Irrigation water utilization coefficient intelligent monitoring method and system based on multi-source perception
By integrating multi-source data and optimizing models, the accuracy and efficiency issues of traditional irrigation water utilization coefficient calculations have been resolved, enabling efficient and precise irrigation water resource management.
Patent Information
- Application Number
- CN202511406150.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-29
- Publication Date
- 2026-01-13
AI Technical Summary
Traditional methods for calculating irrigation water use coefficient rely on manual measurement, which is time-consuming, complex, and inaccurate, making it difficult to fully reflect the water use situation in the irrigation area and resulting in low irrigation efficiency.
Multi-source data is collected using UAV spectral sensing, ground-based IoT sensors, and crop growth models. Feature fusion is performed using a CNN-LSTM-RF model, and irrigation parameters are optimized using the PPO algorithm to construct a three-dimensional digital twin model of the irrigation area. The model can then monitor and output the optimal irrigation strategy in real time.
It improves the monitoring accuracy of irrigation water utilization coefficient, provides scientific irrigation decision-making, saves water resources, and achieves precision irrigation management.
Smart Images

Figure CN121329216A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of agricultural irrigation monitoring technology, specifically to an intelligent monitoring method and system for irrigation water utilization coefficient based on multi-source sensing. Background Technology
[0002] The irrigation water use efficiency coefficient is a key indicator for measuring irrigation water use efficiency and is crucial for the rational utilization and management of agricultural water resources. Traditional methods for calculating the irrigation water use efficiency coefficient mainly rely on manual measurement, such as the typical canal section measurement method and the comprehensive measurement method. These methods have many drawbacks, such as being time-consuming and requiring a large amount of manpower throughout the measurement period; the calculation process is cumbersome, involving numerous parameters and complex formulas, which makes it easy for human calculation errors to occur; and the number of irrigation statistics data samples is insufficient, making it difficult to comprehensively and accurately reflect the water use situation of the entire irrigation area, and the utilization is not sufficient, resulting in inaccurate final coefficient calculation results. Summary of the Invention
[0003] The purpose of this invention is to solve the above problems by designing an intelligent monitoring method and system for irrigation water utilization coefficient based on multi-source sensing.
[0004] The first aspect of this invention provides an intelligent monitoring method for irrigation water use coefficient based on multi-source sensing, the method comprising the following steps:
[0005] Multi-source data is collected through UAV spectral sensing, ground IoT sensors and crop growth models. The multi-source data is then spatiotemporally aligned, outlier removed and standardized to construct a dataset.
[0006] Construct a three-dimensional digital twin model of the irrigation district to simulate the hydrological processes under different irrigation strategies and generate a virtual benchmark value for irrigation water utilization coefficient;
[0007] The target feature data is obtained by fusing multi-source features from the dataset using the CNN-LSTM-RF model.
[0008] The target feature data is input into the PPO algorithm to dynamically optimize irrigation parameters, maximize the actual irrigation water utilization coefficient, and output the optimal irrigation strategy.
[0009] Irrigation is carried out based on the optimal irrigation strategy, and the irrigation water utilization coefficient is monitored in real time. When the irrigation water utilization coefficient is lower than the set threshold, an early warning message is automatically issued.
[0010] Optionally, in a first implementation of the first aspect of the present invention, the collection of multi-source data through UAV spectral sensing, ground-based IoT sensors, and crop growth models includes:
[0011] Data on crop canopy reflectance were obtained using a drone's hyperspectral camera to determine soil moisture content and leaf area index.
[0012] Soil moisture, temperature, humidity, wind speed, and airflow data are acquired through ground-based IoT sensors.
[0013] The actual evapotranspiration of crops was calculated using the Penman-Monteith model, and the crop growth status was assessed by combining the NDVI index.
[0014] Optionally, in a second implementation of the first aspect of the present invention, the step of performing spatiotemporal alignment, outlier removal, and standardization on multi-source data to construct a dataset includes:
[0015] For data from different sources, timestamps and spatial coordinates are extracted, the DTW algorithm is used to calculate the local time offset between each data sequence, and the time axis is aligned by dynamic programming path matching.
[0016] By combining the convolutional gated recurrent unit model and aligning the spatial resolution, low-resolution data is interpolated into the spatial grid of high-resolution data, thus completing the spatiotemporal alignment processing of multi-source data.
[0017] Identify and label the first outlier that exceeds the statistical range in spatiotemporally aligned multi-source data, and use the Isolation Forest algorithm to detect the second outlier by calculating the path length, and remove the outliers that are double-labeled;
[0018] The Z-score standardization method is used to standardize the data after removing outliers to obtain the dataset.
[0019] Optionally, in a third implementation of the first aspect of the present invention, the step of constructing a three-dimensional digital twin model of the irrigation district to simulate hydrological processes under different irrigation strategies and generate a virtual irrigation water use coefficient benchmark value includes:
[0020] By integrating topographic data, soil type distribution maps, and CAD models of irrigation facilities, a three-dimensional digital twin model of the irrigation area is constructed using the Unity3D engine;
[0021] Adjustable irrigation strategy parameters can be set to form 100 strategy combinations. The irrigation strategy parameters include irrigation time, irrigation volume, irrigation method, and rotation irrigation cycle.
[0022] In the three-dimensional digital twin model of the irrigation area, the irrigation process of 100 strategies is simulated in sequence, the virtual irrigation water utilization coefficient is calculated, and the baseline value of the virtual irrigation water utilization coefficient is generated.
[0023] Optionally, in a fourth implementation of the first aspect of the present invention, the step of fusing multi-source features from the data in the dataset using a CNN-LSTM-RF model to obtain target feature data includes:
[0024] The dataset is input into the CNN-LSTM-RF model. The CNN network extracts high-dimensional features step by step through multiple convolutional layers and pooling layers to obtain spatial features. The convolutional layers use convolutional kernels of different sizes to slide and scan the data to extract local correlation features, while the pooling layers reduce the feature dimensionality and enhance spatial invariance.
[0025] In LSTM networks, the input gate, forget gate, and output gate mechanisms in LSTM units dynamically control the transmission and updating of data in the time dimension, capturing long-term dependencies. At each time step, the LSTM unit outputs the hidden feature vector of the current state, and the hidden feature vectors of all time steps are concatenated to obtain the time features.
[0026] The spatial features extracted by the CNN network and the temporal features extracted by the LSTM network are concatenated and fused to form a joint feature vector, which is then output as preliminary target feature data through a fully connected layer.
[0027] The RF algorithm is used to optimize and filter the initial target feature data. Each decision tree randomly selects some features for comparison when splitting at a node. The importance of each feature is evaluated through the splitting path of each tree. The output results of all decision trees are voted on to generate the final target feature data.
[0028] Optionally, in the fifth implementation of the first aspect of the present invention, the joint feature vector is subjected to nonlinear transformation and dimensionality reduction in the fully connected layer. The feature dimension is compressed layer by layer through the multilayer perceptron structure, and the ReLU activation function is used to introduce nonlinear factors in each layer.
[0029] Optionally, in a sixth implementation of the first aspect of the present invention, the step of inputting the target feature data into the PPO algorithm, dynamically optimizing irrigation parameters, maximizing the actual irrigation water utilization coefficient, and outputting the optimal irrigation strategy includes:
[0030] The target feature data is defined as the state space and the irrigation parameters as the action space. A reward function is constructed based on the actual irrigation water utilization coefficient, crop yield increase, energy consumption reduction, and soil salinization risk reduction.
[0031] Construct a policy network and a value network, using the target feature data generated by the CNN-LSTM-RF model as the initial state input. The policy network outputs the action probability distribution, and the value network outputs the expected cumulative reward.
[0032] Run the current strategy in the simulated irrigation district environment, perform operations according to the irrigation actions output by the strategy network, record the state, action, reward and next state of each step, and store them in the experience pool.
[0033] The optimal action of the strategy network is extracted, mapped to the actual irrigation system parameters, and combined with the spatiotemporal distribution characteristics of the irrigation area to generate regional and time-segmented irrigation strategies.
[0034] A second aspect of the present invention provides an intelligent monitoring system for irrigation water use coefficient based on multi-source sensing, the system comprising:
[0035] The data acquisition module is used to collect multi-source data through UAV spectral sensing, ground IoT sensors and crop growth models, and to perform spatiotemporal alignment, outlier removal and standardization on the multi-source data to build a dataset.
[0036] The module is used to build a three-dimensional digital twin model of the irrigation area, simulate the hydrological process under different irrigation strategies, and generate a virtual irrigation water use coefficient benchmark value.
[0037] The feature fusion module is used to perform multi-source feature fusion on the data in the dataset using the CNN-LSTM-RF model to obtain target feature data.
[0038] The dynamic optimization module is used to input target feature data into the PPO algorithm, dynamically optimize irrigation parameters, maximize the actual irrigation water utilization coefficient, and output the optimal irrigation strategy.
[0039] The early warning module is used to irrigate based on the optimal irrigation strategy and monitor the irrigation water utilization coefficient in real time. When the irrigation water utilization coefficient is lower than the set threshold, it will automatically issue an early warning message.
[0040] A third aspect of the present invention provides an intelligent monitoring device for irrigation water utilization coefficient based on multi-source sensing, the intelligent monitoring device for irrigation water utilization coefficient based on multi-source sensing includes a memory and at least one processor, the memory storing instructions; the at least one processor calls the instructions in the memory to cause the intelligent monitoring device for irrigation water utilization coefficient based on multi-source sensing to perform the various steps of the intelligent monitoring method for irrigation water utilization coefficient based on multi-source sensing as described in any of the preceding claims.
[0041] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the intelligent monitoring method for irrigation water utilization coefficient based on multi-source sensing as described in any of the preceding claims.
[0042] The technical solution provided by this invention collects multi-source data through UAV spectral sensing, ground-based IoT sensors, and crop growth models. This multi-source data undergoes spatiotemporal alignment, outlier removal, and standardization to construct a dataset. A three-dimensional digital twin model of the irrigation district is constructed to simulate hydrological processes under different irrigation strategies, generating a virtual baseline value for the irrigation water utilization coefficient. A CNN-LSTM-RF model is used to fuse multi-source features in the dataset to obtain target feature data. This target feature data is input into the PPO algorithm to dynamically optimize irrigation parameters, maximize the actual irrigation water utilization coefficient, and output the optimal irrigation strategy. Irrigation is carried out based on the optimal irrigation strategy, and the irrigation water utilization coefficient is monitored in real time. When the irrigation water utilization coefficient falls below a set threshold, an automatic warning is issued. This invention fully considers the influence of various factors such as soil, weather, crops, and irrigation facilities on the irrigation water utilization coefficient. Compared with traditional methods that rely on only a single or a few parameters for calculation, it greatly improves the accuracy of monitoring results. It can provide managers with scientific and reasonable irrigation decision-making suggestions based on real-time monitoring data, improving irrigation efficiency, saving water resources, and achieving precise monitoring and management of the irrigation water utilization coefficient. Attached Figure Description
[0043] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.
[0044] Figure 1 A flowchart of an intelligent monitoring method for irrigation water utilization coefficient based on multi-source sensing provided in an embodiment of the present invention;
[0045] Figure 2 A schematic diagram of the structure of an intelligent monitoring system for irrigation water utilization coefficient based on multi-source sensing provided in an embodiment of the present invention;
[0046] Figure 3 This is a schematic diagram of the structure of an intelligent monitoring device for irrigation water utilization coefficient based on multi-source sensing, provided in an embodiment of the present invention. Detailed Implementation
[0047] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.
[0048] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The flowchart of the intelligent monitoring method for irrigation water use coefficient based on multi-source sensing provided in this embodiment of the invention includes the following steps:
[0049] Step 101: Collect multi-source data through UAV spectral sensing, ground IoT sensors and crop growth models, perform spatiotemporal alignment, outlier removal and standardization on the multi-source data, and construct a dataset;
[0050] In this embodiment, crop canopy reflectance data is acquired using a drone's hyperspectral camera. To determine soil moisture content and leaf area index, a drone equipped with a 400-1000nm wavelength hyperspectral camera is used. Aerial photography is conducted at a flight altitude of 150 meters, with a frequency of once per hour. At this altitude, the ground resolution reaches 0.05 meters, accurately capturing the subtle spectral characteristics of the crop canopy. The acquired reflectance data undergoes radiometric and atmospheric correction to eliminate interference factors such as lighting conditions and atmospheric scattering. Then, characteristic bands are extracted—an inversion model is constructed using the 650-680nm red band and the 800-900nm near-infrared band. For soil moisture content inversion, the correlation between crop canopy reflectance and soil moisture is analyzed, and the shading effect of vegetation cover on the soil background is considered. A partial least squares regression-based inversion model is established, controlling the inversion error to within 3%. For leaf area index inversion, the nonlinear relationship between the normalized vegetation index and the leaf area index is utilized, and a red-edge band optimization model is introduced to ensure that the coefficient of determination of the inversion results is ≥0.9.
[0051] When acquiring soil moisture, temperature, humidity, wind speed, and flow data using ground-based IoT sensors, a combination of layered deployment and key node placement is employed. For soil monitoring, soil moisture and temperature sensors are deployed in three layers (0-20cm, 20-40cm, and 40-60cm), with an 8-meter spacing between sensors to ensure coverage of different soil textures. The soil moisture sensors utilize the frequency domain reflectance method with a measurement accuracy of ±2%, while the temperature sensors measure from -10 to 60℃ with an accuracy of ±0.3℃, collecting data every 3 minutes to provide real-time updates. The hydrothermal conditions of soil at different depths were investigated. For meteorological data collection, four meteorological monitoring points were evenly distributed throughout the irrigation area. Each monitoring point was equipped with a temperature and humidity sensor and a wind speed sensor. The temperature and humidity sensors could measure temperature from 0-50℃ and relative humidity from 0-100%, with accuracies of ±0.2℃ and ±2%, respectively. The wind speed sensor had a measurement range of 0-30 m / s and an accuracy of ±0.1 m / s. Data was recorded every 5 minutes to provide basic meteorological parameters for evapotranspiration calculation. For irrigation flow monitoring, ultrasonic flow meters were installed at the main irrigation canal, branch canals, and field outlets, with a measurement range of 0.1-10 m³ / s. 3 / s, with an accuracy of ±0.3%, instantaneous flow and cumulative flow data are collected every minute. All sensor data are transmitted to the data center in real time via LoRa wireless transmission technology with a transmission delay of ≤10 seconds, ensuring the timeliness and continuity of the data.
[0052] When calculating actual crop evapotranspiration using the Penman-Monteith model and assessing crop growth status using the NDVI index, meteorological data such as temperature, humidity, and wind speed obtained from ground-based IoT sensors are first input into the model to calculate the reference crop evapotranspiration. Parameters involved in the model, such as saturated vapor pressure and net radiation, are all accurately calculated using measured meteorological data to ensure that the ET0 calculation error is ≤5%. Then, a crop coefficient is determined based on crop type and growth stage, and Kc is dynamically corrected using the leaf area index retrieved from UAVs to obtain the actual crop evapotranspiration, which accurately reflects the crop's current water requirement. Simultaneously, the NDVI index is calculated using UAV hyperspectral data. By analyzing the spatial distribution and temporal changes of the NDVI value, the uniformity and growth status of crop growth are assessed. When the NDVI value is below 0.3, it indicates weak crop growth; when the NDVI value is between 0.5 and 0.8, the crop growth is good. Combining actual evapotranspiration with the NDVI index not only accurately determines whether the crop is under water stress but also provides a scientific basis for adjusting irrigation strategies, enabling on-demand irrigation.
[0053] In this embodiment, timestamps and spatial coordinates are extracted from data from different sources. The DTW algorithm is used to calculate the local time offset between data sequences. When aligning the time axis using dynamic programming path matching, the original timestamps and spatial coordinates are first extracted from multi-source data such as UAV hyperspectral data and ground IoT sensor data. Since the sampling frequencies of different devices differ, the DTW algorithm is used to process the time series differences: the time series of each data source is transformed into a sequence matrix with the sampling time and data value as the core. Using the UAV data time axis as a reference, the cumulative distance between the sensor data and the reference sequence is calculated. The optimal matching path is found through dynamic programming to determine the local time offset. The sensor data timestamps are corrected according to the offset, so that the matching error of all data on the time axis is controlled within 10 seconds, ensuring the consistency of the time dimension in subsequent analysis.
[0054] By combining a convolutional gated recurrent unit (GRU) model to align spatial resolution, low-resolution data is interpolated into the spatial grid of high-resolution data. When performing spatiotemporal alignment of multi-source data, the spatial resolution of each data source is first determined: UAV hyperspectral data has a spatial resolution of 0.05m × 0.05m, ground sensor data corresponds to an 8m × 8m resolution due to the spacing between data points, and meteorological station data covers a larger area. Using the spatial grid of UAV data as the baseline grid, the GRU model processes the low-resolution data: convolutional layers extract spatial features, and the GRU captures potential spatiotemporal correlations, such as the correlation between soil moisture in a certain area and surrounding meteorological conditions. The model outputs the predicted values of the low-resolution data in the high-resolution grid. The predicted values are further optimized using bicubic interpolation to ensure that the spatial error of the low-resolution data in the high-resolution grid is ≤0.5m, ultimately achieving accurate alignment of all data on a 0.05m × 0.05m spatial grid and a unified time axis.
[0055] The algorithm identifies and labels first outliers that exceed the statistical range in spatiotemporally aligned multi-source data. Then, it uses the Isolation Forest algorithm to detect second outliers by calculating path length. When removing double-labeled outliers, the algorithm first determines the statistical range of each parameter based on historical data, traverses the spatiotemporally aligned data, and labels data exceeding this range as first outliers. Next, it constructs an Isolation Forest model, using the spatiotemporally aligned dataset as training samples, including normal data and known outliers. Multiple isolation trees are built by randomly partitioning the feature space, and the average path length of each data point in the isolation tree is calculated—the shorter the path length, the more likely the data point is to be an outlier. A path length threshold is set, and data below the threshold are labeled as second outliers. Finally, the data undergoes double verification, removing only data that is simultaneously labeled as both first and second outliers. This avoids misjudgments by a single method, such as short-term sudden increases in soil moisture which may be a normal phenomenon caused by irrigation, and retains data only if labeled by a single method, thus improving the outlier removal accuracy to over 90%.
[0056] Z-score standardization was used to standardize the outlier-removed data to obtain the dataset. First, the mean and standard deviation of each parameter after outlier removal were calculated, where the mean is the arithmetic mean of all data, and the standard deviation reflects the dispersion of the data. Each data point was then standardized according to a formula, mapping data of different dimensions to a standard normal distribution with a mean of 0 and a standard deviation of 1. Extreme data points exceeding the range [-3, 3] after standardization were truncated to avoid interference from extreme values in subsequent model training. The standardized dataset retains the distribution characteristics and relative relationships of the original data while eliminating dimensional differences, providing a unified data foundation for multi-source data fusion analysis and model computation, thus improving model convergence speed and prediction accuracy.
[0057] Step 102: Construct a three-dimensional digital twin model of the irrigation area to simulate the hydrological process under different irrigation strategies and generate a virtual irrigation water utilization coefficient benchmark value.
[0058] In this embodiment, topographic data, soil type distribution maps, and CAD models of irrigation facilities are integrated to construct a three-dimensional digital twin model of the irrigation area using the Unity3D engine. Adjustable irrigation strategy parameters are set to form 100 sets of strategy combinations. The irrigation strategy parameters include irrigation time, irrigation volume, irrigation method, and rotation irrigation cycle. The irrigation process of the 100 sets of strategies is simulated sequentially in the three-dimensional digital twin model of the irrigation area to calculate the virtual irrigation water utilization coefficient and generate a baseline value for the virtual irrigation water utilization coefficient.
[0059] Step 103: Use the CNN-LSTM-RF model to perform multi-source feature fusion on the data in the dataset to obtain the target feature data;
[0060] In this embodiment, the dataset is input into the CNN-LSTM-RF model. The CNN network extracts high-dimensional features step-by-step through multiple convolutional and pooling layers to obtain spatial features. The convolutional layers use kernels of different sizes to perform sliding scans on the data, extracting locally correlated features. The pooling layers reduce feature dimensionality and enhance spatial invariance. The LSTM network dynamically controls the transmission and updating of data over time through input gates, forget gates, and output gates within the LSTM units, capturing long-term dependencies. At each time step, the LSTM unit outputs the hidden feature vector of the current state. The hidden feature vectors from all time steps are concatenated to obtain the temporal features. The spatial features extracted by the CNN network and the temporal features extracted by the LSTM network are concatenated and fused to form a joint feature vector. This joint feature vector is then subjected to nonlinear transformation and dimensionality reduction in a fully connected layer. The feature dimension is compressed layer by layer using a multilayer perceptron structure, with each layer employing the ReLU activation function to introduce nonlinearity. Preliminary target feature data is then output through the fully connected layer. The Random Forest (RF) algorithm is used to optimize and filter this preliminary target feature data. Each decision tree randomly selects some features for comparison during node splitting, and the importance of each feature is evaluated based on the splitting path of each tree. Finally, the outputs of all decision trees are voted on to generate the final target feature data.
[0061] Step 104: Input the target feature data into the PPO algorithm to dynamically optimize irrigation parameters, maximize the actual irrigation water utilization coefficient, and output the optimal irrigation strategy.
[0062] In this embodiment, the target feature data is defined as a state space, with each state represented by a multi-dimensional feature vector. Subsequently, irrigation parameters such as irrigation duration, frequency, water allocation ratio, and sprinkler / drip irrigation mode are defined as discrete or continuous action spaces to ensure that the action dimensions match the adjustable parameters of the actual irrigation system. Finally, a state transition function is set to describe the changes in soil moisture and crop growth status over time under specific irrigation actions. The main reward is defined as the actual irrigation water utilization coefficient, i.e., the ratio of effective irrigation water volume to total water supply, quantified using sensor data or simulation results. Secondly, auxiliary reward items are introduced, such as increased crop yield, reduced energy consumption, and reduced soil salinization risk, which are weighted and combined to form a comprehensive reward function. Finally, penalty items are set, such as water waste due to over-irrigation and crop yield reduction due to insufficient irrigation, to ensure that the reward function comprehensively reflects the advantages and disadvantages of the strategy.
[0063] A policy network and a value network are constructed to generate irrigation actions and evaluate state value, respectively. The policy network takes state features as input and outputs an action probability distribution, while the value network takes the state as input and outputs the expected cumulative reward. Hyperparameters are then set: pruning range, discount factor, λ value of GAE, and learning rate. Finally, target feature data generated by a pre-trained CNN-LSTM-RF model is loaded as the initial state input. The current policy is run in a simulated irrigation environment, performing operations such as opening valves and adjusting water volume based on the irrigation actions output by the policy network. The state, action, reward, and next state are then recorded for each step. Finally, the complete round of experience data, including the state sequence, action sequence, and reward sequence, is stored in an experience pool for subsequent policy updates.
[0064] Based on the value network, the value function of the current state is predicted. Combining the actual reward with the value estimate of the next state, the time difference error is calculated. Then, the TD errors from multiple time steps are weighted and summed using the GAE method to generate the advantage function for each action, which measures the superiority of that action relative to the average policy. Finally, the bias and variance balance of the advantage estimate are adjusted by combining the discount factor and the λ parameter to ensure its effectiveness in long-term policy optimization. The probability ratio of the new and old policies is calculated, and the credibility of the action selection is evaluated through importance sampling. Subsequently, a pruning objective function is constructed: taking the minimum value between the non-pruned and pruned terms to limit the policy update magnitude. Finally, the objective function is maximized using gradient ascent to update the policy network parameters, ensuring that the new policy gradually improves within the trust region.
[0065] The algorithm calculates the error between the predicted value of the value network and the actual cumulative return; then, it constructs a value loss function; finally, it minimizes this loss function using gradient descent to adjust the value network parameters, enabling it to more accurately assess state value and providing a reliable basis for policy optimization; it evaluates key indicators such as water use efficiency and crop yield of the current policy on a validation set; then, it compares the performance differences between the new and old policies; if the new policy is significantly better than the old policy, it retains the update; otherwise, it rolls back to the historical best policy; finally, it sets the policy update frequency to avoid premature convergence or frequent perturbations, ensuring the stability of the algorithm.
[0066] The optimal actions of the strategy network under a given state, such as the optimal irrigation duration and water allocation ratio, are extracted. Then, the actions are mapped to actual irrigation system parameters, such as valve opening and pump start-up and shutdown times. Finally, combined with the spatiotemporal distribution characteristics of the irrigation area, such as the differences in water demand among different fields, regional and time-segmented irrigation strategies are generated and executed in real time through an intelligent irrigation control system to dynamically adjust irrigation decisions to maximize water use efficiency.
[0067] Step 105: Irrigate based on the optimal irrigation strategy and monitor the irrigation water utilization coefficient in real time. When the irrigation water utilization coefficient is lower than the set threshold, an early warning message will be automatically issued.
[0068] Please see Figure 2 A schematic diagram of the structure of the intelligent monitoring system for irrigation water utilization coefficient based on multi-source sensing provided in this embodiment of the invention. The system includes:
[0069] The data acquisition module is used to collect multi-source data through UAV spectral sensing, ground IoT sensors and crop growth models, and to perform spatiotemporal alignment, outlier removal and standardization on the multi-source data to build a dataset.
[0070] The module is used to build a three-dimensional digital twin model of the irrigation area, simulate the hydrological process under different irrigation strategies, and generate a virtual irrigation water use coefficient benchmark value.
[0071] The feature fusion module is used to perform multi-source feature fusion on the data in the dataset using the CNN-LSTM-RF model to obtain target feature data.
[0072] The dynamic optimization module is used to input target feature data into the PPO algorithm, dynamically optimize irrigation parameters, maximize the actual irrigation water utilization coefficient, and output the optimal irrigation strategy.
[0073] The early warning module is used to irrigate based on the optimal irrigation strategy and monitor the irrigation water utilization coefficient in real time. When the irrigation water utilization coefficient is lower than the set threshold, it will automatically issue an early warning message.
[0074] Figure 3 This is a schematic diagram of the structure of an intelligent monitoring device for irrigation water utilization coefficient based on multi-source sensing, provided in an embodiment of the present invention. The intelligent monitoring device 300 for irrigation water utilization coefficient based on multi-source sensing can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) for storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the intelligent monitoring device 300 for irrigation water utilization coefficient based on multi-source sensing. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute the series of instruction operations in the storage media 330 on the intelligent monitoring device 300 for irrigation water utilization coefficient based on multi-source sensing to implement the method provided in the above embodiment.
[0075] The intelligent monitoring device 300 for irrigation water utilization coefficient based on multi-source sensing may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The structure of the intelligent monitoring device for irrigation water utilization coefficient based on multi-source sensing shown does not constitute a limitation on the computer device provided by the present invention. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.
[0076] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the various steps of the intelligent monitoring method for irrigation water utilization coefficient based on multi-source sensing provided in the above embodiments.
[0077] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described equipment or apparatus / unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.
[0078] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0079] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. A method for intelligent monitoring of irrigation water use coefficient based on multi-source sensing, characterized in that, The method includes the following steps: Multi-source data is collected through UAV spectral sensing, ground IoT sensors and crop growth models. The multi-source data is then spatiotemporally aligned, outlier removed and standardized to construct a dataset. Construct a three-dimensional digital twin model of the irrigation district to simulate the hydrological processes under different irrigation strategies and generate a virtual benchmark value for irrigation water utilization coefficient; The target feature data is obtained by fusing multi-source features from the dataset using the CNN-LSTM-RF model. The target feature data is input into the PPO algorithm to dynamically optimize irrigation parameters, maximize the actual irrigation water utilization coefficient, and output the optimal irrigation strategy. Irrigation is carried out based on the optimal irrigation strategy, and the irrigation water utilization coefficient is monitored in real time. When the irrigation water utilization coefficient is lower than the set threshold, an early warning message is automatically issued.
2. The intelligent monitoring method for irrigation water use coefficient based on multi-source sensing as described in claim 1, characterized in that, The method of collecting multi-source data through UAV spectral sensing, ground-based IoT sensors, and crop growth models includes: Data on crop canopy reflectance were obtained using a drone's hyperspectral camera to determine soil moisture content and leaf area index. Soil moisture, temperature, humidity, wind speed, and airflow data are acquired through ground-based IoT sensors. The actual evapotranspiration of crops was calculated using the Penman-Monteith model, and the crop growth status was assessed by combining the NDVI index.
3. The intelligent monitoring method for irrigation water use coefficient based on multi-source sensing as described in claim 1, characterized in that, The process of performing spatiotemporal alignment, outlier removal, and standardization on multi-source data to construct a dataset includes: For data from different sources, timestamps and spatial coordinates are extracted, the DTW algorithm is used to calculate the local time offset between each data sequence, and the time axis is aligned by dynamic programming path matching. By combining the convolutional gated recurrent unit model and aligning the spatial resolution, low-resolution data is interpolated into the spatial grid of high-resolution data, thus completing the spatiotemporal alignment processing of multi-source data. Identify and label the first outlier that exceeds the statistical range in spatiotemporally aligned multi-source data, and use the Isolation Forest algorithm to detect the second outlier by calculating the path length, and remove the outliers that are double-labeled; The Z-score standardization method is used to standardize the data after removing outliers to obtain the dataset.
4. The intelligent monitoring method for irrigation water use coefficient based on multi-source sensing as described in claim 1, characterized in that, The construction of a three-dimensional digital twin model of the irrigation district, simulating hydrological processes under different irrigation strategies to generate virtual irrigation water use coefficient benchmark values, includes: By integrating topographic data, soil type distribution maps, and CAD models of irrigation facilities, a three-dimensional digital twin model of the irrigation area is constructed using the Unity3D engine; Adjustable irrigation strategy parameters can be set to form 100 strategy combinations. The irrigation strategy parameters include irrigation time, irrigation volume, irrigation method, and rotation irrigation cycle. In the three-dimensional digital twin model of the irrigation area, the irrigation process of 100 strategies is simulated in sequence, the virtual irrigation water utilization coefficient is calculated, and the baseline value of the virtual irrigation water utilization coefficient is generated.
5. The intelligent monitoring method for irrigation water use coefficient based on multi-source sensing as described in claim 1, characterized in that, The process of fusing multi-source features from the dataset using a CNN-LSTM-RF model to obtain target feature data includes: The dataset is input into the CNN-LSTM-RF model. The CNN network extracts high-dimensional features step by step through multiple convolutional layers and pooling layers to obtain spatial features. The convolutional layers use convolutional kernels of different sizes to slide and scan the data to extract local correlation features, while the pooling layers reduce the feature dimensionality and enhance spatial invariance. In LSTM networks, the input gate, forget gate, and output gate mechanisms in LSTM units dynamically control the transmission and updating of data in the time dimension, capturing long-term dependencies. At each time step, the LSTM unit outputs the hidden feature vector of the current state, and the hidden feature vectors of all time steps are concatenated to obtain the time features. The spatial features extracted by the CNN network and the temporal features extracted by the LSTM network are concatenated and fused to form a joint feature vector, which is then output as preliminary target feature data through a fully connected layer. The RF algorithm is used to optimize and filter the initial target feature data. Each decision tree randomly selects some features for comparison when splitting at a node. The importance of each feature is evaluated through the splitting path of each tree. The output results of all decision trees are voted on to generate the final target feature data.
6. The intelligent monitoring method for irrigation water use coefficient based on multi-source sensing as described in claim 5, characterized in that, In the fully connected layer, the joint feature vector is subjected to nonlinear transformation and dimensionality reduction. The feature dimension is compressed layer by layer through the multilayer perceptron structure, and the ReLU activation function is used to introduce nonlinear factors in each layer.
7. The intelligent monitoring method for irrigation water use coefficient based on multi-source sensing as described in claim 1, characterized in that, The process of inputting target feature data into the PPO algorithm to dynamically optimize irrigation parameters, maximize the actual irrigation water utilization coefficient, and output the optimal irrigation strategy includes: The target feature data is defined as the state space and the irrigation parameters as the action space. A reward function is constructed based on the actual irrigation water utilization coefficient, crop yield increase, energy consumption reduction, and soil salinization risk reduction. Construct a policy network and a value network, using the target feature data generated by the CNN-LSTM-RF model as the initial state input. The policy network outputs the action probability distribution, and the value network outputs the expected cumulative reward. Run the current strategy in the simulated irrigation district environment, perform operations according to the irrigation actions output by the strategy network, record the state, action, reward and next state of each step, and store them in the experience pool. The optimal action of the strategy network is extracted, mapped to the actual irrigation system parameters, and combined with the spatiotemporal distribution characteristics of the irrigation area to generate regional and time-segmented irrigation strategies.
8. An intelligent monitoring system for irrigation water use coefficient based on multi-source sensing, characterized in that, The system includes: The data acquisition module is used to collect multi-source data through UAV spectral sensing, ground IoT sensors and crop growth models, and to perform spatiotemporal alignment, outlier removal and standardization on the multi-source data to build a dataset. The module is used to build a three-dimensional digital twin model of the irrigation area, simulate the hydrological process under different irrigation strategies, and generate a virtual irrigation water use coefficient benchmark value. The feature fusion module is used to perform multi-source feature fusion on the data in the dataset using the CNN-LSTM-RF model to obtain target feature data. The dynamic optimization module is used to input target feature data into the PPO algorithm, dynamically optimize irrigation parameters, maximize the actual irrigation water utilization coefficient, and output the optimal irrigation strategy. The early warning module is used to irrigate based on the optimal irrigation strategy and monitor the irrigation water utilization coefficient in real time. When the irrigation water utilization coefficient is lower than the set threshold, it will automatically issue an early warning message.
9. An intelligent monitoring device for irrigation water utilization coefficient based on multi-source sensing, characterized in that, The intelligent monitoring device for irrigation water utilization coefficient based on multi-source sensing includes a memory and at least one processor. The memory stores instructions. The at least one processor calls the instructions in the memory to cause the intelligent monitoring device for irrigation water utilization coefficient based on multi-source sensing to perform each step of the intelligent monitoring method for irrigation water utilization coefficient based on multi-source sensing as described in any one of claims 1-7.
10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the various steps of the intelligent monitoring method for irrigation water utilization coefficient based on multi-source sensing as described in any one of claims 1-7.
Citation Information
Cited By
Farmland intelligent irrigation and drainage decision-making method and equipment based on multi-source perception and reinforcement learning
CN121386434A
Ecological model fused drip irrigation mode evaluation and optimization method and system
CN122066309A
