Evaluation method of sponge city potential based on neural network and fuzzy mathematical evaluation

Abstract

The sponge index is the core of the sponge city flood forecast. Whether the model is reasonable or not directly affects the final forecast result. The study of classification problems using neural network models is an important branch of the artificial neural network application field. The classification and pattern recognition functions can be used to achieve flood classification and sponge index monitoring. In this paper, the author analyze the evaluation method of sponge city potential based on neural network and fuzzy mathematical evaluation. After training, the BP neural network model can effectively evaluate the potential of the sponge city, and based on the input of special information on rain conditions, it can analyze and calculate the flood risk level. It can be seen that this network model has a high mapping capability and can be correctly classified. Therefore, it is feasible to use BP neural network to solve the real-time classification of flood risk. The sponge city potential method and underground drainage system proposed in this paper can provide a reference for promoting sponge city construction.

Keywords

Sponge indicator monitoring and tracking neural network algorithm internet of things

1 Introduction

With the development of the national economy and the continuous improvement of urbanization, urban diseases have also emerged. The problem of floods in urban diseases is particularly prominent [1]. The floods caused economic losses and casualties in the country. Cities urgently need to find a sustainable development road that can solve urban floods. Based on the above background, the sponge city came into being [2, 3]. At present, most sponge city planning and construction management decisions still rely on the traditional use of two-dimensional GIS or picture data, etc. It is difficult to timely, intuitively and effectively transmit information, and there are limitations and inefficiencies in project construction process and operation tracking management [4]. Limitations of traditional planning and construction operation and maintenance management and control technology Currently, sponge city planning and construction operation and maintenance are still managed using traditional measures [5, 6]. Multi-party coordination relies on the “sponge office” for coordination and communication. The project process relies on scattered two-dimensional drawings and file information transfer. The control over the actual construction situation is site investigation and summary of form information [7]. The traditional method is a stage that is undergoing in the management of sponge city planning and construction. It inevitably has the following limitations: First, the information is difficult to transfer intuitively and effectively, and the two-dimensional planar information is abstracted. Intuitive; Second, the lack of a timely and effective overall coordination and information platform, the information exchange efficiency of the participants is low, which is not convenient for the continuous tracking and feedback of the construction effect and operation of the project; the third is the inability to query the urban hydrological waterlogging [8, 9].

As sponge city is a domestic livelihood project that has emerged in China in recent years, domestic and foreign scholars’ research and reporting on sponge city is at an early stage. Some studies have elaborated the origin, development, connotation and construction method system of the concept of “sponge city” by using theories such as ecosystem services and landscape security pattern, and put forward the future research direction of “sponge city” [10]. Some studies suggest that the construction of sponge cities is closely related to flood control. There is a close relationship between sponge cities and ecological cities, low-carbon cities, and smart cities. Some studies have taken the sponge city planning and construction in Ningxiang County as a case, and analyzed empirically the issues such as the concept and planning of sponge city construction in the rainy areas of the south [11, 12]. At the same time, in order to reduce urban floods, a research and design of a fully permeable pavement composition system, combined with the actual application of the project, has been obtained to solve urban floods, effectively replenish groundwater, and build a new structure design mode of “sponge city". Throughout the research results at home and abroad, most of the research areas in sponge cities involve underground pipe network planning and rainwater ecological management, and some guiding suggestions have been put forward [13 –15]. No in-depth research has been conducted on regional sponge city potential methods. Based on this, this paper innovatively combines mathematical statistics, fuzzy neural network, fuzzy comprehensive evaluation, satellite image recognition and other methods to scientifically start from indicators such as water resources, water pollution, flood disasters, and ecological green space to build a regional sponge city potential method system [16]. Aiming at the flooding problem in the areas with the most potential for sponge cities, an underground drainage system based on sponge cities was developed, and the effect of field monitoring was analyzed. The research ideas and results can provide the basis and reference for the in-depth study of sponge cities in the future, and can also provide a reference for the potential of other cities to build sponge cities.

2 Related work

The construction of the sponge city is a large-scale comprehensive project with complex construction content and extensive coverage [17, 18]. It is necessary to coordinate the intensive construction of the districts, parks, green spaces, roads, rivers, various pipelines, etc. in the city, which is difficult to manage; The regular supervision and evaluation system requires accurate and real-time tracking and management of the entire construction process from the initial planning stage to the later construction operation and maintenance [19].

The core of BIM technology is a three-dimensional visualization model. This three-dimensional model not only serves as the most intuitive expression form of the engineering project, but also serves as the information carrier of the engineering project [20]. It stores the relevant information of each stage in the construction of the sponge city and expresses it visually. In addition, the sponge city BIM should be able to be expanded, that is, the information of different stages is integrated with the information of the current stage to form a new model, and then passed down to ensure the retention and utilization of information in each stage of the project [21]. The creation method of sponge city BIM is very important to visually manage the construction and operation and maintenance of sponge facilities from the source-process-end process in three-dimensional (underground + above-ground). The 3D model of BIM requires the help of database for the storage of engineering project information. The data of the model itself also needs to be stored in the database. Sponge City’s life-cycle engineering information spans multiple stages of planning, design, construction, and operation. There are many categories and a large number of projects [22]. Therefore, it is necessary to design and create a large-scale relational database for the entire life cycle of Sponge City construction, integrating the required information. And realize the organization, classification and association of data. The sponge city related information includes sponge city planning data, GIS data, terrain DEM data, satellite map data, hydrological and waterlogging data, sponge facility basic information, sponge facility construction operation and maintenance information, and static information such as drawings, text and numbers, and physical properties [23]. Dynamic information acquisition of networked sensing systems.

The purpose of information management is to make use of the stored information and carry out related work. For different divisions of labor and needs, each participant should be able to operate and control through the corresponding functional modules of the sponge city BIM platform. This requires the application function design and R & D implementation of the sponge city BIM platform according to the characteristics and requirements of the operation and maintenance management of the sponge city planning and construction. The sponge city BIM platform shall realize the analysis, processing, release and management of the three-dimensional visual integrated model of the sponge city, in order to obtain the information on the effectiveness of the construction and operation and maintenance of the sponge facilities, perform the three-dimensional dynamic and accurate simulation of the urban hydrological and waterlogging situation, and the working status of the sponge facilities And sponge indicators for monitoring and tracking. GIS system as show in Fig. 1.

Fig. 1

GIS system.

The indicators that are most closely related to the construction of the sponge city are rainfall, water quality, flood disasters, and green space. In contrast, the four factors of water resources, water quality, flood disaster risk, and green space area are infiltration and storage with the six core goals of the sponge city. “, Stagnation, net, use, row” are closely related. Therefore, a comprehensive evaluation is carried out based on the introduction of water resources, water quality, flood disasters, and green space areas in the assessment of the potential of a regional sponge city. Based on the mathematical statistics method, this paper systematically explores the rainfall distribution and water resources status of tomorrow’s Fuxin District from three aspects: the temporal, spatial distribution and spatial distribution of water resources. Sufficient water resources but incapable of drainage and comprehensive utilization can easily lead to water pollution. Therefore, this paper analyzes six pollution indicators by introducing fuzzy neural networks to obtain the overall situation of water quality in each region. The area of green space affects the distribution of urban water resources and the absorption and discharge of water. Therefore, the area of green space is also an important indicator of the potential of the pilot city of sponges. At this time, a GIS program is introduced to derive 12-level accuracy (32 m×32 m), and images are written using MATLAB The recognition algorithm analyzes the green area. The level of regional flood disaster risk affects the selection of pilot cities in sponge cities. For example, low-lying roads and other places are more likely to cause flood disasters, but it inevitably corresponds to the first-level indicators such as the pregnancy environment, hazard factors, and disaster carriers. It goes deeper. Finally, the four major indicators of rainfall, water quality level, green area, and flood disaster risk were analyzed for correlation degree to obtain the potential ranking of regional sponge cities.

3 The establishment of model

3.1 BP neural network

For the analysis of time series data, the first step is to use statistical learning method to analyze according to the characteristics of time series. Later, with the development of machine learning methods, researchers at home and abroad gradually use machine learning methods into time series research. Later, with the emergence of neural network method, using neural network to analyze time series became popular. With the continuous improvement of computer hardware level, a neural network model with more complex structure and stronger expression ability has emerged. With the development of neural network model with multi-layer hidden layer, that is, deep neural network model, the analysis of time series has become a hot topic for scholars at home and abroad. At the same time, in the process of development, these research methods are combined with the professional knowledge in their respective fields, so that scholars need to have the professional knowledge in the corresponding research fields when conducting research. In summary, the current time series prediction methods mainly include statistical learning method, traditional machine learning method, feedforward neural network method, cyclic neural network and their variants (including LSTM, The neural network has the ability of general approximation, powerful computing and expression. It is an ideal rule and pattern discovery and learning device, which can be used to develop more advanced predictors. By studying the working principle of neural network, we can adjust and improve its structure, and combine its models, so as to improve the accuracy of time series prediction. It is a hot spot of current research, and also a very valuable and practical thing. Neural network model as shown in Fig. 2. Neural network topology diagram as show in Fig. 3.

Fig. 2

BP neural network.

Fig. 3

Neural network topology diagram.

Neural network is a kind of operation model composed of a large number of operation nodes (i.e. neurons) connected with each other. Each node represents a specific output function, the excitation function. Each connection between two nodes represents a weighted value of the signal passing through the connection, which is called the weight, which is equivalent to the memory of the artificial neural network. The output of the network depends on the connection mode of the network, the weight value and the excitation function. The network itself is usually an approximation to some algorithm or function in nature, or an expression of a logic strategy. $β = \sum_{m = 1}^{n} α_{mn} x_{n}$ (1) $y = f (β_{j} - λ_{j})$ (2)

Compared with the common neural network structure, RNN has been modified in the circulation layer. There is no connection between the same layer of the feedforward neural network. The hidden layer of RNN not only receives the signals from the neurons in the previous layer, but also the signals from the nodes in the previous hidden layer, so the dependence between the time nodes is considered. Compared with feedforward neural network, RNN greatly improves the analysis ability of time series. Neural network is a black box model. Its learning process is the process of adjusting the connection weights between neurons according to the input training data. To train feedforward neural network, the most classical method is back propagation algorithm: $f (x) = {\begin{matrix} 1, x \geq 0 \\ 0, x < 0 \end{matrix}$ (3)

(2) Operation algorithm

In standard RNN, this module is tanh function. The hidden layer of LSTM is also such a repeated chain module, but its internal structure is more complex. The model is based on the learning black box model. As long as one has the measurement data, the proposed method of modeling and predicting data center power consumption can be easily used in other more complex situations. These simplifications in the simulation can simplify data collection without compromising the generality of modeling methods. $y = f (\sum_{m = 0}^{n} α_{mn} x_{mn})$ (4)

In the field of computer, data mining technology is a hot research topic, and time series analysis and prediction technology is an important branch of data mining. Time series refers to a series of observation values recorded in time sequence in the process of production and scientific research. It is a random data formed by a variable or multiple variables at different times, reflecting the law of development and change of the phenomenon. Time series has three characteristics: randomness, continuity and periodicity.

In the process of the input layer, the output data source of the output layer is: $\begin{matrix} v = \sum_{m = 0}^{n} t_{mj} y_{m} \\ L = f (v_{j}) = f (\sum_{m = 0}^{n} t_{mn} y_{m}) \end{matrix}$ (5)

The data of current time point will be interfered by many external factors, so the time series is random. In terms of fine-grained time, the values of adjacent time series are continuous and time-dependent, so they are continuous. That is to say, the data at the current time point is likely to be affected by one or more previous time points, which are called short-term dependency and long-term dependency respectively. Periodicity refers to the phenomenon that time series data often presents periodic changes or certain trends as a whole. The above three characteristics not only show the feasibility of prediction time series, but also reflect the difficulty of prediction. Therefore, time series is of great significance. $E = x_{m}^{j} (1 - x_{m}^{j}) (x_{m}^{j} - y_{m}^{j})$ (6)

The number of layers is the same as the number of verification.

If they are different, the calculation error exists in each layer. $E = x_{m}^{n} (1 - x_{m}^{n}) (x_{m}^{n} - y_{m}^{n})$ (7)

Time series forecasting model is widely used in power demand forecasting, financial market forecasting, sales performance forecasting, environmental weather forecasting, medical diagnosis and other fields. $e = \frac{1}{2} \sum_{j = 1}^{n} {(r_{j} - L_{k})}^{2}$ (8)

For different time scales, the prediction of time series can be divided into two categories: (1) fine-grained prediction. (2) Coarse grained prediction. Fine-grained prediction is to predict the specific value of the next time point according to the historical data. Coarse-grained prediction is to predict multiple time points in the future, that is, the mean range of the next period of time. This process is also called trend prediction. At present, time series prediction methods mainly include statistical learning method, traditional machine learning method, feedforward neural network method and cyclic neural network method. $e = \frac{1}{2} \sum_{j = 1}^{n} {[r_{j} - f (\sum_{m}^{n} t_{mj} t_{m})]}^{2}$ (9)

Error signal to the output layer is expressed by the following equation. $e = \frac{1}{2} \sum_{j = 1}^{n} {(r_{j} - f [\sum_{m = 0}^{n} t_{mj} f (\sum_{m}^{n} α_{mj} x_{m})])}^{2}$ (10)

3.2 Error revision

These statistical learning methods usually assume strict preconditions, such as the requirement of stationarity, and the variance of data cannot be too large. However, due to the randomness and instability of time series, the applicability of such models is limited. Traditional machine learning methods, such as linear regression, support vector machine, random forest, generally use the multi-dimensional characteristics of the data to construct the function equation from the dependent variable characteristics of the time series to the target prediction value, and then construct the regression model by constructing the loss function and optimizing it. However, the traditional machine learning method only considers the characteristics of the current time point of the time series and ignores the time dependence of the characteristics, so the prediction effect is often poor. Feedforward neural network is a kind of network composed of multiple stacked neurons, which is extracted by the layered network structure. The adjustment formula is as follows. $\partial_{ij} (v + 1) = \partial_{ij} (v) - Δ e_{i}^{k} x_{j}^{k - 1}$ (11)

(2) Partial revision. Taking into account the existence of the BP neural network, the convergence rate is slow and prone to local errors and other issues, momentum factor is increased, partial problem is adjusted. The problem can be solved by the traingda, traingdm function in the MATLAB software.

The formula for solving the local problem is as follows. $Δ x (v + 1) = ɛ Δ x (v) + γ (1 - ɛ) \frac{\partial e (v)}{\partial x (v)}$ (12) $x (v + 1) = x (v) + Δ x (v + 1)$ (13)

The hidden layer of RNN is modified on the basis of ordinary computer network. The hidden layer not only receives the output from the neuron of the previous layer, but also receives the output from the hidden layer of the previous time point. Therefore, the internal neurons construct the connection between different time data.

To solve the convergence problem, the adjustment formula is as follows. $γ (v + 1) = {\begin{matrix} 1 γ (v), e (v + 1) < e (v) \\ 0.5 γ (v), e (v + 1) > 1 e (v) \\ γ (v), others \end{matrix}$ (14)

Compared with the feedforward neural network, RNN greatly improves the analysis ability of time series. However, when analyzing the long-term dependent scenarios of time series, RNN has the problems of gradient explosion and gradient disappearance in optimization, so it is not suitable for analyzing the long-term dependent time series. $γ_{mn} = γ_{lk} - α \frac{\partial e}{\partial γ}$ (15)

There is the following equation. $\frac{\partial e}{\partial v_{mn}} = \sum_{j = 1}^{m} \frac{\partial e}{\partial v_{mn}}$ (16)

Long and short term memory network (LSTM) is an improved variant of RNN, which solves the problem of gradient explosion and gradient disappearance, and is more suitable for analyzing and predicting long-term dependent time series data. $X^{2} = \frac{\sum_{i = 1} (x_{i} - \bar{x_{i}}) (\overset{\land}{x_{i} - \overset{\bar{\land}}{x_{i}}})}{^{\sqrt{\sum_{i = 1} {(x_{i} - \bar{x_{i}})}^{2} {(\overset{\land}{x_{i} - \overset{\bar{\land}}{x_{i}}})}^{2}}}}$ (17)

Among them, there are the following equations. $\bar{x_{i}} = \frac{1}{n} \sum_{i = 1}^{n} x_{i}$ (18) $\bar{\overset{\land}{x_{i}}} = \frac{1}{n} \sum_{i = 1}^{n} \overset{\land}{x_{i}}$ (19)

4 Construction of a fuzzy neural network model

4.1 Model establishment

In order to obtain the water quality grade data results more accurately, a fuzzy neural network structure is comprehensively analyzed and established, from which it can be obtained that the structure is composed of an anterior network and an anterior network. The antecedent network is used to match the antecedent of the fuzzy rule; the posterior network is used to generate the antecedent of the fuzzy rule.

4.2 Model training

Because the number of training samples has 6 indicators (ammonia nitrogen, dissolved oxygen, chemical oxygen demand, potassium permanganate index, total phosphorus, total nitrogen), the number of input nodes of the fuzzy neural network is 6, and the final output evaluation value is only one. Dimension, so there is one output node. Therefore, the structure of the fuzzy neural network is 6-12-1, that is, there are 12 membership functions, and the center value C and width b of the fuzzy membership function are randomly generated.

Considering the lack of original data, this paper uses equal interval uniform distribution interpolation method to generate 400 sets of training samples from surface water environmental quality standards and comprehensive water quality grade determination standard data, of which 350 groups are selected as training groups and 50 groups are used as prediction groups. Results can be obtained more accurately. In order to realize the learning process of fuzzy neural network based on T-S model, it is generally transformed into an adaptive network, namely the adaptive neural fuzzy inference system (ANFIS). This paper combines the network structure initialized by the water quality grade neural network and its related parameters to obtain the water quality grade ANFIS structure.

Through the above fuzzy neural network training, the water sample index value of each water quality sampling mouth is obtained in this paper, and then the water quality grade index and the comprehensive water quality evaluation standard are obtained based on the predicted value of the neural network. The comprehensive water quality evaluation standards: □ (0, 1. 5], □ (1 5, 2.5), □ (2.5, 3.5), IV (3.5, 4.5) and □ [4.5, 6). The specific grading standards are shown in Table 1. Urban rainfall and flood risk as show in Fig. 4.

Table 1
Comprehensive water quality evaluation standards

Category Ammonia nitrogen Dissolved oxygen Cod Potassium permanganate index Total phosphorus Total nitrogen

□ ≤0.25 ≥7.25 ≤20 ≤2.0 ≤0.10 ≤0.20

□ ≤0.75 ≥6.25 ≤25 ≤4.0 ≤0.15 ≤0.60

□ ≤1.25 ≥5.50 ≤30 ≤6.0 ≤0.25 ≤0.75

□ ≤1.25 ≥3.25 ≤35 ≤10 ≤0.35 ≤1.25

□ ≤1.75 ≥2.25 ≤40 ≤20 ≤0.40 ≤2.00

Category	Ammonia nitrogen	Dissolved oxygen	Cod	Potassium permanganate index	Total phosphorus	Total nitrogen
□	≤0.25	≥7.25	≤20	≤2.0	≤0.10	≤0.20
□	≤0.75	≥6.25	≤25	≤4.0	≤0.15	≤0.60
□	≤1.25	≥5.50	≤30	≤6.0	≤0.25	≤0.75
□	≤1.25	≥3.25	≤35	≤10	≤0.35	≤1.25
□	≤1.75	≥2.25	≤40	≤20	≤0.40	≤2.00

Fig. 4

Urban rainfall and flood risk.

Through MATLAB programming, a comprehensive water quality rating model using a modular neural network is used. The training results are the standard training curve for water quality indicators and the standard prediction curve for water quality indicators. The prediction output of the training data for the comprehensive water quality rating model is basically consistent with the actual output. The error is acceptable. Within the range. At the same time, the actual output of the test data prediction is also more consistent with the prediction. Based on the above results, the comprehensive water quality evaluation model based on the T-S fuzzy neural network was successfully trained.

4.3 Establishment of fuzzy evaluation model

Based on the generalized flood disaster risk assessment model, this paper combines the characteristics of terrain and land features unique to Area A to determine three secondary indicators of the disaster environment, the hazard factors, and the hazard body. Among them, the altitude, slope, Three levels of indicators: surface water resource density, forest coverage, etc.; disaster-causing factors include three levels of heavy rain days, maximum daily rainfall, and average annual rainfall days; disaster victims include population density, birth rate, mortality, GDP density, Three-level indicators such as the proportion of cultivated land area, disaster resistance and disaster relief capacity.

The values of various indicators for flood risk levels are ambiguous, so it is reasonable to use the degree of membership to divide the classification limits. Therefore, the single-factor evaluation matrix is a matrix of single-factor membership r, which is a row element. The degree of membership is determined by calculating the membership function. According to the characteristics of the data distribution, a “half trapezoidal” function is selected here. The membership function of each index is made according to the five-level standard of each factor. According to the above entropy weight theory, Determine the weight of the secondary indicators, see Table 2.

Table 2
Weighting index of flood disaster risk assessment index

Secondary indicators Tertiary indicators Index weight

Flood environment factor(B1) Elevation(C11) 0.1812

slope(C12) 0.1970

Surface water resource density(C13) 0.2146

Forest cover rate(C14) 0.3397

Disaster victim factor(B2) Population density(C21) 0.0985

Birth rate(C22) 0.1276

Mortality rate(C23) 0.1650

GDP density(C24) 0.0899

Proportion of cultivated land area(C25) 0.0942

Disaster resistance(C26) 0.1580

Hazard factor(B3) Heavy rain days(C31) 0.2718

Maximum daily rainfall(C32) 0.3263

Average annual rainfall days(C33) 0.4256

Secondary indicators	Tertiary indicators	Index weight
Flood environment factor(B1)	Elevation(C11)	0.1812
	slope(C12)	0.1970
	Surface water resource density(C13)	0.2146
	Forest cover rate(C14)	0.3397
Disaster victim factor(B2)	Population density(C21)	0.0985
	Birth rate(C22)	0.1276
	Mortality rate(C23)	0.1650
	GDP density(C24)	0.0899
	Proportion of cultivated land area(C25)	0.0942
	Disaster resistance(C26)	0.1580
Hazard factor(B3)	Heavy rain days(C31)	0.2718
	Maximum daily rainfall(C32)	0.3263
	Average annual rainfall days(C33)	0.4256

This paper uses linear weighted grey correlation to calculate four indicators of precipitation, water pollution, flood disaster risk, and green area. By performing dimensionless processing on the original data, and using the coefficient of variation method to obtain the weight of each indicator, the potential index is obtained. The index weights of the evaluation model are shown in Table 3. Sponge urban rainfall dynamic evaluation as show in Fig. 5.

Table 3

Index weights of the potential index evaluation model

Index	Average	Standard deviation	Coefficient of variation	Weights/%
Precipitation	42.57	36.94	0.754	14.2
Water pollution	3.41	2.37	0.412	8.53
Flood	1.75	0.82	0.312	6.54
Green area	55.45	27.43	3.298	62.7

Fig. 5

Sponge urban rainfall dynamic evaluation: (a) rainfall data; (b) Inverse Distance Weighting (IDW) interpolated rainfall surface and (c) reclassified rainfall data.

5 Case analysis

5.1 Sponge indicator monitoring and tracking

Sponge indicator monitoring and tracking is an important aspect of the construction and operation of sponge cities. This module mainly provides: (1) Based on the Internet of Things, big data and other technologies, it realizes the simultaneous monitoring and display of water quantity and water quality data, covering the three dimensions of facilities-projects-catchment basins. Such monitoring data can be connected to the system, and can be monitored and managed by manual sampling review. (2) Dynamic display in the form of charts, which can query, analyze and visualize information management of various data to realize intelligent management of drainage facilities; (3) Based on management The network liquid level and inland river water level monitoring are based on a certain threshold and a warning is issued when the warning value is exceeded.

5.2 Judgment matrix

Neural network has general approximation ability, powerful computing ability and expression ability. It is an ideal rule and pattern discovery and learning device, which can be used to develop more advanced predictors. By studying the working principle of neural network, we can adjust and improve its structure, and combine its models, so as to improve the accuracy of time series prediction. It is a hot spot of current research, and also a very valuable and practical thing. Judgment matrix as shows in Tables 4 –8.

Table 4
A-B Judgment matrix

Matrix B1 B2 B3 Factor

B-1 1 3 1 0.2150

B-2 3 1/3 3 0.5764

B-3 1/5 1/3 1 0.1428

Matrix	B1	B2	B3	Factor
B-1	1	3	1	0.2150
B-2	3	1/3	3	0.5764
B-3	1/5	1/3	1	0.1428

Table 5

B1-C Judgment matrix

A-B	C11	C12	C13	C14	Factor
C11	1	1/5	1/3	5	0.2071
C12	1	1/3	3	3	0.3368
C13	3	1/5	1/1	1/5	0.1542
C14	5	1	1/5	3	0.2698

Table 6

B2-CJudgment matrix

A-B	C21	C22	C23	C24	C25	C26	Factor
C21	1	1/7	1/5	1/5	1/5	1/9	0.0582
C22	7	1	3	3	1/3	1/3	0.2653
C23	5	1/3	1	1	1/3	1/3	0.1542
C24	5	1/3	1	1	1/3	1/3	0.1325
C25	5	3	3	3	1	1/3	0.3999
C26	9	3	3	3	1	1	0.24671

Table 7

B3-CJudgment matrix

A-B	C31	C32	C33	Factor
C31	2	1	2	0.1428
C32	1	1/3	3	0.4179
C33	3	1/3	3	0.3241

Table 8

A-B-C Judgment matrix

A-B	C31
C11	0.0612
C12	0.0846
C13	0.0217
C14	0.5123
C21	0.0428
C22	0.1293
C23	0.0943
C24	0.0874
C25	0.2252
C25	0.1634
C31	0.0223
C32	0.0685
C33	0.0587

Compared with the real data center, the experimental system can be simplified in many aspects. Because our model is based on learning black box model, as long as one has measurement data, the proposed method of modeling and predicting data center power consumption can be easily used in other more complex situations. These simplifications in the simulation can simplify data collection without compromising the generality of modeling methods. The similation result as: $λ = 3.6422, CI = 0.0325, CR = 0.0536 < 0.01$

There is a trade-off between noise reduction and information loss. After analyzing the noise mode of power consumption sequence, we choose the corresponding noise reduction algorithm through appropriate parameter settings. Finally, the noise is reduced by smoothing the data. $λ = 45422, CI = 0.0236, CR = 0.0287 < 0.01$

Through the operation of MATLAB software, the results are obtained. $λ = 5.6784, CI = 0.0125, CR = 0.0115 < 0.01$

Through the operation of MATLAB software, the results are obtained. $λ = 5.5263, CI = 0.0153, CR = 0.0121 < 0.01$

5.3 Normalization processing

When the sample data changes rapidly, the LSTM will deviate greatly from the real value. However, due to the correction effect of additional correction layer, the prediction results of LSTM corr will not deviate greatly, and the prediction effect is obviously better than LSTM. $PN = (P - MIN) / MAX - MIN$ (26)

Principal component analysis is a very widely used data compression method. The main idea of principal component analysis is to reduce the dimension of the data set composed of a large number of or a small number of interrelated variables, and at the same time to retain the changes in the data set to the maximum extent. This is done by converting variables into a set of new variables, which are called the main components, and are orthogonal and orderly, so that when we move down in order, the change retention rate in the original variables will be reduced. Therefore, in this way, the first principal component retains the largest change in the original component. Principal components are eigenvectors of covariance matrix, so they are orthogonal. Some deep learning encoders are also included in the commonly used sequence encoding methods. $W = \sqrt{I + O} + E / 2$ (27)

5.4 MATLAB operation result and analysis

By testing the model, $c = \frac{\sqrt{\frac{\sum (y_{i} - \bar{y)^{2}}}{n - 1}}}{\sqrt{\frac{\sum ([Δ_{i} - \bar{Δ})^{2}}{n - 1}}}$ (28)

The model is based on bilinear recurrent neural network which has strong ability in modeling and predicting time series. By using the learning algorithm based on multi-resolution to train the bilinear recurrent neural network, the learning process is improved, so that it is more robust to the prediction of time series data. $P = P ((Δ_{i} - \bar{Δ})) ≺ μ \sqrt{\frac{\sum ([Δ_{i} - \bar{Δ})^{2}}{n - 1}}$ (29)

The coding series method can make better use of massive historical data to get better prediction results, and it is also easy to train. The resampling based method can not make full use of all the small-scale data; on the other hand, because of the large dimension of the input data, it is not practical to directly use the original sequence to train the neural network. At the same time, more information is retained than the simple average used in the resampling method. It can be seen from the experimental results that the coarse-grained prediction scheme based on coding can provide smaller prediction error and better stability. Operation results as shown in Fig. 6.

Fig. 6

Analysis of operation results.

According to the results of the sponge city potential index, a real area is selected in the most typical Shuangliu area to design the sponge city. From the results, it can be found that precipitation, water pollution, risk of flood and waterlogging disasters and greening area, especially in water resources and rainfall, are the most likely to cause flood and waterlogging disasters.

6 Conclusions

The fuzzy comprehensive evaluation was used to study the risk of waterlogging disasters. Correlation analysis was performed on the four indicators of precipitation, water pollution, floodwaterlogging disaster risk, and green area, and the sponge city potential index ranking was obtained. In areas with potential, priority will be given to the construction of Shuangliu District in the pilot process. Based on the potential assessment method and the design of the drainage system, through the concept of the sponge city, from the perspective of energy saving and emission reduction, a drainage system based on the concept of low-impact development was proposed, and the system carried out indicators such as rainwater runoff, relief of pipe network pressure, and suspended solids in the water Research on the benefits of on-site monitoring, the design and transformation of the drainage system will bring significant effects to the flood resistance of the community, effectively solve the problem of drainage and waterlogging, not only effectively improve the environment, but also take into account economic benefits. The design and transformation of the drainage system will bring significant effects to the flood resistance of the community and effectively solve the problem of drainage and waterlogging.

Footnotes

Acknowledgment

This paper was supported by (1) Project funded by China Postdoctoral Science Foundation; (2) Project funded by the Project (017/2018/A) of FDCT; (3) Project funded by the Project of Macao Foundation.

References

Shao

, Fu

, Li

, Altan

and Cheng

, Remote sensing monitoring of multi-scale watersheds impermeability for urban hydrological evaluation, Remote Sensing of Environment 232 (2019), 111–338.

, Wang

, He

and Shao

, User Identification across Asynchronous Mobility Trajectories, Sensors 19(9) (2019), 2102–2111.

Shao

, Wu

, Wang

, Du

and Li

, Sea Ships: A Large-Scale Precisely Annotated Dataset for Ship Detection, IEEE Transactions on Multimedia 20(10) (2018), 2593–2604.

Shao

, Yang

and Zhou

, Performance Evaluation of Single-Label and Multi-Label Remote Sensing Image Retrieval Using a Dense Labeling Dataset, Remote Sensing 10(6) (2018), 964–977.

Shao

and Cai

, Remote Sensing Image Fusion with Deep Convolutional Neural Network, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 11(5) (2018), 1656–1669.

Zhang

, Weng

and Shao

, An evaluation of monthly impervious surface dynamics by fusing Landsat and MODIS time series in the Pearl River Delta, China, from 2000 to 2015, Remote Sensing of Environment 201 (2017), 99–114.

, Shao

, Fu

and Cheng

, The Dynamic Analysis between Urban Nighttime Economy and Urbanization Using the DMSP/OLS Nighttime Light Data in China from 1992 to 2012, Remote Sensing 9(5) (2017), 416–420.

Wang

, Hu

, Jiang

, Han

and Shao

, Heteroskedasticity tuned mixed-norm sparse regularization for face hallucination, Multimedia Tools and Applications 75(24) (2016), 17273–17301.

Ding

, Shao

, Zhang

, Xu

and Wu

, A Comprehensive Evaluation of Urban Sustainable Development in China Based on the TOPSIS-Entropy Method, Sustainability 8(8) (2016), 746–758.

10.

Yin

, Cheng

, Wang

and Shao

, ‘Big data’ for pedestrian volume: Exploring the use of Google Street View images for pedestrian counts, Applied Geography 63 (2015), 337–345.

11.

Niu

, Zhang

, Shao

and Cheng

, Web-based geological hazard monitoring in the three gorges area of China, Photogrammetric Engineering and Remote Sensing 73(6) (2007), 709–719.

12.

, Yao

, Shao

and Wang

, From digital Earth to smart Earth, Chinese Science Bulletin 59(8) (2014), 722–733.

13.

Shao

and Li

, Image City sharing platform and its typical applications, Science China-Information Sciences 54(8) (2011), 1738–1746.

14.

and Shao

, The new era for geo-information, Science in China Series F-Information Sciences 52(7) (2009), 1233–1242.

15.

Pastorelli

, Matteo

and Pietro

, Os-assisted task preemption for hadoop, Distributed Computing Systems Workshops (ICDCSW), 2014 IEEE 34th International Conference. (2014), 14–20.

16.

Hijawi

H.M.

and Saheb

, Sequence Pattern Mining in Data Streams, Computer and Information Science 8(3) (2015), 64–68.

17.

Jia

, An Interactive Sequential Patterns Mining Algo-rithm Based on Frequent Sequence Tree, Computer Technology and Development 5 (2012), 10–17.

18.

Wang

, et al., TS-Hadoop: Handling access skew in Map Reduce by using tiered storage infrastructure, Information and Communication Technology Convergence (ICTC), 2014 International Conference on. IEEE, (2014).

19.

Verma

, Bankim

and Atul

, Big data analysis: recommendation system with Hadoop framework, Computational Intelligence & Communication Technology (CICT), 2015 IEEE International Conference. (2015), 22–25.

20.

Angelo

, Palhares

, Filho

and Maia

, A new fault classification approach applied to tennessee eastman benchmark process, Applied Soft Computing 49 (2016), 676–686.

21.

Cerrada

, Sánchez

, Pacheco

and Cabrera

, Hierar-chical feature selection based on relative dependency for gear fault diagnosis, Applied Intelligence 44 (2016), 687–703.

22.

Parimala

and Sathiyabama

, SPMLS: An Efficient Sequential Pattern Mining Algorithm with candidate Generation and Frequency Testing, International Journal on Computer Science and Engineering 4(4) (2012), 590–601.

23.

Schweizer

, Zehnder

, Wache

, Witschel

H.F.

, Zanatta

and Rodriguez

, Using Consumer Behavior Data to Reduce Energy Consumption in Smart Homes: Applying Machine Learning to Save Energy without Lowering Comfort of Inhabitants, 2015 IEEE 14th International Conference on Machine Learning and Applications (ICMLA), (2015), 1123–1129.

Evaluation method of sponge city potential based on neural network and fuzzy mathematical evaluation

Abstract

Keywords

1 Introduction

2 Related work

3.1 BP neural network

4.1 Model establishment

4.2 Model training

5.1 Sponge indicator monitoring and tracking

5.2 Judgment matrix

Table 4 A-B Judgment matrix Matrix B1 B2 B3 Factor B-1 1 3 1 0.2150 B-2 3 1/3 3 0.5764 B-3 1/5 1/3 1 0.1428

Footnotes

Acknowledgment

References

Table 4
A-B Judgment matrix

Matrix B1 B2 B3 Factor

B-1 1 3 1 0.2150

B-2 3 1/3 3 0.5764

B-3 1/5 1/3 1 0.1428