Full-automatic quantitative drip irrigation system

Abstract

The invention discloses a full-automatic quantitative drip irrigation system, and relates to the technical field of agricultural irrigation. The system comprises a humidity monitoring module which adopts a distributed sensor network to measure humidity data of a target planting area, introduces an adaptive calibration algorithm, and is used for correcting humidity data deviation, performing secondary mean value processing and generating a humidity value of the target planting area; a meteorological analysis module, a water demand prediction module, an intelligent control module and an effect evaluation module; the technical key points are as follows: continuous optimization of an irrigation strategy is realized by adopting a technical scheme of a reinforcement learning optimization strategy and an effect evaluation module; the reinforcement learning optimization strategy allows the system to continuously adjust the irrigation strategy according to the current state and historical data, the purpose is to maximize accumulated rewards, the closed-loop mechanism ensures long-term efficient operation of the irrigation system, the efficiency and sustainability of agricultural production are remarkably improved, and the defect that an irrigation system in the past cannot be automatically optimized and adjusted is overcome.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural irrigation, in particular to a full-automatic quantitative drip irrigation system. Background Art

[0002] Agricultural irrigation mainly refers to irrigation operations in agricultural cultivation areas. Agricultural irrigation methods generally include traditional surface irrigation, ordinary sprinkler irrigation, and micro-irrigation. In ancient times, agricultural irrigation relied on precipitation from the sky and rivers on the ground. Farming was mainly concentrated in areas with abundant precipitation and well-developed river networks. Farmers in these areas often engage in agricultural production according to the solar terms. For sugarcane planting areas, they attach great importance to water, so it is necessary to design a precise quantitative irrigation system.

[0003] Traditional irrigation technology has several significant problems:

[0004] First, fixed-time or fixed-amount irrigation strategies often lead to water waste or insufficient irrigation. For example, in the dry season, farmers may over-water to prevent crops from suffering from water shortages, but neglect to adjust during the rainy season, resulting in water waste. Second, the lack of real-time monitoring and analysis of soil moisture and meteorological data makes it difficult to accurately predict crop water requirements, which makes irrigation decisions unscientific and affects crop growth. For example, misestimation of rainfall may lead to unnecessary irrigation or insufficient irrigation, affecting crop yields. In addition, sensor aging or errors will affect the consistency and accuracy of data and interfere with irrigation management. For example, deviations in soil moisture sensor readings will lead to irrigation plan errors and affect the assessment of crop root water absorption capacity. Furthermore, traditional irrigation systems usually do not have self-optimization capabilities and cannot dynamically adjust irrigation strategies according to crop growth status and environmental changes. The above problems have not been effectively solved in the current drip irrigation system. Summary of the Invention

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0006] Fully automatic quantitative drip irrigation system, including:

[0007] The humidity monitoring module uses a distributed sensor network to measure the humidity data of the target planting area, introduces an adaptive calibration algorithm to correct the humidity data deviation, and performs quadratic mean processing to generate the humidity value of the target planting area;

[0008] The meteorological analysis module uses the acquired weather forecast information as a basis and adopts an improved support vector machine model to predict the meteorological data of the target planting area. The goal of the improved support vector machine model is to minimize the set loss function.

[0009] The water demand prediction module is based on humidity and meteorological data, and integrates the crop growth data of the target planting area. The data is input into a pre-built neural network model and outputs the predicted water demand of the target planting area;

[0010] The intelligent control module executes the predetermined drip irrigation operation based on the predicted water demand;

[0011] The system selectively decides whether to trigger a reinforcement learning-based optimization strategy based on the drip irrigation effect to adjust the operating status of the drip irrigation system in the target planting area, including the time to open / close the valve and the flow rate.

[0012] The effect evaluation module continuously tracks the drip irrigation effect after the optimization strategy is implemented, observes the changing trends of crop growth indicators and humidity values, and generates effect evaluation indicators based on the changes in crop growth indicators, humidity values, and irrigation volume. When the effect evaluation indicator is lower than the preset threshold, the optimization strategy based on reinforcement learning is triggered; otherwise, the established drip irrigation operation continues.

[0013] Furthermore, let H i (t) represents the humidity data measured by the i-th humidity sensor at time t, H t ′(t) is the humidity data after calibration, and the calibration process is expressed as:

[0014] H i ′(t)=H i (t)+α(H_avg(t)-H i (t));

[0015] Where H_avg(t) represents the average reading of all humidity sensors in the sensor network at time point t, that is, the average humidity data, and α represents the calibration coefficient, which ranges from 0 to 1.

[0016] Furthermore, temperature, rainfall and wind speed; let X be the first input eigenvector and y be the predicted target variable, then the improved support vector machine model aims to minimize the following loss function:

[0017]

[0018] The input first eigenvector X includes a number of historical weather data points, and the historical weather data points include at least: the average temperature of the past several days, the average relative humidity of the past several days, the accumulated precipitation of the past several days, and the average wind speed of the past several days;

[0019] Let X = [x1, x2, ..., x m ], where m represents the number of first eigenvectors;

[0020] In the above formula, w represents the weight vector, b represents the bias term, C represents the dynamic penalty factor, i represents the i-th training sample, n represents the total number of training samples, and y i The true label of the i-th sample; x i The feature vector of the i-th sample.

[0021] Furthermore, in the improved support vector machine model, the penalty factor in the target loss function is derived based on the adaptive learning rate and loss function feedback mechanism;

[0022] The operation process of the feedback mechanism is: Initialization: the penalty factor C of all samples i Set to the same initial value C0; Dynamic adjustment: In each iteration, adjust the corresponding penalty factor C according to the performance of each sample based on the improved support vector machine model i Feedback mechanism: This feedback mechanism is introduced to use historical data and the performance of the current improved support vector machine model to predict future adjustment directions.

[0023] Furthermore, the crop growth data includes at least: the root water absorption capacity value and the leaf transpiration rate of the crop; wherein the root water absorption capacity value is calculated by weighting the soil moisture and the maximum depth of the root distribution; and the leaf transpiration rate is measured by a plant lysimeter.

[0024] Furthermore, in the water demand forecasting module, data integration is performed to obtain the second eigenvector;

[0025] The second eigenvector is input into the neural network model to output the predicted water requirement of the target planting area;

[0026] The neural network model is expressed as:

[0027] Q = f(X2; W);

[0028] Where Q represents the output variable, i.e., the predicted water demand; X2 is the second eigenvector set, which includes at least humidity values, meteorological data, and crop growth data; and W represents the weight parameter in the neural network model. The model is adjusted continuously until the error between the neural network model output and the actual observation value is minimized, and the water demand prediction result can be obtained.

[0029] Furthermore, the optimization strategy based on reinforcement learning is:

[0030] Let R(a v , s v ) Take action a in the current state v The reward value obtained after

[0031] Among them, the reward value is used to evaluate the quality of any action; vAs the objective function, at the state of time step v, the goal of reinforcement learning is to find where γ is the discount factor, ranging from 0 to 1.

[0032] Furthermore, the change in crop growth indicators is expressed as follows: the change in the comprehensive value obtained by weighting plant height and leaf area index LAI;

[0033] The changes are: the value after drip irrigation minus the value before drip irrigation;

[0034] The formula for generating the performance evaluation index is:

[0035]

[0036] Where Er represents the effect evaluation index, δ1 and δ2 represent weight coefficients, both ranging from 0 to 1, ΔGr represents the change in crop growth index before and after drip irrigation, ΔH represents the change in humidity before and after drip irrigation, and Ir represents the actual irrigation amount.

[0037] The present invention provides a fully automatic quantitative drip irrigation system, which has the following beneficial effects:

[0038] (1) By adopting the adaptive calibration algorithm and data integration technology, the consistency and accuracy of sensor data are achieved, which solves the problem of data inaccuracy caused by sensor errors in traditional systems, improves the reliability of the system, enhances the stability of the system, and provides a solid foundation for subsequent water demand forecasting;

[0039] (2) By dynamically adjusting the penalty factor, the model can better focus on samples with large prediction errors, thereby improving the overall prediction accuracy. This solves the problem that when the penalty factor is fixed in the traditional way, some samples may be ignored, affecting the overall performance of the model. It introduces a feedback mechanism based on adaptive learning rate and loss function, and uses historical data to predict the future adjustment direction, thereby improving and ensuring the generalization ability. On the other hand, it can achieve a better fitting effect in a shorter time, thereby accelerating the convergence speed.

[0040] (3) By adopting the scheme design of precise irrigation management and intelligent control, the water demand of crops can be accurately predicted and dynamically adjusted, achieving the effect of efficient water resource utilization and solving the problem of water resource waste in traditional irrigation methods. Combining the data of humidity monitoring, meteorological analysis and water demand prediction modules, the irrigation strategy can be dynamically adjusted according to the actual environmental conditions. Compared with the traditional fixed time or fixed amount irrigation method, this linkage mechanism not only improves the efficiency of water resource utilization, but also reduces the resource waste caused by over-watering or under-irrigation.

[0041] (4) By adopting the technical solution of reinforcement learning optimization strategy and effect evaluation module, the continuous optimization of irrigation strategy is achieved, the effect of closed-loop management system is achieved, and the problem of traditional irrigation system lacking self-optimization ability is solved; the reinforcement learning optimization strategy allows the system to continuously adjust the irrigation strategy according to the current status and historical data, with the goal of maximizing the cumulative reward, and at the same time track the drip irrigation effect after execution, and provide timely feedback and adjust the strategy. This closed-loop mechanism ensures the long-term and efficient operation of the irrigation system, significantly improves the efficiency and sustainability of agricultural production, and makes up for the defect that the irrigation system in the past could not be automatically optimized and adjusted. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The figure is a flow chart of the operation process of the fully automatic quantitative drip irrigation system in the present invention. DETAILED DESCRIPTION

[0043] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0044] Example 1:

[0045] See also Figure 1 This embodiment provides a fully automatic quantitative drip irrigation system. This system, a product of agricultural technological advancement, enables precision irrigation, a key means of increasing crop yields and improving water resource utilization efficiency. Water management is particularly important for sugarcane cultivation, as it directly affects the growth rate, sugar accumulation, and ultimate yield of the sugarcane. This system utilizes a series of intelligent functional modules to achieve precise, efficient, and continuous irrigation management of sugarcane fields. An overview of the system is as follows:

[0046] Humidity monitoring module:

[0047] A distributed sensor network is used to measure the humidity data of the target planting area, and an adaptive calibration algorithm is introduced to correct the humidity data deviation. The quadratic mean processing is then performed to generate the humidity value of the target planting area.

[0048] The multi-point distributed sensor network is deployed at selected test points within the target planting area. The feedback results from the test points represent the results of the entire target planting area. The sensor network includes several humidity sensors evenly distributed at the test points to measure the humidity of the crop soil at the test points. The data obtained is the humidity data.

[0049] Specifically, each humidity sensor provides highly accurate humidity readings and transmits these data wirelessly to a central control center for aggregation. To ensure data consistency and accuracy, an adaptive calibration algorithm is implemented. This algorithm periodically checks each humidity sensor's reading and compares it with surrounding sensors to correct for any discrepancies (e.g., due to sensor aging).

[0050] Let H i (t) represents the humidity data measured by the i-th humidity sensor at time t, H t ′(t) is the humidity data after calibration, and the calibration process is expressed as:

[0051] H i ′(t)=H i (t)+α(H_avg(t)-H i (t));

[0052] Where H_avg(t) represents the average reading of all humidity sensors in the sensor network at time point t, that is, the average humidity data; α represents the calibration coefficient, which ranges from 0 to 1 and is used to adjust the calibration strength; a secondary mean calculation is performed on the calibrated humidity data to obtain the humidity value of the target planting area; a representative time point is usually selected to represent the humidity of the target planting area; the corresponding humidity values at each time point t can also be further averaged within a period T as needed to obtain the final required humidity value; this method can ensure that the humidity data of the entire system is highly consistent, thereby providing a reliable basis for subsequent water demand forecasting.

[0053] Effect description: By adopting the technical solution of adaptive calibration algorithm and data integration, the consistency and accuracy of sensor data are achieved, the reliability of the system is improved, and the problem of inaccurate data caused by sensor errors in traditional systems is solved; among them, the improved adaptive calibration algorithm regularly corrects the humidity sensor readings and combines meteorological data and crop growth data for secondary mean processing, ensuring the high quality of the input data of the entire system. This provides a solid foundation for subsequent water demand forecasting, enhances the stability and reliability of the system, and overcomes the measurement deviation problem caused by sensor aging or other factors in the past.

[0054] Meteorological analysis module:

[0055] Based on the acquired weather forecast information, an improved support vector machine model is used to predict the short-term weather trend of the target planting area, i.e., meteorological data. The goal of the improved support vector machine model is to minimize the set loss function.

[0056] The weather forecast information comes from: local weather stations or public internet resources;

[0057] Weather forecast information includes key parameters such as temperature, rainfall, and wind speed;

[0058] Let X be the first eigenvector of the input and y be the predicted target variable. The goal of the improved support vector machine model is to minimize the following loss function:

[0059]

[0060] The input first eigenvector X contains several historical weather data points, and these historical weather data points include but are not limited to: temperature: the average temperature of the past few days; humidity: the average relative humidity of the past few days; precipitation: the cumulative precipitation of the past few days; wind speed: the average wind speed of the past few days;

[0061] Let X = [x1, x2, ..., x m ], where m represents the number of first eigenvectors;

[0062] In the above formula, w represents the weight vector, which is used to describe the direction of the hyperplane; b represents the bias term, which is used to adjust the position of the hyperplane; C represents the dynamic penalty factor, which is used to balance the relationship between maximizing the classification interval and the classification error; i represents the i-th training sample (the i here only represents this meaning in the above formula), n represents the total number of training samples; y i The true label of the i-th sample; x i The feature vector of the i-th sample; w*x i represents the dot product operation, which represents the inner product of the weight vector and the first eigenvector;

[0063] Specifically, the first part of the above formula (corresponding to the part before C) is the regularization term, which is used to prevent overfitting; the second part (except the first part) is the loss function, which is used to measure how well the model fits the training data;

[0064] Among them, the penalty factor is derived based on the adaptive learning rate and loss function feedback mechanism;

[0065] The operation process of this mechanism is as follows: Initialization: At the beginning, the penalty factor C of all samples i Set to the same initial value C0; Dynamic adjustment: In each iteration, adjust the corresponding penalty factor C according to the performance of each sample based on the improved support vector machine model i Feedback mechanism: This feedback mechanism is introduced to use historical data and the performance of the current improved support vector machine model to predict future adjustment directions to further improve the generalization ability and robustness of the model;

[0066] Specifically, regarding initialization: Let C i (0) = C0 represents the penalty factor of the i-th sample at the 0th iteration. The initial value C0 is a fixed constant. Regarding dynamic adjustment: in each iteration t, the prediction error e of the improved support vector machine model for the i-th sample is calculated. i (t) :

[0067] e i (t) =|y i -(w (t) ·x i +b (t) )|;

[0068] Then, the penalty factor C is adjusted according to the prediction error i :

[0069] C i (t+1) =C i (t) ·exp(λ·e i (t) );

[0070] Where λ represents the learning rate, which ranges from 0 to 1 and is used to control the adjustment amplitude; e i (t) represents the prediction error of the i-th sample in the t-th iteration;

[0071] Regarding the feedback mechanism: the following linear regression model is used to predict the direction of future adjustments:

[0072]

[0073] Where k represents the size of the historical data window; β represents the adjustment coefficient, which ranges from 0 to 1 and controls the degree of influence of historical change trends on current adjustments; the larger the value, the more sensitive the model is to historical trends; the smaller the value, the more conservative it is; C i (t-j) represents the penalty factor of the i-th sample at the tj-th iteration, which represents the penalty weight value of a certain sample at a certain moment in the past, and j is the number of steps counted from the current time point (e.g., j = 1 represents the previous iteration);

[0074] The final penalty factor update formula is:

[0075] C i (t+1) =C i (t) ·exp(λ·e i (t) )+ΔC i(t+1) ;

[0076] Logical explanation: Through the prediction error e i (t) To adjust the penalty factor C i , which makes the model pay more attention to samples with larger prediction errors, thereby improving the fitting accuracy of these samples. By introducing a feedback mechanism of historical data (i.e., based on an adaptive learning rate and loss function feedback mechanism), the future adjustment direction can be predicted more accurately, avoiding the problem of over-adjustment or under-adjustment, thereby improving the overall performance of the model.

[0077] Effect description: By dynamically adjusting the penalty factor, the model can better focus on samples with large prediction errors, thereby improving the overall prediction accuracy. This solves the problem that when the penalty factor is fixed in the traditional way, some samples may be ignored, affecting the overall performance of the model. Although the fixed penalty factor makes the model show better fitting ability on some data sets, it has poor generalization ability on other data sets. Therefore, a feedback mechanism based on adaptive learning rate and loss function is further introduced, and historical data is used to predict the future adjustment direction, thereby improving and ensuring the generalization ability. Unlike the original dynamic or fixed penalty factor, the model adjusted by the feedback mechanism is more stable when facing outliers or noisy data, reducing the risk of overfitting. On the other hand, it can achieve a better fitting effect in a shorter time, thereby accelerating the convergence speed.

[0078] In summary, by dynamically adjusting the penalty factor and combining it with a feedback mechanism, the overall solution not only improves the model's prediction accuracy and generalization ability, but also brings unexpected results in terms of stability and convergence speed. This gives the solution significant advantages in complex application scenarios such as meteorological data analysis and can better support the development of precision agriculture. For example, during sugarcane planting, the system can more accurately predict precipitation, thereby formulating more scientific and reasonable irrigation plans, reducing water waste and increasing crop yields.

[0079] Water demand forecasting module (based on the data of the above two modules to predict the water demand of crops, such as sugarcane):

[0080] Based on humidity values (which may also include humidity data) and meteorological data, and integrated with crop growth data of the target planting area, the data is input into a pre-built neural network model to fit the complex nonlinear relationship between crop water requirements and environmental conditions; the predicted water requirements of the target planting area are output;

[0081] Crop growth data includes the root water absorption capacity and leaf transpiration rate of crops at the test point (if there are multiple test points, the average is taken as the corresponding crop growth data). As mentioned in the humidity monitoring module, the test point can represent the entire target planting area. The root water absorption capacity can be calculated by weighting soil moisture and the maximum root distribution depth; the leaf transpiration rate can be directly measured using a plant lysimeter.

[0082] After data integration, the second eigenvector is obtained;

[0083] The second eigenvector is input into the neural network model to output the predicted water requirement of the target planting area;

[0084] The form of the neural network model can be expressed as:

[0085] Q = f(X2; W);

[0086] Where Q represents the output variable, i.e., the predicted water demand; X2 is the second eigenvector set (including humidity values, meteorological data, and crop growth data); and W represents the weight parameter in the neural network model. By continuously adjusting until the error between the neural network model output and the actual observation value is minimized, a more accurate water demand prediction result can be obtained.

[0087] Intelligent control module:

[0088] Based on the predicted water demand, the system executes the planned drip irrigation operation. Based on the drip irrigation results, it selectively decides whether to trigger a reinforcement learning-based optimization strategy to automatically adjust the operating status of the drip irrigation system within the target planting area, including the time to open / close the valve and the flow rate. While running this optimization strategy, the system continuously tries different irrigation plans in the solution library and adjusts the next optimization decision-making method based on the effect feedback to approach or reach the optimal solution.

[0089] The core of the above optimization strategy is to minimize water waste while meeting the needs of crop growth;

[0090] The optimization strategy based on reinforcement learning is:

[0091] Let R(a v , s v ) Take action a in the current state v The reward value obtained after the action is completed; the reward value is used to evaluate the quality of an action; for example, in a drip irrigation system, if the irrigation amount is appropriate and the soil moisture is kept in an appropriate range, the reward value is high; on the contrary, if the irrigation is too much or too little, resulting in inappropriate soil moisture, the reward value is low; v is the objective function, the state of the system at time step v; then the goal of reinforcement learning is to find The strategy of maximizing γ is a discount factor that is used to balance the importance of short-term and long-term rewards, and its value ranges from 0 to 1. v The exponential form of the discount factor is used to discount future rewards. During this process, the system will continuously try different irrigation plans and adjust the direction of the next decision based on the actual effect feedback, thus gradually approaching the optimal solution.

[0092] By adopting the above scheme, the system can continuously adjust the irrigation strategy according to the current status and historical data to maximize the cumulative reward, which not only improves the efficiency of water resource utilization, but also ensures that crops (for example, sugarcane) are always in the best growth state. The above dynamic adjustment of irrigation strategy adapts to different environmental conditions and significantly improves the efficiency and sustainability of agricultural production. Specifically, the technical solution of precise irrigation management and intelligent control module is adopted to realize the accurate prediction and dynamic adjustment of crop water demand, achieve the effect of efficient water resource utilization, and solve the problem of water resource waste in traditional irrigation methods. Combined with the data of humidity monitoring, meteorological analysis and water demand prediction module, the intelligent control module can dynamically adjust the irrigation strategy according to actual environmental conditions. Compared with the traditional fixed time or fixed amount irrigation method, this linkage mechanism not only improves the efficiency of water resource utilization, but also reduces the resource waste caused by over-watering or insufficient irrigation.

[0093] Effect evaluation module:

[0094] Continuously track the drip irrigation effect after implementing the optimization strategy, observe the changing trends of crop growth indicators and humidity values, and generate effect evaluation indicators based on the changes in crop growth indicators, humidity values, and irrigation volume;

[0095] When the effect evaluation index is lower than the preset threshold, it means that the current drip irrigation effect is poor, triggering the optimization strategy based on reinforcement learning; when the effect evaluation value is not lower than the preset threshold, it means that the current drip irrigation effect meets the standard, and the original drip irrigation operation can be executed;

[0096] The crop growth index is similar to the crop growth data mentioned above, but the content of the crop growth index is different. The change in the crop growth index is represented by the change in the comprehensive value obtained by weighting the plant height and leaf area index (LAI) (both obtained within the test point, similar to the description of the test point above);

[0097] The changes are: the value after drip irrigation minus the value before drip irrigation;

[0098] The formula for generating the performance evaluation index is:

[0099]

[0100] In the formula, Er represents the effect evaluation index, δ1 and δ2 represent weight coefficients, both ranging from 0 to 1, ΔGr represents the change in crop growth index before and after drip irrigation, ΔH represents the change in humidity before and after drip irrigation, and Ir represents the actual irrigation amount. Logic explanation: The numerator in the formula is the comprehensive result obtained by weighted calculation, which can comprehensively evaluate the effect of the drip irrigation strategy. The denominator represents the actual irrigation amount used. Using it as the denominator ensures that the evaluation results reflect the effect of each unit of irrigation amount, facilitating subsequent comparisons. It should be noted that the weight coefficient is determined using the coefficient of variation method, which assigns weights to each evaluation indicator based on the degree of variation between the current value and the target value of each evaluation indicator. If the numerical difference of an indicator is large, it can clearly distinguish the evaluated objects, indicating that the indicator has rich discriminative information and should be given a larger weight. Conversely, if the numerical difference of the evaluated objects on a certain indicator is small, then the ability of this indicator to distinguish the evaluated objects is weak and should be given a smaller weight. This method directly utilizes the information contained in each indicator to calculate the indicator weight, thus being objective.

[0101] Results: By adopting a reinforcement learning optimization strategy and an effect evaluation module, the system achieves continuous optimization of irrigation strategies, achieving the effect of a closed-loop management system and resolving the problem of traditional irrigation systems lacking self-optimization capabilities. The reinforcement learning optimization strategy allows the system to continuously adjust irrigation strategies based on current status and historical data, with the goal of maximizing cumulative rewards. At the same time, the effect evaluation module tracks the effects of drip irrigation after execution, providing timely feedback and adjusting strategies. This closed-loop mechanism ensures the long-term and efficient operation of the irrigation system, significantly improving the efficiency and sustainability of agricultural production, and addressing the previous problem of irrigation systems being unable to automatically optimize and adjust.

[0102] In summary, the fully automatic quantitative drip irrigation system forms a complete closed-loop management system through the close collaboration between the above modules. It not only greatly improves agricultural production efficiency and water resource utilization, but also effectively solves many challenges existing in traditional irrigation technology.

[0103] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0104] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0105] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. Fully automatic quantitative drip irrigation system, characterized by: include: The humidity monitoring module uses a distributed sensor network to measure the humidity data of the target planting area, introduces an adaptive calibration algorithm to correct the humidity data deviation, and performs quadratic mean processing to generate the humidity value of the target planting area; The meteorological analysis module uses the acquired weather forecast information as a basis and adopts an improved support vector machine model to predict the meteorological data of the target planting area. The goal of the improved support vector machine model is to minimize the set loss function. The water demand prediction module is based on humidity and meteorological data, and integrates the crop growth data of the target planting area. The data is input into a pre-built neural network model and outputs the predicted water demand of the target planting area; The intelligent control module executes the predetermined drip irrigation operation based on the predicted water demand; The system selectively decides whether to trigger a reinforcement learning-based optimization strategy based on the drip irrigation effect to adjust the operating status of the drip irrigation system in the target planting area, including the time to open / close the valve and the flow rate. The effect evaluation module continuously tracks the drip irrigation effect after the optimization strategy is implemented, observes the changing trends of crop growth indicators and humidity values, and generates effect evaluation indicators based on the changes in crop growth indicators, humidity values, and irrigation volume. When the effect evaluation indicator falls below the preset threshold, the optimization strategy based on reinforcement learning is triggered. Otherwise, the established drip irrigation operation will continue.

2. The fully automatic quantitative drip irrigation system according to claim 1, characterized in that: The content of the adaptive calibration algorithm is: Let H i (t) represents the humidity data measured by the i-th humidity sensor at time t, H t ′(t) is the humidity data after calibration, and the calibration process is expressed as: H i ′(t)=H i (t)+α(H_avg(t)-H i (t)); Where H_avg(t) represents the average reading of all humidity sensors in the sensor network at time point t, that is, the average humidity data, and α represents the calibration coefficient, which ranges from 0 to 1.

3. The fully automatic quantitative drip irrigation system according to claim 1, characterized in that: Weather forecast information includes at least temperature, rainfall, and wind speed. Let X be the first input eigenvector and y be the predicted target variable. The goal of the improved support vector machine model is to minimize the following loss function: The input first eigenvector X includes a number of historical weather data points, and the historical weather data points include at least: the average temperature of the past several days, the average relative humidity of the past several days, the accumulated precipitation of the past several days, and the average wind speed of the past several days; Let X = [x1, x2, ..., x m ], where m represents the number of first eigenvectors; In the above formula, w represents the weight vector, b represents the bias term, C represents the dynamic penalty factor, i represents the i-th training sample, n represents the total number of training samples, and y i The true label of the i-th sample; x i The feature vector of the i-th sample.

4. The fully automatic quantitative drip irrigation system according to claim 3, characterized in that: In the improved support vector machine model, the penalty factor in the target loss function is derived based on the adaptive learning rate and loss function feedback mechanism; The operation process of the feedback mechanism is: Initialization: the penalty factor C of all samples i Set to the same initial value C0; Dynamic adjustment: In each iteration, adjust the corresponding penalty factor C according to the performance of each sample based on the improved support vector machine model i Feedback mechanism: This feedback mechanism is introduced to use historical data and the performance of the current improved support vector machine model to predict future adjustment directions.

5. The fully automatic quantitative drip irrigation system according to claim 1, characterized in that: Crop growth data at least include: the root water absorption capacity value and leaf transpiration rate of the crop; wherein, the root water absorption capacity value is calculated by weighting the soil moisture and the maximum depth of the root distribution; the leaf transpiration rate is measured by a plant lysimeter.

6. The fully automatic quantitative drip irrigation system according to claim 1, characterized in that: In the water demand forecasting module, the second eigenvector is obtained after data integration; The second eigenvector is input into the neural network model to output the predicted water requirement of the target planting area; The neural network model is expressed as: Q = f(X2; W); Where Q represents the output variable, i.e., the predicted water demand; X2 is the second eigenvector set, which includes at least humidity values, meteorological data, and crop growth data; and W represents the weight parameter in the neural network model. The model is adjusted continuously until the error between the neural network model output and the actual observation value is minimized, and the water demand prediction result can be obtained.

7. The fully automatic quantitative drip irrigation system according to claim 1, characterized in that: The optimization strategy based on reinforcement learning is: Let R(a v , s v ) Take action a in the current state v The reward value obtained after Among them, the reward value is used to evaluate the quality of any action; v As the objective function, at the state of time step v, the goal of reinforcement learning is to find where γ is the discount factor, ranging from 0 to 1.

8. The fully automatic quantitative drip irrigation system according to claim 1, characterized in that: The change in crop growth indicators is expressed as: the change in the comprehensive value obtained by weighting plant height and leaf area index LAI; The changes are: the value after drip irrigation minus the value before drip irrigation; The formula for generating the performance evaluation index is: Where Er represents the effect evaluation index, δ1 and δ2 represent weight coefficients, both ranging from 0 to 1, ΔGr represents the change in crop growth index before and after drip irrigation, ΔH represents the change in humidity before and after drip irrigation, and Ir represents the actual irrigation amount.

Images