Method and system for estimating agricultural irrigation water consumption in irrigated area

Abstract

The invention relates to the technical field of agricultural irrigation management, in particular to a method and system for estimating the agricultural irrigation water consumption of an irrigation area. The method comprises the following steps: collecting soil, weather and crop data through a sensor and a detector, processing the collected data through an edge computing node, and then selecting an optimal transmission path to transmit the data through an Internet of Things gateway based on an https protocol; an irrigation area agricultural irrigation water consumption prediction model is constructed based on an intelligent algorithm of deep learning, the model stores, preprocesses and analyzes received data, then modeling and prediction are carried out on soil humidity, meteorological conditions and crop growth conditions, and finally an irrigation water consumption estimation result is obtained; and updating the irrigation area agricultural irrigation water consumption prediction model to ensure the accuracy and timeliness of model prediction. According to the method, through an advanced sensor technology, the Internet of Things, big data analysis and a machine learning algorithm, accurate estimation and optimal management of the agricultural irrigation water consumption of the irrigation area can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural irrigation management, and in particular to a method and system for estimating the amount of water used for agricultural irrigation in an irrigation area. Background Art

[0002] Traditional agricultural irrigation management methods mainly rely on historical experience and manual measurement, lacking accurate water monitoring and dynamic control methods, resulting in low estimation accuracy of irrigation water consumption and low water resource utilization efficiency. In addition, there are obvious spatial and temporal differences in factors such as soil moisture, climate conditions, and crop water requirements in the irrigation area. The traditional average water consumption estimation method is difficult to adapt to the diverse environmental needs of the irrigation area and lacks the mining of the correlation between multiple types of data. At the same time, with the expansion of agricultural irrigation and the increasing severity of water shortage problems, the demand for efficient use of water resources in irrigation areas has become more urgent. Traditional irrigation methods are unable to meet the requirements of efficient and sustainable development of modern agriculture in terms of water-saving effects and accuracy.

[0003] Therefore, the present invention provides an irrigation area agricultural irrigation water consumption estimation method and system to solve the above problems. Summary of the invention

[0004] In view of the shortcomings of the existing technology, the present invention develops a method and system for estimating the water consumption of agricultural irrigation in irrigation areas. The main purpose is to achieve accurate estimation and optimized management of the water consumption of agricultural irrigation in irrigation areas through advanced sensor technology, the Internet of Things, big data analysis and machine learning algorithms.

[0005] The technical solution to the technical problem of the present invention is, on the one hand, a method for estimating the amount of water used for agricultural irrigation in an irrigation area, comprising the following steps: S1. Collect soil, weather, and crop data through sensors and detectors, package and encapsulate the collected data into a standardized format through edge computing nodes, and then select the best transmission path to transmit the data through the IoT gateway based on the https protocol; S2. Construct an irrigation area agricultural irrigation water consumption prediction model based on a deep learning intelligent algorithm. The data processing center of the irrigation area agricultural irrigation water consumption prediction model processes the data in step S1, and stores, pre-processes and analyzes the received data, and then models and predicts soil moisture, meteorological conditions and crop growth conditions, and finally obtains the irrigation water consumption estimation result; S3. The irrigation area agricultural irrigation water consumption prediction model updates the internal parameters of the model with new input data at regular intervals. At the same time, the updated irrigation area agricultural irrigation water consumption prediction model is deployed to the edge computing node, and this process is iterated repeatedly.

[0006] In a specific implementation manner, S1 is as follows: The collected soil, meteorological and crop data include soil moisture, temperature, air humidity, precipitation, and crop growth data. The data collected at each moment is expressed as , , Respectively Soil moisture, temperature, air humidity, precipitation, and crop growth data collected at all times; The data in the data set is packaged through the edge computing node and encapsulated into a standardized format. The encapsulated data includes data timestamp, sensor ID, data type and value of the data in the data set.

[0007] In a specific implementation manner, S2 is as follows: S2.1. Save the received data and then perform preprocessing. The preprocessing operation includes data cleaning and data smoothing. Data cleaning is performed by removing outliers and noise data in the input data, and the data is smoothed by the sliding average method. The specific calculation is as follows: , in, represents the sliding window size, An index representing the size of the sliding window, express The data collected at any time, Represents the preprocessed data.

[0008] S2.2. Estimation of evapotranspiration: Estimating evapotranspiration from crops , the calculation process is as follows: , in, express The evapotranspiration at the time, express The rate of change of saturated water vapor pressure with temperature at a given moment is express The soil heat flux density measured at all times, represents the psychrometer constant, express The temperature collected at any time, express The wind speed measured at the time, express The saturated water vapor pressure measured at the time, express The actual water vapor pressure measured at any moment, express The net radiation at that moment.

[0009] S2.3. Extracting spatiotemporal features: The preprocessed data is processed by the CNN-mLSTM deep learning neural network and evapotranspiration Processing is performed to extract spatiotemporal features. The specific calculation is as follows: , in, express The spatial and temporal characteristics of the moment, Represents the operation of the CNN-mLSTM model; In CNN network: The preprocessed data and evapotranspiration After being input into the CNN-mLSTM model, the spatial features are first extracted through the CNN network, where the CNN network uses the ResNet pre-trained model. In mLSTM long short-term memory networks with multiple sequence lengths: (1) Based on LSTM, a matrix-based memory unit and covariance update rule are introduced. The memory unit of mLSTM is a matrix, denoted as , , Represents the dimension by memorizing the cell matrix Multiplication performs storage and retrieval of information; The storage mechanism in mLSTM is to store , storage key Sum , at time , through the query vector Retrieve Value , represents the increasing time step; (2) The calculation process within mLSTM is as follows: (2-1) Introducing dilated convolution, the unit state update of the fused dilated convolution operation is expressed as follows: , in, express The matrix memory cell state at the moment, express The door of forgetfulness of time, express The input gate of the moment, and Respectively The value input vector and key input vector at each moment, represents the convolution operation, represents the convolution kernel, express The elements, represents the convolution kernel size, The index representing the convolution kernel size, Represents the expansion rate.

[0010] (2-2) Normalizer state update: , in, express The normalizer state at time , express The normalizer state at time; (2-3) Hidden status update: , , in, express The hidden state of the moment, express The hidden state normalized at each moment, express The output gate at time, represents dot product; (2-4) Query input: , in, express The query vector at time, and Indicates the adjustment parameters of the query input; (2-5) key input: , in, express Key input at the time, , and Indicates three different adjustment parameters of key input; (2-6) Value input: , in, express The value input at the time, and Two different tuning parameters representing the query input; (2-7) Gating mechanism: The output of the input gate at any time : ; The output of the always-forget gate : ; Output of the time output gate : ; Among them, the output gate outputs That is The spatial temporal features at each moment are combined with the spatial temporal features extracted by the CNN network Splice to get spatiotemporal features , represents the Sigmoid activation function, represents transpose, represents the bias parameter of the input gate network, represents the neuron weight parameters of the input gate network, represents the bias parameter of the forget gate network, represents the neuron weight parameter of the forget gate network, represents the bias parameter of the output gate network, Represents the neuron weight parameters of the output gate network.

[0011] S2.4. Estimation of irrigation water consumption: The spatiotemporal characteristics Input to the residual-like spatiotemporal attention module, which consists of two branches; In the first branch, First, two parallel pooling layers are used. and , then add the output results and multiply them by the weight parameter , and then input the result into the activation function layer. After obtaining the calculated result, multiply it by the weight parameter and weight parameters , then input the two calculation results into two activation function layers set in parallel, and then multiply the results, and then Perform point multiplication to obtain the calculation result of the first branch; In the second branch, Through the pooling layer , multiplied by the weight parameter and weight parameters , then through the activation function layer, and then with Perform point multiplication to obtain the calculation result of the second branch; Finally, the results of the two branches are added together, and then passed through the batch normalization layer and the activation function layer to finally get the estimated irrigation water consumption. The specific calculation is as follows: , in, Indicates estimated The irrigation water consumption at the time, Represents the operation of the activation function layer, using the Sigmoid activation function, and BN represents the batch normalization operation. represents dot product, represents the pooling operation of the temporal feature branch in the first branch, represents the pooling operation of the spatial feature branch in the first branch, represents the pooling operation in the second branch, represents matrix addition, represents matrix multiplication, , , , and Represents five different weight parameters.

[0012] The present invention also provides an irrigation area agricultural irrigation water consumption estimation system, which implements an irrigation area agricultural irrigation water consumption estimation method, which includes the following parts: Soil moisture sensor module, meteorological monitoring module, crop growth monitoring module, Internet of Things gateway, data transmission module, edge computing node, data processing center and user interface. The above eight parts perform data observation or node management separately.

[0013] The effects provided in the content of the invention are only the effects of the embodiments, rather than all the effects of the invention. The above technical solution has the following advantages or beneficial effects: The present invention proposes an efficient and accurate method and system for estimating the amount of water used for agricultural irrigation in irrigation areas. The irrigation area agricultural irrigation water consumption estimation system provided by the present invention can obtain multi-source data such as weather, soil moisture, crop growth status, etc. in the irrigation area in real time by integrating advanced sensor technology, the Internet of Things and big data analysis technology, and input these data into a CNN-MLSTM model fused with dilated convolution for feature extraction, and construct a residual-type spatiotemporal attention module to analyze the features to realize the water consumption prediction of the irrigation area, fully considering the time correlation and spatial sparsity of the irrigation area data, and realizing the accurate estimation of the irrigation area agricultural irrigation water consumption. Compared with traditional methods, the system can not only utilize the spatial distribution characteristics of irrigation data, dynamically adjust irrigation strategies, and improve the efficiency of water resource utilization, but also deeply mine historical data through a time series deep learning algorithm, optimize the irrigation model, and make irrigation management more scientific and intelligent. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0015] Figure 1 It is a schematic diagram of the process of the present invention.

[0016] Figure 2 Line graph comparing different prediction methods for estimating agricultural water use in irrigated areas. DETAILED DESCRIPTION

[0017] In order to clearly illustrate the technical features of the present invention, the present invention is described in detail below through specific implementation methods and in conjunction with the accompanying drawings.

[0018] Example 1 like Figure 1 As shown in Figure 1, a method for estimating agricultural irrigation water consumption in an irrigation area is as follows: S1. Collect soil, weather, and crop data through sensors and detectors, package and encapsulate the collected data into a standardized format through edge computing nodes, and then select the best transmission path to transmit the data through the IoT gateway based on the https protocol; S2. Construct an irrigation area agricultural irrigation water consumption prediction model based on a deep learning intelligent algorithm. The data processing center of the irrigation area agricultural irrigation water consumption prediction model processes the data in step S1, and stores, pre-processes and analyzes the received data, and then models and predicts soil moisture, meteorological conditions and crop growth conditions, and finally obtains the irrigation water consumption estimation result; S3. The irrigation area agricultural irrigation water consumption prediction model updates the internal parameters of the model with new input data at regular intervals. At the same time, the updated irrigation area agricultural irrigation water consumption prediction model is deployed to the edge computing node, and this process is iterated repeatedly.

[0019] In a specific implementation manner, S1 is as follows: The collected soil, meteorological and crop data include soil moisture, temperature, air humidity, precipitation, and crop growth data. The data collected at each moment is expressed as , , Respectively Soil moisture, temperature, air humidity, precipitation, and crop growth data collected at all times; The data in the data set is packaged through the edge computing node and encapsulated into a standardized format. The encapsulated data includes data timestamp, sensor ID, data type and value of the data in the data set.

[0020] In a specific implementation manner, S2 is as follows: S2.1. Save the received data and then perform preprocessing. The preprocessing operation includes data cleaning and data smoothing. Data cleaning is performed by removing outliers and noise data in the input data, and the data is smoothed by the sliding average method. The specific calculation is as follows: , in, represents the sliding window size, An index representing the size of the sliding window, express The data collected at any time, Represents the preprocessed data.

[0021] S2.2. Estimation of evapotranspiration: Estimating evapotranspiration from crops , the calculation process is as follows: , in, express The evapotranspiration at the time, express The rate of change of saturated water vapor pressure with temperature at a given moment is express The soil heat flux density measured at all times, represents the psychrometer constant, express The temperature collected at any time, express The wind speed measured at the time, express The saturated water vapor pressure measured at the time, express The actual water vapor pressure measured at any moment, express The net radiation at that moment.

[0022] S2.3. Extracting spatiotemporal features: The preprocessed data is processed by the CNN-mLSTM deep learning neural network and evapotranspiration Processing is performed to extract spatiotemporal features. The specific calculation is as follows: , in, express The spatial and temporal characteristics of the moment, Represents the operation of the CNN-mLSTM model; In CNN network: The preprocessed data and evapotranspiration After being input into the CNN-mLSTM model, the spatial features are first extracted through the CNN network, where the CNN network uses the ResNet pre-trained model. In mLSTM long short-term memory networks with multiple sequence lengths: (1) Based on LSTM, a matrix-based memory unit and covariance update rule are introduced. The memory unit of mLSTM is a matrix, denoted as , , Represents the dimension by memorizing the cell matrix Multiplication performs storage and retrieval of information; The storage mechanism in mLSTM is to store , storage key Sum , at time , through the query vector Retrieve Value , represents the increasing time step; (2) The calculation process within mLSTM is as follows: (2-1) Introducing dilated convolution, the unit state update of the fused dilated convolution operation is expressed as follows: , in, express The matrix memory cell state at the moment, express The door of forgetfulness of time, express The input gate of the moment, and Respectively The value input vector and key input vector at each moment, represents the convolution operation, represents the convolution kernel, express The elements, represents the convolution kernel size, The index representing the convolution kernel size, Represents the expansion rate.

[0023] (2-2) Normalizer state update: , in, express The normalizer state at time , express The normalizer state at time; (2-3) Hidden status update: , , in, express The hidden state of the moment, express The hidden state normalized at each moment, express The output gate at time, represents dot product; (2-4) Query input: , in, express The query vector at time, and Indicates the adjustment parameters of the query input; (2-5) key input: , in, express Key input at the time, , and Indicates three different adjustment parameters of key input; (2-6) Value input: , in, express The value input at the time, and Two different tuning parameters representing the query input; (2-7) Gating mechanism: The output of the input gate at any time : ; The output of the always-forget gate : ; Output of the time output gate : ; Among them, the output gate outputs That is The spatial temporal features at each moment are combined with the spatial temporal features extracted by the CNN network Splice to get spatiotemporal features , represents the Sigmoid activation function, represents transpose, represents the bias parameter of the input gate network, represents the neuron weight parameters of the input gate network, represents the bias parameter of the forget gate network, represents the neuron weight parameter of the forget gate network, represents the bias parameter of the output gate network, Represents the neuron weight parameters of the output gate network.

[0024] S2.4. Estimation of irrigation water consumption: The spatiotemporal characteristics Input to the residual-like spatiotemporal attention module, which consists of two branches; In the first branch, First, two parallel pooling layers are used. and , then add the output results and multiply them by the weight parameter , and then input the result into the activation function layer. After obtaining the calculated result, multiply it by the weight parameter and weight parameters , then input the two calculation results into two activation function layers set in parallel, and then multiply the results, and then Perform point multiplication to obtain the calculation result of the first branch; In the second branch, Through the pooling layer , multiplied by the weight parameter and weight parameters , then through the activation function layer, and then with Perform point multiplication to obtain the calculation result of the second branch; Finally, the results of the two branches are added together, and then passed through the batch normalization layer and the activation function layer to finally get the estimated irrigation water consumption. The specific calculation is as follows: , in, Indicates estimated The irrigation water consumption at the time, Represents the operation of the activation function layer, using the Sigmoid activation function, and BN represents the batch normalization operation. represents dot product, represents the pooling operation of the temporal feature branch in the first branch, represents the pooling operation of the spatial feature branch in the first branch, represents the pooling operation in the second branch, represents matrix addition, represents matrix multiplication, , , , and Represents five different weight parameters.

[0025] Example 2 A system for estimating the amount of water used for agricultural irrigation in an irrigation area, which implements a method for estimating the amount of water used for agricultural irrigation in an irrigation area, comprises the following parts: Soil moisture sensor module, meteorological monitoring module, crop growth monitoring module, Internet of Things gateway, data transmission module, edge computing node, data processing center and user interface. The above eight parts can perform data observation or node management separately.

[0026] Example 3 As shown in Table 1, in order to further verify the beneficial effects of the present invention, the present invention is compared with the existing method in terms of prediction accuracy based on the data set collected by the present invention. The calculation results of the root mean square error, the mean absolute error and the estimation accuracy show that the error of the present invention is small, and therefore, the prediction accuracy of the present invention is higher.

[0027] Figure 2 A comparative line graph of the method of the present invention and the existing convolutional neural network model and Lstm model for estimating the amount of water used for irrigation in irrigation areas is given in Figure 2 It can be seen that the CNN-mLSTM model (red line) proposed in the present invention overlaps more with the actual observed value (black line), which further illustrates that the prediction accuracy of the method of the present invention is higher.

[0028] Table 1 Comparison results of this method with other methods Although the above describes the specific implementation mode of the invention in conjunction with the drawings, it is not intended to limit the scope of protection of the invention. Based on the technical solution of the present invention, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present invention.

Claims

1. A method for estimating water consumption for agricultural irrigation in an irrigated area, characterized in that: The following steps are involved: S1. Collect soil, weather, and crop data through sensors and detectors, package and encapsulate the collected data into a standardized format through edge computing nodes, and then select the best transmission path to transmit the data through the IoT gateway based on the https protocol; S2. Construct an irrigation area agricultural irrigation water consumption prediction model based on a deep learning intelligent algorithm. The data processing center of the irrigation area agricultural irrigation water consumption prediction model processes the data in step S1, and stores, pre-processes and analyzes the received data, and then models and predicts soil moisture, meteorological conditions and crop growth conditions, and finally obtains the irrigation water consumption estimation result; S3. The irrigation area agricultural irrigation water consumption prediction model updates the internal parameters of the model with new input data at regular intervals. At the same time, the updated irrigation area agricultural irrigation water consumption prediction model is deployed to the edge computing node, and this process is iterated repeatedly.

2. A method for estimating water consumption for agricultural irrigation in an irrigation area according to claim 1, characterized in that: S1 is as follows: The collected soil, meteorological and crop data include soil moisture, temperature, air humidity, precipitation, and crop growth data. The data collected at each moment is expressed as , , Respectively Soil moisture, temperature, air humidity, precipitation, and crop growth data collected at all times; The data in the data set is packaged through the edge computing node and encapsulated into a standardized format. The encapsulated data includes data timestamp, sensor ID, data type and value of the data in the data set.

3. A method for estimating water consumption for agricultural irrigation in an irrigation area according to claim 2, characterized in that: S2 is as follows: S2.

1. Save the received data and then perform preprocessing. The preprocessing operation includes data cleaning and data smoothing. Data cleaning is performed by removing outliers and noise data in the input data, and the data is smoothed by the sliding average method. The specific calculation is as follows: ， in, represents the sliding window size, An index representing the size of the sliding window, express The data collected at any time, Represents the preprocessed data.

4. A method for estimating water consumption for agricultural irrigation in an irrigation area according to claim 3, characterized in that: S2.

2. Estimation of evapotranspiration: Estimating evapotranspiration from crops , the calculation process is as follows: ， in, express The evapotranspiration at the time, express The rate of change of saturated water vapor pressure with temperature at a given moment is express The soil heat flux density measured at all times, represents the psychrometer constant, express The temperature collected at any time, express The wind speed measured at the time, express The saturated water vapor pressure measured at the time, express The actual water vapor pressure measured at any moment, express The net radiation at that moment.

5. A method for estimating water consumption for agricultural irrigation in an irrigation area according to claim 4, characterized in that S2 The details are as follows: S2.

3. Extracting spatiotemporal features: The preprocessed data is processed by the CNN-mLSTM deep learning neural network and evapotranspiration Processing is performed to extract spatiotemporal features. The specific calculation is as follows: ， in, express The spatial and temporal characteristics of the moment, Represents the operation of the CNN-mLSTM model; In CNN network: The preprocessed data and evapotranspiration After being input into the CNN-mLSTM model, the spatial features are first extracted through the CNN network, where the CNN network uses the ResNet pre-trained model. In mLSTM long short-term memory networks with multiple sequence lengths: (1) Based on LSTM, a matrix-based memory unit and covariance update rule are introduced. The memory unit of mLSTM is a matrix, denoted as , , Represents the dimension by memorizing the cell matrix Multiplication performs storage and retrieval of information; The storage mechanism in mLSTM is to store , storage key Sum , at time , through the query vector Retrieve Value , represents the increasing time step; (2) The calculation process within mLSTM is as follows: (2-1) Introducing dilated convolution, the unit state update of the fused dilated convolution operation is expressed as follows: ， in, express The matrix memory cell state at the moment, express The door of forgetfulness of time, express The input gate of the moment, and Respectively The value input vector and key input vector at each moment, represents the convolution operation, represents the convolution kernel, express The elements, represents the convolution kernel size, The index representing the convolution kernel size, represents the expansion rate; (2-2) Normalizer state update: ， in, express The normalizer state at time , express The normalizer state at time; (2-3) Hidden status update: ， ， in, express The hidden state of the moment, express The hidden state normalized at each moment, express The output gate at time, represents dot product; (2-4) Query input: ， in, express The query vector at time, and Represents the adjustment parameters of the query input; (2-5) key input: ， in, express Key input at the time, , and Indicates three different adjustment parameters of key input; (2-6) Value input: ， in, express The value input at the time, and Two different tuning parameters representing the query input; (2-7) Gating mechanism: The output of the input gate at any time : ； The output of the forget gate : ； Output of the time output gate : ； Among them, the output gate outputs That is The spatial temporal features at each moment are combined with the spatial temporal features extracted by the CNN network Splice to get spatiotemporal features , represents the Sigmoid activation function, represents transpose, represents the bias parameter of the input gate network, represents the neuron weight parameters of the input gate network, represents the bias parameter of the forget gate network, represents the neuron weight parameter of the forget gate network, represents the bias parameter of the output gate network, Represents the neuron weight parameters of the output gate network.

6. A method for estimating water consumption for agricultural irrigation in an irrigation area according to claim 5, characterized in that S2 The details are as follows: S2.

4. Estimation of irrigation water consumption: The spatiotemporal characteristics Input to the residual-like spatiotemporal attention module, which consists of two branches; In the first branch, First, two parallel pooling layers are used. and , then add the output results and multiply them by the weight parameter , and then input the result into the activation function layer. After obtaining the calculated result, multiply it by the weight parameter and weight parameters , then input the two calculation results into two activation function layers set in parallel, and then multiply the results, and then Perform point multiplication to obtain the calculation result of the first branch; In the second branch, Through the pooling layer , multiplied by the weight parameter and weight parameters , then through the activation function layer, and then with Perform point multiplication to obtain the calculation result of the second branch; Finally, the results of the two branches are added together, and then passed through the batch normalization layer and the activation function layer to finally get the estimated irrigation water consumption. The specific calculation is as follows: ， in, Indicates estimated The irrigation water consumption at each time Represents the operation of the activation function layer, using the Sigmoid activation function, and BN represents the batch normalization operation. represents dot product, represents the pooling operation of the temporal feature branch in the first branch, represents the pooling operation of the spatial feature branch in the first branch, represents the pooling operation in the second branch, represents matrix addition, represents matrix multiplication, , , , and Represents five different weight parameters.

7. A system for estimating the amount of water used for agricultural irrigation in an irrigation area, which executes a method for estimating the amount of water used for agricultural irrigation in an irrigation area as claimed in any one of claims 1 to 6, characterized in that: Includes the following sections: Soil moisture sensor module, meteorological monitoring module, crop growth monitoring module, Internet of Things gateway, data transmission module, edge computing node, data processing center and user interface. The above eight parts perform data observation or node management separately.

Images