A farmland irrigation method based on the multilayer perceptron model of WOA-LSO algorithm

Abstract

The present invention discloses a farmland irrigation method based on a multilayer perceptron model of the WOA-LSO algorithm, and relates to the technical field of agricultural irrigation. The method comprises: obtaining meteorological data factors of multiple samples and determining a plurality of factor combinations. The factor combination most correlated with soil transpiration is screened out through correlation testing, and then the number of nodes in the input layer, output layer, and hidden layer of the multilayer perceptron (MLP) model is determined; an MLP prediction model optimized based on the WOA-LSO algorithm is constructed, and the algorithm obtains the optimal sequence and assigns it to the model by optimizing the weight and threshold sequence of the MLP; the target meteorological data factors of the study area are input into the model to predict the soil transpiration in the future period; the effective soil water content and transpiration are input into a fuzzy inference system to obtain an irrigation coefficient, and the amount of irrigation water required for the farmland is calculated accordingly. The present invention solves the problem of inaccurate soil transpiration prediction caused by scarce and incomplete data in plateau areas, and significantly improves the accuracy of soil transpiration prediction.

Description

Technical Field

[0001] The present application relates to the field of agricultural irrigation technology, and in particular to a farmland irrigation method based on a multilayer perceptron model of a WOA-LSO algorithm. Background Art

[0002] Due to their unique geographical location and climatic conditions, plateau regions face severe water shortages. Many plateau regions are often classified as semi-arid, with extremely fragile ecosystems, exacerbated by water scarcity. This water shortage significantly constrains local agricultural development, particularly irrigation water availability, which has become a key factor impacting agricultural production.

[0003] Existing crop growth analysis methods based on satellite data and machine learning combine remote sensing data with ground-based measurements to establish a relationship between transpiration and vegetation indices. Machine learning methods are then used to optimize transpiration prediction models, enabling real-time monitoring of crop growth. However, this approach fails to fully account for the complexity of various environmental factors and meteorological data in plateau regions, resulting in inaccurate transpiration predictions and thus impacting the effectiveness of irrigation decisions.

[0004] Therefore, there is an urgent need for an intelligent irrigation method to accurately predict irrigation volume under the complex environmental factors and meteorological data in plateau areas. Summary of the Invention

[0005] Based on this, it is necessary to provide a farmland irrigation method based on a multi-layer perceptron model of the WOA-LSO algorithm to address the above technical problems, so as to solve the problem of difficulty in accurately predicting irrigation volume under the complex environmental factors and meteorological data in plateau areas.

[0006] This manual adopts the following technical solutions:

[0007] This specification provides a farmland irrigation method based on a multilayer perceptron model of the WOA-LSO algorithm, including:

[0008] Obtaining meteorological data factors; the meteorological data factors include: maximum temperature, average temperature, minimum temperature, average relative humidity, day number, sunshine hours and wind speed;

[0009] According to a correlation test between meteorological data factors and soil transpiration, the number of nodes in the multilayer perceptron network, including the input layer, the hidden layer, and the output layer, is determined, and a multilayer perceptron prediction model optimized based on the WOA-LSO algorithm is constructed to predict soil transpiration; the multilayer perceptron prediction model optimized based on the WOA-LSO algorithm encodes the weight sequence and the threshold sequence of the multilayer perceptron model through the WOA-LSO algorithm, and iteratively calculates the fitness value of the WOA-LSO algorithm until the maximum number of iterations of the WOA-LSO algorithm or the preset value of the fitness value is reached, thereby obtaining the optimal weight sequence and the optimal threshold sequence of the multilayer perceptron model and assigning them to the multilayer perceptron model;

[0010] Inputting the meteorological data factors into the multilayer perceptron prediction model based on the WOA-LSO algorithm for training to obtain a multilayer perceptron prediction model based on the WOA-LSO algorithm; and using the multilayer perceptron prediction model based on the WOA-LSO algorithm to predict soil transpiration to obtain soil transpiration;

[0011] The existing effective soil moisture content and predicted soil transpiration are input into the fuzzy inference system to obtain the irrigation coefficient; and the irrigation coefficient is used to formulate an irrigation strategy to calculate the amount of irrigation water required for farmland.

[0012] Preferably, determining the number of nodes in the multilayer perceptron network including the input layer, the hidden layer, and the output layer specifically includes:

[0013] Determine the number of nodes in the input layer of a multilayer perceptron network, including:

[0014] Using a multilayer perceptron model, testing the correlation between each meteorological data factor and soil transpiration, determining the meteorological data factors involved in the final soil transpiration prediction, and determining the number of nodes in the input layer of the multilayer perceptron model based on the number of meteorological data factors involved in the final soil transpiration prediction;

[0015] Determine the number of nodes in the hidden layer of a multilayer perceptron network, including:

[0016] The range of the number of nodes in the hidden layer is determined by an empirical formula. Within this range, the network error of the multilayer perceptron network corresponding to each number of nodes is repeatedly tested. The network error of the multilayer perceptron network corresponding to each number of nodes is compared, and the number of nodes with the smallest error is selected as the number of nodes in the hidden layer.

[0017] Determine the number of nodes in the output layer of a multilayer perceptron network, including:

[0018] The number of prediction targets is the number of nodes in the output layer of the multilayer perceptron network.

[0019] Preferably, determining the number of nodes in the input layer of the multilayer perceptron network comprises the following steps:

[0020] The raw meteorological data factors are grouped and input into an existing multi-layer perceptron prediction model to predict soil transpiration, thereby obtaining a prediction result of soil transpiration;

[0021] According to the prediction results of soil transpiration, the correlation test between each original meteorological data factor and soil transpiration is carried out, and based on the test results, the meteorological data factors that are finally involved in the soil transpiration prediction are determined;

[0022] The grouping of the original meteorological data factors includes:

[0023] The first group of meteorological data factors includes: all meteorological data factors, meteorological data factors excluding the lowest temperature item, and meteorological data factors excluding the highest temperature and the lowest temperature;

[0024] The second set of meteorological data factors includes: based on the meteorological data factors after excluding the maximum temperature and the minimum temperature, four of the remaining five meteorological data factors are randomly arranged and combined, and all combinations thereof constitute the second set of meteorological data factors;

[0025] The third group of meteorological data factors includes: on the basis of determining the best combination of the second group of meteorological data factors, randomly eliminating one meteorological data factor, and all the resulting combinations are the third group of meteorological data factors.

[0026] Preferably,

[0027] The first set of meteorological data factors is used to test the correlation between the maximum temperature, average temperature and minimum temperature and soil transpiration;

[0028] The second group of meteorological data factors is used to test the correlation between average temperature, average relative humidity, day number, wind speed and sunshine hours and soil evaporation;

[0029] The third group of meteorological data factors is used to determine the meteorological data factors with the strongest correlation.

[0030] Preferably, determining the number of nodes in the hidden layer of the multilayer perceptron network specifically includes:

[0031] According to the relationship between the number of hidden layer nodes, the number of input layer nodes, and the number of output layer nodes, the formula is:

[0032] ;

[0033] Where, is the minimum number of hidden layer nodes, is the maximum number of hidden layer nodes, is the number of input layer nodes, is the number of output layer nodes, is the number of hidden layer nodes;

[0034] Determine the range of the number of hidden layer nodes ;

[0035] In the range of the number of hidden layer nodes The error of the multilayer perceptron network is calculated in sequence, and the number of nodes corresponding to the minimum value of the network error is selected as the number of hidden layer nodes.

[0036] Preferably, the WOA-LSO algorithm specifically includes:

[0037] The WOA-LSO algorithm specifically includes:

[0038] Initialize algorithm parameters, including:

[0039] Initialize the number of whale populations and the maximum number of iterations. The formula is:

[0040] ;

[0041] Where, It is a linearly decreasing parameter used to control the balance between global search and local search; is the current iteration number, is the maximum number of iterations;

[0042] ;

[0043] ;

[0044] ;

[0045] Where, and is a random number between [0, 1], parameter A Used to control the direction of movement of the whale, parameters C Used to control the whale's moving distance, p Used to select predation strategies;

[0046] Update the position of each whale and the current optimal solution for global search. The formula is:

[0047] ;

[0048] Where, For the The current optimal solution at the time of update, For the The optimal solution at the time of update, is the initial random solution, is the initialization parameter, is the search range, The optimal solution at the time of update,

[0049] Perform random perturbations on the current optimal solution and update the current optimal solution again to perform local search. The formula is:

[0050] ;

[0051] Where, is the latest solution after random perturbation, For the The current optimal solution at the time of update, is the disturbance coefficient, express A random number between

[0052] Update the optimal solution, the formula is:

[0053] ;

[0054] Where, is the current optimal solution, is the latest solution after random perturbation, For the The current optimal solution at the time of update, for The fitness value of for The fitness value of .

[0055] Preferably, encoding the weight sequence and threshold sequence of the multilayer perceptron model by the WOA-LSO algorithm specifically includes:

[0056] Assume that the number of weights of the multilayer perceptron model is n , the threshold number is m , then the weight sequence of the model is , the threshold sequence of the model is ;

[0057] The model's weight sequence and threshold sequence are encoded as n+m dimensional vector, specifically:

[0058] ;

[0059] Where, is the encoding sequence of model weights and thresholds, is the weight sequence of the model, is the threshold sequence of the model;

[0060] Preferably, obtaining the optimal weight sequence and threshold sequence of the multilayer perceptron model specifically includes:

[0061] Initialize the population individual positions, population size and maximum number of iterations of the WOA-LSO algorithm; the solution corresponding to the population individual positions is a coding sequence of weights and thresholds of the multilayer perceptron model;

[0062] The mean square error is selected as the fitness function of the WOA-LSO algorithm, the positions of the individuals in the population and their corresponding solutions are iteratively updated, and the fitness value of the WOA-LSO algorithm is iteratively calculated;

[0063] When the fitness value of the WOA-LSO algorithm reaches the maximum number of iterations or the preset value of the fitness value, the iteration stops, and the optimal position of the population individual and its corresponding optimal solution are obtained. The corresponding optimal solution is the optimal weight sequence and optimal threshold sequence of the multilayer perceptron model.

[0064] Preferably, the fuzzy inference system comprises: an input layer, a fuzzification layer, a rule layer, an aggregation layer and a defuzzification layer;

[0065] The input layer has two nodes for inputting the existing soil effective water content and the predicted soil transpiration;

[0066] In the fuzzification layer, the input existing soil effective water content and predicted soil transpiration are fuzzified using a Gaussian membership function, that is, the actual input set is converted into a fuzzy set using a Gaussian membership function;

[0067] By using the predefined fuzzy rules in the rule layer, the fuzzy set is reasoned and the fuzzy output is obtained;

[0068] Through the aggregation layer, all fuzzy outputs are aggregated into a comprehensive fuzzy set;

[0069] Through the defuzzification layer, the fuzzy output is clarified into the watering coefficient using the centroid method.

[0070] Preferably, the irrigation coefficient is used to formulate an irrigation strategy to calculate the amount of irrigation water required for the farmland, specifically including:

[0071] Define the input variables of the fuzzy inference system and set the output variable of the system as the watering coefficient;

[0072] Establish a fuzzy rule base and use the training of fuzzy neural network to optimize the fuzzy rules and membership functions of the fuzzy inference system;

[0073] The soil transpiration predicted in real time by the multi-layer perceptron model optimized based on the WOA-LSO algorithm is input into the fuzzy inference system to predict the irrigation coefficient. The irrigation amount required for the farmland is calculated in combination with the irrigation quota formula to assist in irrigation decision-making.

[0074] At least one of the above technical solutions adopted in this specification can achieve the following beneficial effects:

[0075] This specification provides a farmland irrigation method based on a multilayer perceptron model of the WOA-LSO algorithm, and constructs a multilayer perceptron model optimized based on the WOA-LSO algorithm. First, the correlation test between meteorological data factors and soil transpiration is performed to determine the number of nodes in the multilayer perceptron network, including the input layer, hidden layer, and output layer. Since there are a large number of meteorological factors that affect the prediction results and there are hundreds of input combinations, the present invention only considers a type of meteorological data factors that are highly correlated with soil transpiration as model input, thereby reducing the computational load of the model. By reducing the number of nodes in the network layer, the model is simplified and the prediction efficiency is improved. Secondly, the prediction model optimizes the model weights and thresholds through the WOA-LSO algorithm, and utilizes the global search capability of the WOA-LSO algorithm to effectively avoid the local optimal solution problem encountered by traditional machine learning models in the face of multiple environmental factors and the complexity of meteorological data in plateau areas.

[0076] In summary, given the complexities of multiple environmental factors and meteorological data in plateau regions, this paper simplifies the network structure by testing the correlation between meteorological data factors and soil evaporation. Furthermore, it optimizes the model's weights and thresholds using the WOA-LSO algorithm. This solves the problem of inaccurate soil evaporation predictions in plateau regions due to the complexity of environmental factors and meteorological data, significantly improving the accuracy of soil evaporation predictions. Furthermore, this paper constructs a fuzzy inference system that, based on accurate soil evaporation predictions, applies the predicted soil evaporation to calculate the required irrigation water for farmland. This provides strong support for irrigation decisions in plateau regions, effectively improves irrigation efficiency, and achieves effective water conservation, laying the foundation for sustainable agricultural development. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0078] Figure 1 A schematic diagram of a farmland irrigation method based on a multilayer perceptron model of the WOA-LSO algorithm provided in this specification;

[0079] Figure 2 This is a flow chart for predicting soil transpiration using a multilayer perceptron model based on the WOA-LSO algorithm for farmland irrigation provided in this specification;

[0080] Figure 3 The best network error graph provided for this manual;

[0081] Figure 4 The fuzzy irrigation decision-making system modeling flow chart provided in this manual. DETAILED DESCRIPTION

[0082] To make the purpose, technical solutions, and advantages of this specification more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this specification and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0083] The following describes in detail the technical solutions provided by various embodiments of the present application in conjunction with the accompanying drawings.

[0084] Figure 1 This is a flowchart of a multilayer perceptron model method based on the WOA-LSO algorithm in this specification, which specifically includes the following steps:

[0085] S101: Acquire multiple sample meteorological data factors.

[0086] Specifically, the multiple sample meteorological data factors include: maximum temperature, average temperature, minimum temperature, average relative humidity, day number, sunshine hours and wind speed.

[0087] Specifically, this embodiment uses the monthly meteorological data from July to November 2024 in a certain plateau area and the corresponding soil evaporation obtained through the actual PM equation as meteorological factor data. The system hardware includes a sensor installation module, a data acquisition module, a data analysis module, an irrigation plan setting module, an irrigation execution module, an effect evaluation module, a maintenance module and a user interaction module. It is characterized in that the sensor installation module includes a soil temperature sensor, a soil moisture sensor, a soil nutrient sensor, a light intensity sensor, a wind speed and direction sensor, and an atmospheric pressure sensor. All sensors can perform data transmission monitoring.

[0088] Additionally, we designed a visual decision support platform, providing an intuitive interface and data display to help users understand the system's decision logic and results. This will improve user acceptance and operational convenience, promoting technology adoption. Furthermore, this cloud platform will be connected to sensor data, providing users with a clear and transparent view of all sensor data.

[0089] S102: Based on multiple sample meteorological data factors, determine multiple sample meteorological data factor combinations, and perform correlation tests between each sample meteorological data factor combination and soil transpiration, determine the target sample meteorological data factor combination, and determine the number of nodes in the input layer, output layer, and hidden layer of the multilayer perceptron model based on the target sample meteorological data factor combination.

[0090] Optionally, the multiple sample meteorological data factor combinations include a first group of meteorological data factor combinations and a second group of meteorological data factor combinations; based on the multiple sample meteorological data factors, the multiple sample meteorological data factor combinations are determined, specifically including: based on the multiple sample meteorological data factors, the first group of meteorological data factor combinations includes all meteorological data factor combinations, meteorological data factor combinations excluding the lowest temperature item, and meteorological data factor combinations excluding the highest temperature and the lowest temperature; among the five meteorological data factors in the meteorological data factor combination excluding the highest temperature and the lowest temperature, four are selected for random arrangement and combination, and all their combinations are determined as the second group of meteorological data factor combinations.

[0091] Among them, the first group of meteorological data factors is used to test the correlation between the maximum temperature, average temperature and minimum temperature and soil transpiration; the second group of meteorological data factors is used to test the correlation between the average temperature, average relative humidity, day number, wind speed and sunshine hours and soil transpiration; the third group of meteorological data factors is used to determine the meteorological data factors with the strongest correlation.

[0092] Optionally, a correlation test is performed on each sample meteorological data factor combination and soil transpiration to determine the target sample meteorological data factor combination, including: inputting each meteorological data factor combination into the initial multi-layer perceptron model to obtain the predicted soil transpiration under each meteorological data factor combination; determining the correlation between each meteorological data factor combination and soil transpiration based on the predicted soil transpiration and actual soil transpiration under each meteorological data factor combination; obtaining the candidate meteorological data factor combination with the greatest correlation in the second group of meteorological data factor combinations, and successively eliminating each meteorological data factor in the candidate meteorological data factor combination to obtain a third group of meteorological data factor combinations; obtaining the correlation between each meteorological data factor combination in the third group of meteorological data factor combinations and soil transpiration; and determining the meteorological data factor combination with the greatest correlation among all meteorological data factor combinations as the target sample meteorological data factor combination.

[0093] Optionally, the number of nodes in the input layer, output layer, and hidden layer of the multilayer perceptron model is determined based on the target sample meteorological data factor combination.

[0094] Optionally, the number of sample meteorological data factors in the target sample meteorological data factor combination is determined as the number of nodes in the input layer; the number of nodes in the output layer is determined to be 1; and the number of nodes in the hidden layer is determined based on the number of nodes in the input layer and the number of nodes in the output layer.

[0095] Specifically, determining the number of nodes in the hidden layer according to the number of nodes in the input layer and the number of nodes in the output layer includes:

[0096] According to the relationship between the number of hidden layer nodes, the number of input layer nodes, and the number of output layer nodes, the range of the number of hidden layer nodes is determined by the formula:

[0097] ;

[0098] Where, is the minimum number of hidden layer nodes, is the maximum number of hidden layer nodes, is the number of input layer nodes, is the number of output layer nodes, is the number of hidden layer nodes;

[0099] In the range of the number of hidden layer nodes The prediction accuracy of the multilayer perceptron prediction model under each hidden layer node number is obtained in turn, and the number of nodes corresponding to the highest prediction accuracy is determined as the number of hidden layer nodes.

[0100] Specifically, choose a framework and tools. Select an appropriate deep learning framework and then define the model architecture. The input layer: define the number of input layer nodes based on the number of features. Hidden layers: select 12 hidden layers, with an appropriate number of nodes per layer. Output layer: use one node to output soil evaporation based on the prediction target, using a linear activation function. Train the model using the training set and monitor performance using the validation set. Set appropriate epochs and batch_size, and use early stopping to prevent overfitting. In the WOA-LSO algorithm, first set the whale population size to 30, the initial spectral population to 30, and the maximum number of iterations to 100 to limit the algorithm's run time. Next, define the search space, specifically the range of hyperparameters to be optimized, such as the learning rate and the number of hidden layer nodes. To evaluate the effectiveness of different hyperparameter configurations, define a fitness function, typically using a performance metric (such as mean squared error (MSE)) on the validation set. This fitness function guides the whale population to adjust its position in the search space to find the optimal solution by training the model based on the current hyperparameter configuration and calculating the loss on the validation set. Within the maximum number of iterations, the whale's position is updated, the fitness is calculated, and the optimal solution is maintained. The optimal learning rate and number of hidden layer nodes are found by iteratively optimizing hyperparameters.

[0101] Specifically, many factors influence crop reference transpiration. It is generally believed that seven factors—maximum temperature, average temperature, minimum temperature, average relative humidity, day number, sunshine hours, and wind speed—are the primary influencing factors. The number of input layer nodes should be selected based on minimizing experimental error and maximizing prediction accuracy. However, due to the large number of influencing factors and the hundreds of possible input combinations, we considered comparing the predicted results of highly correlated factors and selecting only one as the network input variable. This approach reduces the network's computational load and the number of inputs, while also simplifying the model and improving prediction efficiency.

[0102] Specifically, according to the analysis, since the three factors of maximum temperature, average temperature and minimum temperature are highly correlated, it should be confirmed that only one of the three factors can be retained as input to obtain a more accurate prediction value; secondly, the influence of five influencing factors such as average relative humidity, average temperature, sunshine hours, day number and wind speed on the predicted value of transpiration should be further considered; finally, according to the experimental results of the second step, we can eliminate two influencing factors from the factors that have a greater impact on the prediction results determined in the previous step, run the network, and compare several groups of experiments to observe whether the prediction results can meet the prediction requirements.

[0103] Specifically, the prediction results of the three groups of input methods are compared to determine the correlation between the maximum temperature, average temperature, and minimum temperature in the network model, including:

[0104] The second group, based on the first group (3), examined the effects of average temperature, average relative humidity, day number, wind speed, and sunshine hours on the output results. Four of the five influencing factors were selected, and the permutations and combinations were performed to compare the experimental results. Specifically: The four influencing factors, including average temperature, day number, sunshine hours, and wind speed, were used as input to establish a prediction model, and the experimental results were observed. The four influencing factors, including average relative humidity, day number, sunshine hours, and wind speed, were used as input to establish a prediction model, and the experimental results were observed. The four influencing factors, including average temperature, average relative humidity, day number, sunshine hours, and wind speed, were used as input to establish a prediction model, and the experimental results were observed. The four influencing factors, including average temperature, average relative humidity, day number, sunshine hours, and wind speed, were used as input to establish a prediction model, and the experimental results were observed. The four influencing factors, including average temperature, average relative humidity, day number, and wind speed, were used as input to establish a prediction model, and the experimental results were observed.

[0105] By comparing and analyzing the above experimental results, we can identify four factors that have a greater impact on the predicted value of transpiration.

[0106] The third group randomly eliminated one item based on the above results, ran the network, and observed the experimental results.

[0107] For convenience, the above nine combinations of different meteorological factors are recorded as Scheme 1, Scheme 2, Scheme 3, ..., Scheme 9, etc. By comprehensively comparing the experimental results of the above schemes, we can intuitively select the input combination that can achieve the forecast accuracy requirements and then determine the number of nodes in the input layer.

[0108] Specifically, based on meteorological data such as average temperature, average relative humidity, and wind speed, the number of nodes in the output layer of the predicted transpiration is 1. The number of nodes in the hidden layer is determined, specifically including:

[0109] First, determine an appropriate range of the number of hidden layer nodes through an empirical formula, then repeat the experiment one by one within this range, compare the size of the network error, and select the number of hidden layer nodes corresponding to the minimum error. In general, the number of hidden layer nodes has the following relationship with the number of input layer nodes and the number of output layer nodes. Suppose the number of nodes in the input layer, hidden layer and output layer of the three-layer neural network are 、 、 。

[0110]

[0111] Then the number of hidden layer nodes should be within the interval [a,b].

[0112] Based on this, we can derive the appropriate range of the number of hidden layer nodes corresponding to different numbers of input and output layer nodes, see Table 1;

[0113] Table 1 The appropriate range of the number of hidden layer nodes when selecting different numbers of input and output layer nodes

[0114]

[0115] When the number of input quantities is different or the number of input quantities is the same but the combination is different, the experiment should be repeated in the above range in sequence, and the number of hidden layer nodes corresponding to the minimum value of the network error should be taken.

[0116] Based on the meteorological data from July to November 2024, a prediction model was established with seven influencing factors, including maximum temperature, average temperature, minimum temperature, average relative humidity, day number, sunshine hours, and wind speed. At this time, the appropriate range of the number of hidden layer nodes is [4, 18]. Three experiments were conducted at each integer point in this interval, and the median of the three experimental errors was recorded as the experimental result. Figure 3 When the number of hidden layer nodes is 13, the network achieves the minimum error of 0.020307. Therefore, the number of hidden layer nodes should be 13. Repeating schemes 2-9 yields the results, see Table 2:

[0117] Table 2 Network prediction errors corresponding to different numbers of nodes

[0118]

[0119] See also Figure 2 The following is a flow chart for soil evaporation prediction. Establishing a soil evaporation prediction model primarily involves programming the MLP model and optimizing the model's weights and thresholds using the WOA-LSO algorithm. A three-layer MLP grid structure is generated through programming, with the network's weights and threshold sequences encoded. The WOA-LSO algorithm is then used to optimize the network's weights and thresholds. The MLP model is then assigned weights and thresholds that meet the required accuracy or achieve the maximum number of evolutionary generations. The trained MLP model is then used for prediction.

[0120] S103: Based on the multilayer perceptron model with the determined number of nodes, a multilayer perceptron prediction model optimized based on the WOA-LSO algorithm is constructed; the multilayer perceptron model optimized based on the WOA-LSO algorithm is obtained by optimizing the weight sequence and threshold sequence of the multilayer perceptron model through the WOA-LSO algorithm, obtaining the optimal weight sequence and the optimal threshold sequence and assigning them to the multilayer perceptron model.

[0121] Optionally, the weight sequence and threshold sequence of the multilayer perceptron model are optimized by the WOA-LSO algorithm, specifically including: initializing the population individual position, population size and maximum number of iterations of the WOA-LSO algorithm; the population individual position is a coding sequence obtained by encoding the weight sequence and threshold sequence of the multilayer perceptron model; selecting the mean square error of the multilayer perceptron model for predicting soil transpiration as the fitness function value of the WOA-LSO algorithm, iteratively updating the population individual position, and iteratively calculating the fitness function value of the WOA-LSO algorithm; when the WOA-LSO algorithm reaches the maximum number of iterations or the fitness function value is greater than a preset threshold, the iteration stops and the optimal population individual position is obtained, the optimal population individual position is the optimal coding sequence, and the optimal coding sequence is decoded to obtain the optimal weight sequence and the optimal threshold sequence.

[0122] Specifically, the coding sequence is:

[0123] ;

[0124] Where, is the coding sequence, is the weight sequence of the multilayer perceptron model, is the threshold sequence of the multilayer perceptron model, n and m are the number of weights and thresholds of the multilayer perceptron model, respectively.

[0125] Specifically, in the WOA-LSO algorithm, the whale population size is first set to 30, the initial spectral population is 30, and the maximum number of iterations is set to 100 to limit the running time of the algorithm. Next, the search space needs to be defined, including the range of hyperparameters to be optimized, such as the learning rate and the number of hidden layer nodes. In order to evaluate the effect of different hyperparameter configurations, a fitness function is defined, and the performance indicator of the model on the validation set (such as mean squared error, MSE) is usually used as the fitness value. This fitness function guides the whale population to adjust its position in the search space to find the optimal solution by training the model according to the current hyperparameter configuration and calculating the loss on the validation set. Within the maximum number of iterations, the position of the whale is updated, the fitness is calculated, and the optimal solution is maintained. By iteratively optimizing the hyperparameters, the optimal learning rate and number of hidden layer nodes are found.

[0126] S104: Input multiple target meteorological data factors of the study area into the multi-layer perceptron prediction model to obtain the soil transpiration amount in the future period.

[0127] Specifically, the inconsistency of the physical meaning and units of the input vector will affect the network training process. Therefore, the input vector and the expected output vector values ​​are normalized according to the formula.

[0128] ;

[0129] The anti-normalization formula is:

[0130] ;

[0131] Where, are the maximum and minimum values ​​in the original sample, For the original sample and the corresponding processed data, according to the characteristics of the reference crop transpiration prediction problem, the following indicators are selected for the established MLP, standard deviation, and the formula is:

[0132] ;

[0133] The average value of the absolute value of the relative error is:

[0134] ;

[0135] The prediction qualification rate is set as the absolute value of the relative error The percentage of samples in the total number of samples; are the coefficients of the linear regression equation; is the coefficient of determination, that is, the square of the correlation coefficient; in order to verify the stability of the model, we repeated the experiment independently and recorded the prediction relative error, prediction standard deviation, and prediction pass rate of each experimental result.

[0136] Specifically, seven influencing factors were input: maximum temperature, average temperature, minimum temperature, average relative humidity, day number, sunshine hours, and wind speed. An MLP model was established with a 7-13-1 network structure. The WOA-LSO algorithm was used to optimize network weights and thresholds, with the fitness function taking the sum of the absolute values ​​of the training errors. The trained MLP network model was then used to predict transpiration of a reference crop. The evaluation indicators and prediction results are as follows:

[0137] Standard Deviation: ; Average value of absolute value of relative error: ; The pass rate is 94.36%; the coefficient of the linear regression equation is =0.987, square of correlation coefficient ;

[0138] In summary, the results of schemes 2-9 are as shown in Table 3:

[0139] Table 3 Prediction accuracy results

[0140]

[0141] Comparing the prediction results of Schemes 1, 2, and 3 shows that all three schemes perform well, with very close numerical values ​​for each evaluation indicator. This indicates a strong correlation between the three influencing factors: maximum temperature, average temperature, and minimum temperature. Simply selecting the average temperature among these three and combining it with the other four influencing factors as input yields highly accurate predictions. Comparing the prediction results of Schemes 4, 5, 6, 7, and 8 shows that, based on Scheme 3, by reducing the input factors to four, Scheme 7 achieves prediction results and indicators that are nearly identical to those of Scheme 3 and significantly superior to those of Schemes 4, 5, 6, and 8. This suggests that the day number has little influence on the prediction results, while the four influencing factors of average temperature, average relative humidity, wind speed, and sunshine hours are essential for predicting reference transpiration. Comparing the prediction results of Scheme 9 with those of Schemes 4, 5, 6, 7, and 8 further demonstrates the essential nature of these four factors when predicting crop reference transpiration.

[0142] In summary, Schemes 1, 2, 3, and 7 performed well. Comparative analysis revealed that Scheme 7 was selected as the evaporation prediction model to reduce network computational load, streamline the network structure, and improve model efficiency. This scheme, which only requires four influencing factors as input—average temperature, average relative humidity, sunshine hours, and wind speed—can generate relatively accurate predictions using the MLP model. The prediction results for Scheme 7 showed an average absolute relative error of 6.02%, a pass rate of 94.52%, and a standard deviation of 0.079.

[0143] S105: Input the effective soil moisture content and soil transpiration of the study area into the fuzzy inference system to obtain the irrigation coefficient, and calculate the irrigation amount required for the farmland based on the irrigation coefficient.

[0144] Optionally, the fuzzy inference system includes: an input layer, a fuzzification layer, a rule layer, an aggregation layer and a defuzzification layer; the inputting of the effective soil moisture content and the soil transpiration of the study area into the fuzzy inference system to obtain the irrigation coefficient specifically includes: inputting the effective soil moisture content and the soil transpiration into the fuzzification layer through the input layer; in the fuzzification layer, using a Gaussian membership function to fuzzify the effective soil moisture content and the soil transpiration to obtain an initial fuzzy set; in the rule layer, according to predefined fuzzy rules, the fuzzy set is inferred to obtain multiple fuzzy outputs; in the aggregation layer, all fuzzy outputs are aggregated into a target fuzzy set; in the defuzzification layer, the fuzzy outputs in the target fuzzy set are clarified using the center of gravity method to obtain the irrigation coefficient.

[0145] Optionally, the amount of irrigation water required for the farmland is calculated based on the irrigation coefficient, including: calculating the amount of irrigation water required for the farmland based on the irrigation coefficient in combination with the irrigation quota formula, the formula being:

[0146] ;

[0147] Where, The amount of irrigation water required for farmland, is the irrigation coefficient, is the soil evaporation, is the irrigation water utilization coefficient.

[0148] Specifically, in order to build a model based on Fuzzy Inference System (FIS), see Figure 4 The following is a flowchart for modeling a fuzzy irrigation decision-making system. The irrigation coefficient α is predicted using soil available water content (SAWC) and actual transpiration (AT) from soil (Poa pratensis) as input variables, processed using a fuzzy neural network (FNN). The following are the steps for building the model:

[0149] This model defines input and output variables. The input variables include soil available water content (SAWC) and soil evaporation (AT). The range of SAWC is [0, 100] mm, and the range of AT is [0, 10] mm / day. The output variable is the irrigation coefficient α, which ranges from [0, 1]. To facilitate understanding and analysis, the input and output variables are divided into linguistic variables. SAWC can be categorized into three levels: low, medium, and high. AT is similarly categorized into low, medium, and high. Finally, the irrigation coefficient α is also categorized into low, medium, and high. This categorization helps us better understand the relationships between different variables and their impact on the irrigation coefficient.

[0150] Specifically, in this model, the input variables soil available water content (SAWC) and actual soil transpiration (AT) are described using Gaussian membership functions, which effectively express the ambiguity and uncertainty of the input variables within a specific range. Meanwhile, the output variable, the irrigation coefficient α, is represented using a triangular membership function. A, b, and c are the vertices of the triangle, defining the different levels of the irrigation coefficient. This approach allows for a clearer representation and analysis of the variation of the irrigation coefficient under different input conditions. In this model, a fuzzy neural network (FNN) is used to construct a fuzzy inference system, combining the advantages of fuzzy logic and neural networks. The design proceeds as follows. First, the input layer has two nodes, corresponding to soil available water content (SAWC) and actual soil transpiration (AT), respectively. Next, in the fuzzification layer, the inputs are fuzzified using Gaussian membership functions, converting the actual input values ​​into fuzzy sets. Subsequently, in the rule layer, the fuzzy inputs are inferred according to predefined fuzzy rules (e.g., "If SAWC is high and AT is high, then α is high") to generate fuzzy outputs. Next, in the aggregation layer, the fuzzy outputs of all rules are aggregated to generate a comprehensive fuzzy output. Finally, in the defuzzification layer, the centroid method is used to convert the aggregated fuzzy output into a clear numerical value α. To enhance the user experience, a visualization interface is also provided. Users can graphically display the membership functions of input and output variables, allowing them to intuitively observe the fuzzy inference process and its output results under different input conditions.

[0151] Furthermore, when using a fuzzy neural network (FNN) to predict the irrigation coefficient α, the model must first be trained and validated using historical data to ensure its accuracy. Once training is complete and verified, real-time soil available water content (SAWC) and actual transpiration (AT) of Kentucky bluegrass can be input into the FNN model. The model then outputs the irrigation coefficient α to assist in farmland irrigation management. To enhance decision-making intuitiveness, the prediction results can be graphically displayed, and sensitivity analysis can be performed to examine the impact of different input variables on α. ​​Furthermore, the model can be optimized by adjusting the parameters of the Gaussian membership function and expanding the fuzzy rule base. Including additional influencing factors (such as temperature, humidity, and wind speed) can further improve prediction accuracy. Through these methods, the FNN model can effectively support farmland management decisions.

[0152] Furthermore, to address the complex irrigation challenges in the Tibetan Plateau, we designed an irrigation fuzzy decision-making system based on a fuzzy neural network (FNN) with an irrigation coefficient, an irrigation quota formula, and a water balance equation. This system comprehensively considers soil type, topography, and water source conditions when selecting suitable experimental plots. Multiple experiments were conducted under varying meteorological conditions (such as sunny, cloudy, and soil moisture changes after rainfall), recording irrigation effects and crop growth, including growth rate, yield, and water resource efficiency. In constructing the FNN, input variables (such as soil moisture, temperature, humidity, and precipitation) were first defined, and the output variable was set as the irrigation coefficient α. Next, a fuzzy rule base was established, and the FNN was trained using historical data to optimize the fuzzy rules and membership functions. Subsequently, the irrigation quota was calculated using the irrigation quota formula, incorporating key data such as actual crop evapotranspiration, precipitation, and soil moisture infiltration and consumption. The water balance equation was used to assess soil moisture changes and determine the need for irrigation. Finally, real-time data was input into the FNN to obtain the irrigation coefficient α, which was then combined with the irrigation quota formula to calculate the required irrigation volume. The system's irrigation decisions stabilize soil moisture within the expected range, ensuring healthy crop growth while also conserving water resources. The fuzzy irrigation decision system was used to determine irrigation conditions in the plateau region in November 2024. The results showed that the system's irrigation decisions stabilized soil moisture within the expected range, ensuring healthy crop growth while conserving water resources.

[0153] In addition, based on actual test results, the model and system are evaluated and improvement suggestions are made to further improve the efficiency and effectiveness of irrigation management. By continuously optimizing model parameters and decision rules, the adaptability and reliability of the system under different environmental conditions are ensured.

[0154] In summary, the present invention utilizes a multilayer perceptron model and embeds the WOA-LSO algorithm. Its global search capability effectively avoids the local optimal solution problem that traditional machine learning algorithms may encounter when data is scarce or incomplete. Furthermore, the combination of the WOA and LSO algorithms complements each other and addresses a number of data processing issues with existing technologies. Furthermore, the present invention incorporates fuzzy crop coefficients, allowing the calculation of soil evaporation and soil available moisture to account for multiple uncertainties. This approach effectively reduces the risks associated with data uncertainty in traditional irrigation decision-making, laying a solid foundation for future irrigation decisions. Comprehensively considering available moisture and water demand allows for accurate calculation of irrigation requirements, reducing water waste. Compared to traditional algorithms, this method achieves higher accuracy and faster data processing for soil evaporation prediction, enabling better irrigation decision-making and enhancing the adaptability and flexibility of irrigation decisions. Specifically designed for the extreme climate and water shortage challenges of the Qinghai-Tibet Plateau, the present invention incorporates efficient data collection methods to ensure feasibility and effectiveness in specific environments. The WOA-LSO algorithm is significantly more adaptable than other algorithms.

[0155] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A farmland irrigation method based on a multilayer perceptron model of the WOA-LSO algorithm, characterized in that: include: Obtain multiple sample meteorological data factors; Based on multiple sample meteorological data factors, multiple sample meteorological data factor combinations are determined, and correlation tests are conducted on each sample meteorological data factor combination and soil transpiration to determine the target sample meteorological data factor combination. Based on the target sample meteorological data factor combination, the number of nodes in the input layer, output layer, and hidden layer of the multilayer perceptron model is determined; Based on the multilayer perceptron model with the determined number of nodes, a multilayer perceptron prediction model optimized based on the WOA-LSO algorithm is constructed; the multilayer perceptron model optimized based on the WOA-LSO algorithm is obtained by optimizing the weight sequence and threshold sequence of the multilayer perceptron model using the WOA-LSO algorithm, obtaining the optimal weight sequence and the optimal threshold sequence, and assigning them to the multilayer perceptron model; Multiple target meteorological data factors of the study area are input into the multi-layer perceptron prediction model to obtain the soil evaporation in the future period; The effective soil moisture content and soil transpiration of the study area were input into the fuzzy inference system to obtain the irrigation coefficient, and the required irrigation amount of the farmland was calculated based on the irrigation coefficient.

2. The farmland irrigation method based on the multilayer perceptron model of the WOA-LSO algorithm according to claim 1, characterized in that: The multiple sample meteorological data factors include: maximum temperature, average temperature, minimum temperature, average relative humidity, day number, sunshine hours and wind speed; the multiple sample meteorological data factor combinations include a first group of meteorological data factor combinations and a second group of meteorological data factor combinations; determining the multiple sample meteorological data factor combinations based on the multiple sample meteorological data factors specifically includes: Determine a first group of meteorological data factor combinations based on multiple sample meteorological data factors; the first group of meteorological data factor combinations includes all meteorological data factor combinations, meteorological data factor combinations excluding the lowest temperature item, and meteorological data factor combinations excluding the highest temperature and the lowest temperature; Among the five meteorological data factors in the meteorological data factor combination excluding the maximum temperature and the minimum temperature, four are selected for random arrangement and combination, and all their combinations are determined as the second group of meteorological data factor combinations.

3. The farmland irrigation method based on the multilayer perceptron model of the WOA-LSO algorithm according to claim 2, characterized in that: The correlation test between each sample meteorological data factor combination and soil transpiration is performed to determine the target sample meteorological data factor combination, including: Each combination of meteorological data factors is input into the initial multi-layer perceptron model to obtain the predicted soil transpiration under each combination of meteorological data factors; According to the predicted soil evaporation and actual soil evaporation under each combination of meteorological data factors, the correlation between each combination of meteorological data factors and soil evaporation is determined; Obtain the candidate meteorological data factor combination with the greatest correlation in the second group of meteorological data factor combinations, and eliminate each meteorological data factor in the candidate meteorological data factor combination in turn to obtain a third group of meteorological data factor combinations; Obtain the correlation between each meteorological data factor combination in the third group of meteorological data factor combinations and soil evaporation; The meteorological data factor combination with the greatest correlation among all meteorological data factor combinations is determined as the target sample meteorological data factor combination.

4. The farmland irrigation method based on the multilayer perceptron model of the WOA-LSO algorithm according to claim 1, characterized in that: The method of determining the number of nodes in the input layer, output layer, and hidden layer of the multilayer perceptron model according to the target sample meteorological data factor combination includes: The number of sample meteorological data factors in the target sample meteorological data factor combination is determined as the number of nodes in the input layer; Determine the number of nodes in the output layer to be 1; The number of nodes in the hidden layer is determined according to the number of nodes in the input layer and the number of nodes in the output layer.

5. The farmland irrigation method based on the multilayer perceptron model of the WOA-LSO algorithm according to claim 4, characterized in that: Determining the number of nodes in the hidden layer according to the number of nodes in the input layer and the number of nodes in the output layer specifically includes: According to the relationship between the number of hidden layer nodes, the number of input layer nodes, and the number of output layer nodes, the range of the number of hidden layer nodes is determined by the formula: ； Where, is the minimum number of hidden layer nodes, is the maximum number of hidden layer nodes, is the number of input layer nodes, is the number of output layer nodes, is the number of hidden layer nodes; In the range of the number of hidden layer nodes The prediction accuracy of the multilayer perceptron prediction model under each hidden layer node number is obtained in turn, and the number of nodes corresponding to the highest prediction accuracy is determined as the number of hidden layer nodes.

6. The farmland irrigation method based on the multilayer perceptron model of the WOA-LSO algorithm according to claim 1, characterized in that: The WOA-LSO algorithm is used to optimize the weight sequence and threshold sequence of the multilayer perceptron model, specifically including: Initialize the population individual positions, population size, and maximum number of iterations of the WOA-LSO algorithm; the population individual positions are encoded sequences obtained by encoding the weight sequence and threshold sequence of the multilayer perceptron model; The mean square error of soil transpiration prediction using the multilayer perceptron model is selected as the fitness function value of the WOA-LSO algorithm. The positions of individual populations are iteratively updated, and the fitness function value of the WOA-LSO algorithm is iteratively calculated. When the WOA-LSO algorithm reaches the maximum number of iterations or the fitness function value is greater than the preset threshold, the iteration stops and the optimal population individual position is obtained. The optimal population individual position is the optimal coding sequence, and the optimal coding sequence is decoded to obtain the optimal weight sequence and the optimal threshold sequence.

7. The farmland irrigation method based on the multilayer perceptron model of the WOA-LSO algorithm according to claim 6, characterized in that: The coding sequence is: ； Where, is the coding sequence, is the weight sequence of the multilayer perceptron model, is the threshold sequence of the multilayer perceptron model, n and m are the number of weights and thresholds of the multilayer perceptron model, respectively.

8. The farmland irrigation method based on the multilayer perceptron model of the WOA-LSO algorithm according to claim 1, characterized in that: The fuzzy inference system includes: an input layer, a fuzzification layer, a rule layer, an aggregation layer, and a defuzzification layer; the soil effective water content and soil transpiration of the study area are input into the fuzzy inference system to obtain the irrigation coefficient, which specifically includes: The effective soil water content and soil transpiration are input into the fuzzification layer through the input layer; In the fuzzification layer, the Gaussian membership function is used to fuzzify the soil available water content and soil transpiration to obtain the initial fuzzy set. In the rule layer, fuzzy sets are reasoned according to predefined fuzzy rules to obtain multiple fuzzy outputs; In the aggregation layer, all fuzzy outputs are aggregated into a target fuzzy set; In the defuzzification layer, the centroid method is used to clarify the fuzzy output in the target fuzzy set to obtain the watering coefficient.

9. The farmland irrigation method based on the multilayer perceptron model of the WOA-LSO algorithm according to claim 1, characterized in that: The calculation of the amount of irrigation required for farmland based on the irrigation coefficient includes: According to the irrigation coefficient and the irrigation quota formula, the amount of irrigation water required for farmland is calculated as follows: ； Where, The amount of irrigation water required for farmland, is the irrigation coefficient, is the soil evaporation, is the irrigation water utilization coefficient.

Images