Optimization-assisted water supplement mechanism with energy efficiency in IoT based greenhouse

In this section of the paper, we describe our proposed objective function formulation for optimization in detail. An optimization algorithm is a procedure which is activated iteratively by comparing existing solutions until an optimum, or a satisfactory solution is found. Optimization algorithms have been widely used in various areas for minimizing the production cost or maximizing the profit [33, 34]. The objective function is the basement of optimization algorithms, which is formulated for finding the optimal solution for a specific domain [35]. Real-world problems consist of multiple objectives, and the overall system requires to be optimized simultaneously [36, 37] using multi-objective optimization. Multi-objective optimization based applications have been applied in many areas, such as engineering, agriculture, logistics, economics, and to name a few [38–40]. We have formulated the multi-objective optimization algorithm for keeping the water and moisture level at an optimal level in order to achieve optimal resource utilization by avoiding overfilling the tank and overwatering the greenhouse. However, the proposed objective function can be applied to the other environmental parameters such as temperature, humidity, pH, and CO₂ level. Let us assume greenhouse (G) and water tank (T) actual parameter readings which are given as GTP a = [ W a , M a ](1)

W _a depicts actual water level, while actual moisture level is considered as M _a . As can be seen from Fig. 1 that, there can be several soil moisture sensors in different locations of the greenhouse because it is difficult to decide actual moisture values based on one or two moisture sensors in the complex greenhouse. Various moisture sensors in the greenhouse environment are labeled as M_a1, M_a2,M_a3,  … M_aN. Our target is improving the overall condition of the greenhouse and water storage environment, so we calculate the average actual soil moisture level (M _A ) and the average water level (W _A ) according to the different sensor nodes’ actual values with the following equation. In our proposed system, there is only one water level sensor so it can be assigned as W _A  = W _a W A = W a M A = ( M a 1 + M a 2 + M a 3 + … + M aK ) / K(2)

Where K describes the number of sensors that are used for measuring the actual soil moisture level. Through the equation mentioned above (2) the average actual sensing values (GTP _A ) can be calculated for each sensing values. The list of symbols used in this formulation is provided in Table 1.

User/Farmer can select the most desirable values for both parameters as GTP d = [ W d , M d ](3)

Let W _min , M _min and W _max , M _max denote the most acceptable the min and the max user-desired ranges for the water and soil moisture levels, which can be selected by the user/farmer according to the greenhouse and water tank condition. W d = [ W min , W max ] M d = [ M min , M max ](4)

The max value describes a highly desirable point for water and soil moisture, and the min value is the least desirable points. If there were not any problems with energy cost, the user would obviously want to set the maximum point for parameters to boom the production. For the compactness, the range length of the minimum and the maximum parameters described as Δ W = | W max - W min | Δ M = | M max - M min |(5)

Suppose that the following equation determines the optimal setting for a compromise between the desired environmental parameters and energy consumption: GTP o = [ W o , M o ](6)

Optimal water and soil moisture levels come between the minimum and the maximum user-desired parameters as described in Equation 7, W o ∈ [ W min , W max ] M o ∈ [ M min , M max ](7)

The total consumed energy for maintaining optimal parameters for the water, and soil moisture levels are described as follows E opt = E W o + E M o(8)

Where E _{W o} , E _{M o} present consumed energy with optimal environmental parameters, which is calculated with the below-given equation. E W o = { P wp 1 × ( W o - W A ) if W A < W min 0 if W max < W A E M o = { P wp 2 × ( M o - M A ) if M A < M min 0 if M max < M A(9)

Where P_wp1 and P_wp2 show the consumed energy by the water tank pump and the greenhouse water pump, per unit change in water level. Above is the total required power for maintaining optimal energy settings for water supply. We can assume the possible minimum and maximum power consumption as below E min = E W min + E M min E max = E W max + E M max(10)

Where E W min = { P wp 1 × ( W min - W A ) if W A < W min 0 if W max < W A E W max = { P wp 1 × ( W max - W A ) if W A < W min 0 ) if W max < W A E M min = { P wp 2 × ( M min - M A ) if M A < M min 0 if M max < M A E M max = { P wp 2 × ( M max - M A ) if M A < M min 0 if M max < M A(11)

It is clear that user/farmer aims to achive a highly productive greenhouse environment with less energy utilization. G _p and G _e present the gain for optimal environmental settings and energy consumption, respectively. User preferences are described with α _p and α _e , if the user wants to give more attention to optimal water level and soil moisture parameters, then the user needs to give more weights to the α _p . If the user aims to increase energy efficiency, then more weight is needed to α _e values, as formulated in Equation 12. These desired α _p and α _e objective weights are given in the range [0, 1]. If the user pays more attention to the optimization of energy consumption, then he or she can give 0.9 value for the α _e , and 0.1 value for the environmental parameter weight. If the energy consumption is not a problem for the user, then the user can boost the production by giving high weight to the optimal environment (α _e  = 0 and α _p  = 1). In this work, we consider the objective weights for the optimal environment and energy consumption as an equal by assigning them α _p  = 0.5 and α _e  = 0.5 values. Max ( α p . G p + α e . G e ) G p = ∑ X ∈ { W , M } w X . ( 1 - D X 2 )(12)

D _M and D _W illustrate the deficiency for attaining the optimal values for the soil moisture and water levels. Assigned weights for the water level and soil moisture are described with w _W and w _M , respectively. According to its significance for crop production such that w _W  + w _M  = 1. In order to increase the gain for the development of the optimal desired environment G _p , the deficiency component of each defined parameter must be minimized by using the below-given equation. D W = { W max - W o Δ W if W A < W min W o - W min Δ W if W max < W A D M = { M max - M o Δ M if M A < M min M o - M min Δ M if M max < M A(13)

To set up optimal water and soil moisture levels to the tank and greenhouse, deficiency component D _X is needed to be minimized. If W _A  < W _min then desired optimal environmental parameters are needed to fix W _o  ≈ W _max . If W _o  → W _max then D _W  → 0 which is likely to support in increasing the desired G _p . The energy consumption gain (G _e ) can be computed with Equation 14, G e = 1 - ( E opt - E min E max - E min ) 2(14)

For maximizing energy efficiency, the minimization of energy utilization is needed E _opt . When E _opt  → E _min then G _e  → 1 that supports the maximization of the energy-efficiency G _e . Therefore, the final objective function for optimization becomes

Max ( α p . ∑ X ∈ { W , H } w X . ( 1 - D X 2 ) + α e . ( 1 - ( E opt - E min E max - E min ) 2 ) )(15)

Constraints: W A ⩽ W min ⩽ W A ⩽ W max for water supplying case to the water tank W min ⩽ W o ⩽ W max < W A for stopping water supplying case to the water tank M A ⩽ M min ⩽ M o ⩽ M max for water supplying case to the greenhouse M min ⩽ M o ⩽ M max < M A for stopping water supplying case to the greenhouse 0 < E min ⩽ E o ⩽ E max(16)

As we mentioned in Section 2, the rule-based system aims to gain knowledge from a human expert and convert this into a number of hardcoded rules to apply to the input values. The widely used basic form of the rule-based system is conditional statements such as if A, then do X, else if B, then do Y. Rule-based systems can be easily adapted and acceptable for some various problems [41]. Using a rule-based system is dependent on the problem area. If the problem area is not remarkably too broad if-then rule statements are appropriate, wide problem area requires too many if-then rules., As a consequence, the system might face some difficulties for calculating and it influences on decreasing performance speed [42]. In our proposed approach, we applied the rule-based control model for computing the optimal water and soil moisture levels comparing with actual sensing data and optimized data. Control module sets flow rate and operational duration for water pumps; Fig. 3 describes an abstract diagram of a rule-based controller; we have input parameters, rule-based controller and output.

OPEN IN VIEWER

Fig. 3

Rule-based control system conceptual diagram for water supplement.

There are four input parameters to the rule-based controller, namely actual sensing and optimized values for the water and soil moisture level, and the rule-based controller sets the flow rate and operational duration for each water pump. Figure 4 illustrates the detailed flow diagram of the proposed rule-based control module. It can be seen that there are two possible cases for the water supplement to the tank through the water pump 1. The first case is when the actual average water level (W _A ) is greater than or equal to the optimal water level (W _o ), then activation is not needed to the water pump actuator. The other case is when the first case fails, and the actual average water level is lower than the optimal water level, then the activation is needed to the water pump. Water pumps’ operational duration (dur) and flow rate (rate) set according to the difference between the actual average and optimal water levels. For the finding flow rate and operational duration, we developed W and M related formula. As can be seen, when we divide into five the multiplication of the water pump operational duration and water pump flow rate, the results need to be higher than W. The same applies to the optimization of the soil moisture level in the greenhouse using water pump 2. If the average soil moisture is greater than or equal to the optimized soil moisture level, then water pump 2 is not activated. Whenever the actual soil moisture level is less than optimal level, then the M is used to calculate the water pump’s flow rate and operation duration to activate the pump.

OPEN IN VIEWER

Fig. 4

Flow diagram of a rule-based control module.

5.1 Deployment of IoT based data acquisition unit

For the IoT based data acquisition, optimization, and control modules, we developed the Apache HTTP based server, and sensing values from the environment can be stored to the MySQL database using an HTTP POST request, as shown in Fig. 8. Collected soil moisture (M₁, M₂) and water level (W) sensing values were saved to the database, and these values were described the system interface in visual charts. Collected data was uploaded to calculate optimal soil moisture and water level values using the proposed objective function. According to the actual and optimized environmental values, the control module calculated the flow rate and operational duration for the water pumps; then, the proposed system sent On or Off commands to the IoT Gateway in the required time for the activation or deactivation of actuators. The measurement units for the soil moisture and water level values were considered in %and mm, respectively.

OPEN IN VIEWER

Fig. 8

IoT based Data Acquisition Unit.

It is important to note that the farmer needs to pay attention to install the driplines to spread the same amount of water for each location through water pipelines. Pumping out various amounts of water to different locations is likely to decrease the effectiveness of the proposed system. Let us imagine one pipeline has more drip holes, and then the more water spreads to this crop line, the soil moisture sensor (M₁) detects the actual soil moisture level as 90%. If in another greenhouse location, the water pipeline has fewer drip holes, then obviously the soil moisture level becomes less, such as the M₂ sensor detects soil moisture as 10%. After that, the actual average moisture level is calculated as M_A=(M_1  + M₂)/2, and the result becomes (90%+10%)/2 = 50%, according to Section 3.1 equation 2. It seems 50%of soil moisture level is enough for greenhouse crop production; however, different locations have various soil moisture levels, which can affect badly for greenhouse crop production.

Figure 9 describes the summary of the actual collected data values during October 2019 with the user desired minimum and maximum points. The X-axes describe 31 days of data collection period; however, the experiments were carried out using one month’s data in which water level and soil moisture sensors had collected this data every hour during the mentioned period. Figure 9(a) shows that minimum and maximum user-desired points were 60%and 80%, respectively. However, actual soil moisture levels were below the minimum user desired threshold in most of the days. If actual soil moisture levels come between user desired points, then we do not need to activate the greenhouse water pump to increase the soil moisture level. The same consideration also applies to the water level values in the water tank. The optimal water level for the tank is between 250 mm and 300 mm. The actual water level data fluctuated throughout the given period. On October 3, 7, and 29 data instances, the actual water level values were acceptable for the user required points. However, in some days, the water level data instances were below the user-desired values. Overwatering also can be seen in 13, 14, 23, and 25 October.

OPEN IN VIEWER

Fig. 9

Actual data instances with user desired points during the month (October 2019).

The user/farmer knows which soil moisture and the water level are the most desired levels and which one is the least acceptable level for keeping the soil moisture and the water level at the optimal point. If the cost of energy power is not an issue, the user sets the maximum desired level and boost the production of the greenhouse. The setting of maximum parameters for the greenhouse is likely to require high energy consumption. For this reason, creating a stable condition between energy consumption and the environment is essential. For the model simplicity, visibility, and straightforward explanation, the optimization module results were described using 150 data instances, as presented in Fig. 10. Actual soil moisture and water level lines show that only some points are acceptable for user desired settings, and other sensing values are outside of the user proposed ranges.

OPEN IN VIEWER

Fig. 10

Optimization module results. (a) Soil moisture level optimization for greenhouse (b) Water level optimization for water tank.

For the sake of comparison of the effectiveness of the proposed objective function, we considered a baseline approach for improving the water and soil moisture levels. The baseline scheme does not include optimization formulations and selects the midpoint of the user-specified range for each parameter if the actual sensing values are outside of the user desired ranges, as shown in Equation 17,

B W = ( W min + W max ) / 2 B M = ( M min + M max ) / 2(17)

If the actual soil moisture and water levels are equal to 30%and 120 mm, respectively, then the baseline scheme results become ((60%+80%)/2)=70%and ((250 mm+300 mm)/2)=275 mm. The baseline scheme is relatively simple, the actual sensing values are continuously monitored using sensors. As the sensing values are outside of user-desired set points, then the water pump actuators are operated automatically until the water and moisture levels reach the midpoint of user-desired ranges. When the user clicks the baseline scheme button from the application, the system calculates the baseline scheme-based optimization results. The formulated objective function for the optimization scheme calculates the optimal soil moisture and water levels for the greenhouse and water tank according to the sensing data. Optimization scheme results represent that optimized sensing values always came inside of the user desired minimum and maximum values. Furthermore, the optimization scheme based results were more optimal comparing with baseline scheme results.

In this section, we discussed the rule-based control module results of the water pumps. According to the flow rate, water pumps are divided into single and variable flow rate pumps. A single-flow rate pump activates at one fixed rate, and a variable flow rate pump can be programmed to run at many different rates [43]. A high flow rate consumes more energy in comparison with the less pumping rate. On the other hand, it is favourable to run a water pump at a higher flow rate to fill the water tank in a short time. In the greenhouse and water tank environments, the quantity of the required energy of the pump activation can be decreased by changing the water pump flow rate, but it demands to activate the pump for additional time, more precisely, if the pump needs to be activated with the flow rate of 60 ft³/min for 8 hours in a day, then operating the same pump with 30 ft³/min requires 16 h of activation in one day. For rule-based control setup analysis, we considered three variable flow rates, namely, minimum, average, and maximum flow rates. Table 4 describes the required energy consumption and operational duration for each pump according to the water pumps’ flow rates. As can be seen, if the water tank pump is activated with a minimum flow rate, then it spends less energy (600–1000 watts) but more time (600–1000 minutes) to increase the actual water level to the optimal level in the tank. If the greenhouse pump is activated with a maximum flow rate, then it requires more energy consumption (210–300) with less operational duration (5–10 minutes) to provide an optimal moisture level to the greenhouse crops.

The rule-based controller module is used to control the operation of water pumps based on optimized parameter values and actual sensing values. When the actual sensing values are lower than minimum desired settings in per minute, the optimization scheme calculates the optimal parameter values for this sensing values and sends these levels to the rule-based control module. The rule-based control module selects the flow rate and operational duration for the water pumps based on actual and optimal sensing values. The water pump is activated automatically until the actual water level reaches the optimal point. If the current water level reaches the optimal level, then the water pump is deactivated. Section 3.2 includes a detailed explanation of the rule-based control module.

Figure 11(a) and (b) describes pumps’ flow rate and operational duration results in the greenhouse and water storage environments, respectively. The water pump actuator was activated in three flow rates, namely minimum, average, and maximum, to increase the actual soil moisture and water level sensing data in both the greenhouse and water tank. As we mentioned above, the soil moisture level of the greenhouse was measured in percentage, and the water level of the tank was considered in mm. As expected, the activation of water pumps with various flow rates required a different operational duration to increase the actual environmental sensing values to the optimal parameters. For example, the greenhouse water pump operational duration was 15, 13.6, and 5.7 minutes based on minimum, average, and maximum flow rates, respectively, between 1 and 5 sensing values. The actual soil moisture levels were below than user-desired minimum range in most of the time, whereas, actual water levels of the tank were between the user desired minimum and maximum ranges. As a result, the greenhouse pump required a more activation period compared with the water tank pump.

OPEN IN VIEWER

Fig. 11

Rule-based control module results: (a) Greenhouse water pump (b) Tank water pump.

According to the water pumps’ flow rate and operational duration results, we were able to compute the overall activation period for each water pump, as described in Table 5. Overall operational duration results are different based on minimum, average, and maximum flow rates. To optimize the actual soil moisture level, the greenhouse water pump required 1900, 1401.7, and 887.5 minutes with the minimum, average, and maximum flow rates, respectively. Whereas, activation durations were 1456.2 (minimum flow rate), 1084.6 (average flow rate), and 691.8 (maximum flow rate) minutes for running the water tank pump to provide an optimal water level to the tank based on actual water level.

There is a significant relationship between water pump flow rate and energy consumption. In order to conduct energy consumption analysis, we need to assign power ratings to the water pumps. As we already explained above, we utilized a variable flow rate water pump which can operate with different power consumption. Table 4 describes the required power ratings for the water pumps’ activation according to the various flow rates. If we take the greenhouse pump as an example, with the minimum, average, and maximum flow rate, the pump needs 75, 165, 255 watts, respectively. We could compute the consumed energy for each pump based on the overall operational duration results and flow rates. Figure 12 presents the quantity of the pumps’ consumed energy in kWh according to the various flow rates.

OPEN IN VIEWER

Fig. 12

Actuators’ energy consumption results based on various flow rates.

It can be seen from the chart; the water tank pump spent more energy compared with the greenhouse pump due to the actuator activation requirement. The water tank pump consumed 1106.8, 1301.5, and 1164.9 kWh energy with maximum, average, and minimum flow rates, respectively. During the maximum flow rate activation period, the water tank pump spent the least energy. Average and minimum flow rate activation required more energy than the maximum flow rate. The greenhouse pump was spending 142.5, 231.3, and 226.3 kWh energy with the minimum, average, and maximum flow rates, respectively, to increase actual soil moisture level to the optimal in terms of 150 sensing data instances which were collected in every 30 minutes from the real greenhouse environment. Optimal energy consumption was achievable with activating the greenhouse pump with a minimum flow rate.