depth of water layer.

(2) The method of calculating the value of water demands (the $K$ method) with output taken as a parameter, and the formula is specified as follows:

$\displaystyle\textit{ET}=K\times Y^{n}+c$ (5)

Where: ET refers to the total water demands (measured in mm or m ${}^{3}$ /hm ${}^{2}$ ) during the entire period of growth of the crops; $Y$ refers to the yield of crops per unit area (kg/hm ${}^{2}$ ); $K$ refers to the water demands per unit yield (m ${}^{3}$ /kg). $K$ , $n$ and $c$ are determined through experiments.

The second method is the method of calculating the water demands based on the evapotranspiration of reference crop. In terms of the reference crop $\textit{ET}_{0}$ , we have merely taken into account the impact imposed by meteorological factors as well as the demands of crops for water. However, we need to properly adjust the actual water demands of crop by taking into account the crop and soil-related factors. The formula adopted in the model is specified as follows:

$\displaystyle\textit{ET}=K_{c}\times\textit{ET}_{0}$ (6)

Where: $\textit{ET}_{0}$ refers to the evapotranspiration of reference crops, measured in mm/d; $K_{c}$ refers to the comprehensive crop coefficient; whereas ET refers to the actual demands of crops for water, measured in mm/d.

3.2.3 Rainfall model

The rainfall model base consists of the models on recursive neural networks, wavelet neural networks and ARIMA models, which are mainly based on the data related to the observed historical rainfall statistic. We have adopted AI algorithms to set up the models of forecasting. In case no relevant theories or empirical formulas are applicable to calculation and prediction, we can adopt the knowledge-based inference models and conduct the qualitative analysis. We have adopted the model on recursive neural networks to set up a model with a recurrent neural network, which features both long and short-term memory units. Moreover, we have updated the network weight by adopting the algorithm of backpropagation through time, so as to address the issue of long-term forecasting of rainfall and to cope with the problem of high-dimensionality, nonlinearity and local minima. The question lies in the forecast of rainfall for the next day based on the historical rainfall data of the previous seven days.

3.2.4 Water balance model

In this model, we have adopted the principle of the water balance for farmland [20], namely:

$\displaystyle I+P+G=\textit{ET}+D+R\pm\textit{ASW}$ (7)

Where: $I$ refers to the amount of irrigation water in the period $T$ ; $P$ refers to the precipitation in the period $T$ ; $G$ refers to the recharge of groundwater in the period $T$ ; ET refers to the evapotranspiration of the reference crop in the period $T$ ; $D$ refers to the deep infiltration; $R$ refers to the surface runoff; ASW refers to the effective storage of soil water at the beginning of the period. During the modeling procedure, we have simplified the balance equation in this paper, while omitting $G$ , $D$ and $R$ , and the formula is specified as follows:

$\displaystyle I+P=\textit{ET}\pm\textit{ASW}$ (8)

Therefore, the amount of irrigation can be expressed as: $I=\textit{ET}\pm\textit{ASW}-P$ .

3.2.5 Model of decision-making on the fertilization amount based on the online support vector regression

During the decision-making procedure on irrigation and fertilization, it is necessary to set up the model of decision-making, so as to determine the amount of fertilization in an intelligent manner. While conducting the irrigation, we have adopted methods of the nutritional diagnosis, the target yield and the fertilizer effect function, which constitute the three major models of decision-making on fertilization.

While training, it is easy for the model to reach the local optimum. However, the over-fitting issue and the slow speed of convergence are the shortcomings existing in the neural network model. Since there is not an effective way yet to forecast the target output, we have taken the amount of fertilizer and soil nutrients as input in the neural network model in general, whereas the resulting value would be the output. During this procedure, we have studied the relationship between the output and the amount of fertilization. In addition, we have adopted the algorithm of support vector regression (SVR) algorithm to facilitate such regression, thereby preventing the neural network from being over-simulated while fitting the situation. During actual application we have added an online SVR algorithm to address the issue of repetitive training of the SVR model. The target yield and the content of soil nutrient serve as the input units, whereas the amount of fertilization is taken as the output unit. The model of decision-making on the fertilization is set up accordingly.

3.3 Design of the knowledge base

The knowledge base refers to the key module that ensures the intelligence of the decision support system (DSS). Developing the knowledge base is regarded as the advanced stage during the development of DSS. In a knowledge-based system, knowledge is constantly organized in a certain manner. In this paper, we have stored the knowledge in a database in the form of a table, while adopting the method of expression of production rules. In addition, in terms of the inference engine, we can leverage technologies of search or pattern matching to perform the design and to obtain solutions to the issue. This rule can be divided into two parts, namely, premise and conclusion, which are used to express causality. Its general form is specified: If A then B, where A refers to the precondition and B refers to the conclusion or action to be taken. A new rule shall comprise the number of the rule, its premise, conclusion and credibility. In the system, we may add, delete or modify the rule without considering its relationship with other rules.

While adopting the KBE-based approach to intelligent system of decision-making on irrigation, we must ensure that the expert knowledge is equivalent as the other experts are in the agricultural field, so as to build the optimal system performance and excellent problem-solving ability [20]. In the system, the knowledge base primarily refers to the knowledge and experience of experts in the sector of farmland water and fertilizer irrigation. In addition, we have extracted the experimental data as well as data on the water and fertilizer irrigation of varying crops in the local or similar areas, which primarily consist of the irrigation systems in favor of water-saving, the relationship between crop water and yield, etc. In terms of the design of the knowledge base in the system, we need to promote the intelligent decision-making during system management on the one hand, whereas on the other hand, it is for farmers to inquire and accurately understand the knowledge related to irrigation.

3.4 Implementation of the intelligent decision-making module

The intelligent decision-making module is designed accordingly to the previous database, knowledge base and model base. By leveraging the data obtained by reasoning and the collection of real-time data information, we are able to conduct real-time analysis and decision-making, while enabling decision-makers to achieve simpler and faster judgment and providing them with access to decision-making information. Based on the crop information and the soil moisture, the amount of irrigation can be determined to enhance the benefits brought along by increased water efficiency.

Ever since the users enter into the KBE-based system of support to intelligent decision-making, they are able to opt for decision options and select the decision strategies of varying subjects of decision-making, so as to ensure that the method and process of decision support are designed based on users’ needs and conditions, as illustrated in Fig. 10. First, we need to call the model base in turn. Through the model of calculation, we shall call the relevant data in the database module for calculation. Subsequently, users would opt for their method in the module of model base, so as to calculate the evapotranspiration and water demands of crops, in addition to the rainfall, etc. Last but not least, we shall adopt the water balance model to determine the demands of crops for the irrigation amount.

Figure 10.

Interface of decision-making on irrigation.

3.5 Visual query subsystem

The visual query subsystem consists of modules such as temperature and humidity, rainfall, amount of irrigation, weather forecast, and knowledge of irrigation. By taking measures to ensure real-time monitoring and graphical display of the irrigation water, we can enable decision makers to reasonably decide upon issues at any time and raise the awareness of users for DSS. Furthermore, we have leveraged the technology of web mining to obtain the data related to the weather forecast next week, and we need to adopt the technology of data visualization to represent the data in the form of graphics and images. Hopefully, our investors are able to gain access to such information and observe the data with more insights in this respect. Last but not least, we need to modify vital information concealed in the data.

4. Result analysis and experiment

4.1 Cloud platform for the intelligent decision support system

Based on the aforementioned research, we have set up the intelligent decision support system (DSS) with the KBE approach, and the python language is used in the decision support technology. The major interface of the system is illustrated in Fig. 11. The KBE approach to intelligent irrigation DSS is mainly composed of the modules such as database, model library, knowledge base, intelligent decision-making and visual query, etc. During the study, we have integrated the expert experience with massive amount of knowledge through computer programs, and we have leveraged the mechanisms of reasoning and explanation, and to imitate agricultural experts for judgment and analysis. Users can opt for the technologies related to data mining, machine learning, and AI, so as to make accurate decisions about irrigation amount, evaluate economic benefits, save water and fertilizer, and increase crop yields and output values.

Figure 11.

Smart system of support to decision-making.

4.2 Results and analysis of decision-making algorithm on fertilization

The water and fertilizer model in the previous model were carried out experimental analysis, under circumstances of the same training set, we trained the BP neural network model and the SVR model. In addition, we have compared the root mean square error of the estimated results of the two models, so as to obtain a comparison chart for the model error, as specified in Fig. 12.

Figure 12.

Model error comparison.

Judging from the comparison of model errors shown in Fig. 11, we may conclude that the root mean square error of potassium, phosphorus, and nitrogen in the online SVR model decision is the smallest, followed by the BP neural network, whereas the method of fertilizer effect function features the largest error during prediction of results. In general, it shall be noted that the algorithm has higher accuracy of prediction than the model of the function method, features the universally applicable characteristics. On the other hand, the method of the fertilizer effect proves to be faced with constraints. The online SVR model can represent the relationship among the amount of fertilization, the content of soil nutrients as well as the target yield. Furthermore, the model is even more consistent with the decision-making module for on-site research on application and the fertilization guided by production.

4.3 Verification and application of the model of decision-making on irrigation and fertilization

4.3.1 Production plan for decision-making

In this study, we have taken tomatoes planting as an example, and we have conducted the experiment on the irrigation and fertilization at the Xinjiang Academy of Agricultural Sciences. In addition, we have input the parameters such as soil information, climatic factors, and production status over the years inputed into the decision-making system for simulation decision-making. As a result, we have generated a set of management plans for tomato fertilization, so as to verify the functions of each module in the system. The result data of fertilization decision-making is composed with the experimental data of high-yield tomato planting and the mode of cultivation in the Technical Regulations for Tomato Cultivation in Xinjiang (DB65/T 2214-2005).

The demonstration bases of irrigation and fertilization are located in Urumqi, Xinjiang. We have installed the equipment of intelligent irrigation in the solar greenhouse, and we have carried out the tomato “Jinpeng No. 1” experiment on irrigation and fertilization. The content of the organic matter of the greenhouse environment roughly amounts to 8 mg/kg. The content of alkaline nitrogen roughly amounts to 50 mg/kg. The available content of phosphorus content roughly amounts to 10 mg/kg. The content of available potassium roughly amounts to 150 mg/kg. The pH value of soil ranges between 6.5 and 8.5, and the it can be found that the content of soil salt actually amounts to less than 3 mg/kg. We have elaborated on the conditions of production environment and studied the requirements in the decision module of the pre-sowing plan in the system, so as to obtain a plan on pre-sowing production suitable to the local production links. We have achieved the results of the recommended varieties, density of planting and breeding plan given by the pre-sowing plan on decision-making and production, as shown in Table 1.

Table 1
Production plan of decision-making before broadcasting

Recommended varieties		Recommended planting density				Recommended seedling program
Crop varieties	Credibility	Planting density/plant $\cdot$ hm ${}^{-2}$	Seeding rate/kg $\cdot$ hm ${}^{-2}$	Plant spacing/m	Average line spacing/m	Seedling sowing date	Seedbed density/plant $\cdot$ hm ${}^{-2}$	Nursery area/m ${}^{2}$
Jinpeng No. 1	0.89	41752	1.295	0.33	0.55	October 3	473.3	1543.92

4.3.2 Verification of the scheme of the irrigation amount

In this paper, we have established the decision-making knowledge base, and we have named the Fintech area as the oasis. Jinpeng No. 1 is as infinite type of growth, fast expansion of fruits, ideal early maturity, sound commercial properties, optimal disease resistance, strong storage resistance, and good heat resistance. The findings of our comparison indicate that while opting for crop varieties through the decision-making system, we are able to better meet the needs of tomato varieties amid the regulations of cultivation technologies. The required planting density ranges between 40,000 and 45,000 plants/hm ${}^{2}$ , whereas the requirements for plant spacing range between 40 and 50 cm. In addition, the seedbed density is suggested to reach 500 plants/hm ${}^{2}$ . Judging from our comparison, the planting density of the decision-making plan can meet the requirements of technical regulations. Furthermore, the average row spacing and seedbed density are slightly less than the requirements on plant spacing.

Based on the aforementioned information, the intelligent system of decision-making on irrigation and fertilization enables us to decide upon the amount of irrigation water allocated for varying growth periods in the cultivation environment, as shown in Table 2.

Table 2
Scheme of decision-making on the amount of irrigation water

Methods	Recommended irrigation quota/m ${}^{3}\cdot$ hm ${}^{-2}$	Irrigation amount at seedling stage/m ${}^{3}\cdot$ hm ${}^{-2}$	Irrigation amount during flowering/m ${}^{3}\cdot$ hm ${}^{-2}$	Irrigation amount during fruit expansion period/m ${}^{3}\cdot$ hm ${}^{-2}$	Irrigation amount during maturity/m ${}^{3}\cdot$ hm ${}^{-2}$
This system	2940	192.3	873.8	1534.7	339.2
Xinjiang Tomato Cultivation Technical Regulations	3600 $\sim$ 4050	150 $\sim$ 170	150 $\sim$ 170	1650 $\sim$ 1855	1650 $\sim$ 1855

The technical regulations on the cultivation of tomatoes in Xinjiang suggest that the irrigation amount for the entire growth period in northern Xinjiang amounts to 3600 $\sim$ 4050 m ${}^{3}$ /hm ${}^{2}$ ; whereas the irrigation amount for drip irrigation in southern Xinjiang amounts to 4350-4950 m ${}^{3}$ /hm ${}^{2}$ . The recommended amount of irrigation in the decision-making scheme amounts to 2940 m ${}^{3}$ /hm ${}^{2}$ , which is compared with the recommended amount of irrigation in the technical regulations of cultivation to obtain an accuracy of 81.7% for decision-making.

4.3.3 Verification of the scheme of decision-making on fertilization amount

The system is able to intelligently identify the status-quo of distribution for the fertilization amount. The results of the total amount of application of nitrogen fertilizer, phosphate fertilizer, and potash fertilizer as well as the proportion of organic fertilizer are specified in Table 3.

Table 3
Scheme of decision-making on the fertilization amount

Methods	Recommended total fertilization/kg $\cdot$ hm ${}^{-2}$			Recommended organic-inorganic nitrogen ratio
	N amount	P ${}_{2}$ O ${}_{5}$ amount	K ${}_{2}$ O amount	Organic nitrogen	Inorganic nitrogen
This system	192.5	119.4	37.7	2.5	7.5
Xinjiang Tomato Cultivation Technical Regulations	180 $\sim$ 225	90 $\sim$ 120	30 $\sim$ 60	2.5	7.5

As specified in the technical regulations on the tomato cultivation in Xinjian, the recommended amount of fertilization for tomatoes amounts to 180 $\sim$ 225 kg/hm ${}^{2}$ for nitrogen fertilizer, 90 $\sim$ 120 kg/hm ${}^{2}$ for phosphate fertilizer, and 30 $\sim$ 60 kg/hm ${}^{2}$ for potassium fertilizer. By comparing the technical regulations on cultivation, it shall be noted that the recommended amount of fertilization in the scheme of decision-making on the fertilization amount is consistent with the requirements of the relevant technical regulations.

4.3.4 Growth index and yield verification

We have selected the schemes of decision-making and local experience of cultivation as the basis for carrying out the experiments on the planting of tomatoes. The planting commenced on October 3, 2019, and we have regularly measured the diameter and the height of plant, and obtained the comparison results on the growth period of crops. Specifically, we have recorded the results from October 10, 2019 to December 29. We have measured the diameter of the stem and the height of the plant on a weekly basis, while measuring and marking 10 measuring points in each row, and two fixed plants for each point of measurement. By comparing the effects imposed by the intelligent schemes of decision-making and local experience of cultivation on the actual growth indicators and yields of tomatoes.

The results of comparison of the growth changes of stem thickness under varying planting schemes are specified in Fig. 13, whereas the results of comparison of the growth changes of plant height under varying planting schemes are illustrated in Fig. 14. Judging from the comparison, we may conclude that the crops under management schemes on water and fertilization obtained by the intelligent system of decision-making feature a thicker stem and a higher height during the growth period than those obtained according to local experience of cultivation. The output of the system of decision-making amounts to 620 kg, which is a relative increase of 5.08% compared with the 590 kg obtained from local experience of cultivation.

Figure 13.

Comparison of tomato stem thickness under varying planting schemes.

Figure 14.

Comparison of tomato plant heights under varying planting schemes.

5. Discussion

We have elaborated on the intelligent technologies of support to irrigation decision-making based on the KBE approach and the purpose of decision-making on the demands of crops for water and fertilization, so as to study the methods of calculating the water demands of crops and the amount of irrigation and to set up the coupling model on water and fertilization accordingly. In addition, we have explored the establishment of the corresponding database, studied the knowledge engineering technologies, and developed the corresponding knowledge based on the python language. Besides the water demands of crops in the scheme of intelligent decision support system (DSS), we have also introduced the detailed functions of the system module in this paper, and applied the system in the demonstration base. Judging from our research findings, there is an optimal accuracy in terms of the application of irrigation decision-making, which will facilitate the popularization and application of the approach.

The irrigation DSS based on knowledge engineering proves to be intelligent due to the fact that the system is capable of leveraging expert knowledge to assist decision-making. In addition, the system is able to modify its behavior as the environment of decision-making is subject changes. Moreover, the system can learn new knowledge while updating its integrated knowledge base as the environment varies. The highly accurate models on the irrigation decision-making, in addition to the supplement and updating of database, model library and knowledge base, enable us to address the intricate and highly non-linear issues arisen from irrigation decision-making. During the application of irrigation engineering in terms of farmland water and fertilization, the technology of “Internet $+$ ” is applied, which is set up on the basis of big data. The management platforms of irrigation help enhance the reliability and reasonableness of decision-making on irrigation.

We have integrated the empirical data of experts in the field of fertilization, and established the method of automatic acquisition for the decision-making on knowledge of irrigation and fertilization. In addition, we have completed the analysis of decision-making on irrigation and fertilization oriented towards knowledge engineering. Moreover, we have put forward the method of acquiring knowledge of irrigation and fertilization by using the web crawler technology and the method of knowledge expression based on ontology, thereby achieving the automatic analysis of the demands for water and fertilizer as well as knowledge related to irrigation and fertilization decision-making. We have set up the knowledge base of DSS for irrigation and fertilization through the acquisition and expression of relevant knowledge. In terms of the amount of applying fertilizers in the decision-making model based on online SVR, we have compared with the results with the method of fertilizer effect function and the BP neural network. Judging from the research results, the error is minor and the model features higher accuracy. We have also analyzed;he relationship between the target yield and the fertilization amount, while recommending the changes to the target yield for fertilization. Based on the influence imposed by the amount of fertilization, we have put forward the decision-making approach on the amount of fertilization based on the maximum benefits, and established the decision model of the amount of fertilization with the optimal comprehensive benefits. In addition, we have achieved the application of the knowledge on tomato irrigation and fertilization into the intelligent decision support system.

Footnotes

Acknowledgments

The authors would like to thank the other team-mates or useful comments and discussions. The authors would also like to thank the anonymous reviewers for their comments. The work was partially supported by Scientific Research Project of Zhejiang Education Department of China under Grant Nos.Y202043638 and Y202045381, National Key R&D Program of China during the 13th Five-Year Plan Period under Grant Nos. 2017YFD0201504.

References

Mérida

G.A.

et al.,Coupling irrigation scheduling with solar energy production in a smart irrigation management system, Journal of Cleaner Production 175(2018), 670–682.

X.Y.

et al.,Current status of irrigation district water using decided supporting system and its prospects,Res Soil Water Conservation,15(4)(2008),28–32.

George

B.A.

, Reddy

B.R.S.

, Raghuwanshi

N.S.

,Decision support system for estimating reference evapotranspiration,J Irrigation Drainage Engin,128(1)(2002),1–10.

E. Giusti and S.A. Marsili-Libelli, Fuzzy Decision Support System for irrigation and water conservation in agriculture. Environmental Modeling & Software 63 (2014), 73–86

Knapp

, Huang

,Do climate factors matter for producers irrigation practices decisions? Journal of Hydrology 552 (2017), 81–91.

Zhang

Y.B.

et al.,Decision support technology for irrigation research based on knowledge engineering,China Agricultural Science and Technology Review,19(8)(2017),47–54.

Zou

, Chen

, Ouyang

Z.X.

,Research and application of a semantic ontology-based network information security management assistant decision-making system for colleges and universities,Software Engineering,23(8)(2020),52–55.

Zeng

S.M.

,Research and application on optimaization methods with auto-body GD&T design process, Mh.D. Dissertation, Shanghai Jiaotong University, 2011.

Peng

Y.H.

, Hu

,KBE Technology and its Application of Product Design, Ph.D. Dissertation, Shanghai Jiaotong University Press, 2007.

10.

Liu

F.H.

,Computer aided conceptual design for satellite based on ontology knowledge, Ph.D. Dissertation, Harbin Industrial University, 2011.

11.

Zhou

X.H.

, Peng

Y.H.

,Modern Mold Design and Manufacture Theory and Technology, Ph.D. Dissertation, Shanghai Jiaotong University Press, 2000.

12.

Y.M.

, Sun

X.Y.

, Zhang

X.L.

,Beyond the knowledge base: knowledge analysis and decision-making – Taking the knowledge management system of Shanghai University of Science and Technology as an example, Library Theory and Practice (4) (2020), 48–52.

13.

Luo

,Research on intelligent design system foe aluminum extrusion and the key technologies, Ph.D. Dissertation, Central South University, 2004.

14.

Khalaf

F.K.

et al.,A review study on the optimizing the performance of soil using nanomaterials,Advances in Industrial Engineering and Management,9(2)(2020),1–10.

15.

Peng

et al.,Knowledge-based resource allocation for collaborative simulation development in a multi-tenant cloud computing environment,IEEE Transactions on Services Computing,11(2)(2018),306–317.

16.

Sun

Y.S.

, Zhang

M.Z.

, Zhu

L.J.

,Domestic knowledge visualization research progress: theoretical analysis, Information Theory and Practice (11) (2015), 126–131.

17.

Peter

G.W.K.

,Value analysis: Justifying decision support systems,MIS Quarterly,5(1)(1981),1–15.

18.

Sprague

R.H.

,A framework for the development of decision system, Mis Quarterly, 1980.

19.

Zhou

X.M.

,A support system for crop water requirement diagnosis and irrigation decision making, Ph.D. Dissertation, Northwest Agriculture and Forestry University of Science and Technology, 2013.

20.

Zhang

Y.B.

et al.,Improved online sequential extreme learning machine for simulation of daily reference evapotranspiration,Tecnologíay Ciencias del Agua,8(2)(2017),127–140.

21.

Zhang

Y.B.

et al.,Applicability evaluation of different algorithms for daily reference evapotranspiration model in KBE system,International Journal of Computational Science and Engineering,18(4)(2019),361–674.

22.

Huang

J.X.

, Li

W.K.

, Zhang

X.P.

, Li

J.J.

,Knowledge acquisition and matrix method of generalized multi-scale information system,Journal of Shanxi University (Natural Science Edition),43(4)(2020),865–873.

23.

Sun

Y.X.

, Huang

X.P.

, Zhou

X.Z.

, Tang

B.J.

,Construction of a knowledge-based naval battlefield situation assessment assistant decision-making system,Command Information System and Technology,11(4)(2020),15–20.

Decision-making technology based on knowledge engineering and experiment on the intelligent water-fertilizer irrigation system

Abstract

Keywords

1. Introduction

2. Design principles based on Knowledge-based Engineering

2.3 Knowledge reasoning

3.2.4 Water balance model

3.3 Design of the knowledge base

3.4 Implementation of the intelligent decision-making module

4. Result analysis and experiment

4.1 Cloud platform for the intelligent decision support system

4.3.1 Production plan for decision-making

Table 1 Production plan of decision-making before broadcasting

Table 2 Scheme of decision-making on the amount of irrigation water

Table 3 Scheme of decision-making on the fertilization amount

Footnotes

Acknowledgments

References

Table 1
Production plan of decision-making before broadcasting

Table 2
Scheme of decision-making on the amount of irrigation water

Table 3
Scheme of decision-making on the fertilization amount