An intelligent improvement method, system, device and medium for an agricultural matching equipment

Abstract

This application relates to a method, system, equipment, and medium for intelligent improvement of supporting equipment in planting. The method includes: acquiring multi-source heterogeneous data from a target planting area, and performing data cleaning and spatiotemporal alignment on the multi-source heterogeneous data to obtain aligned multi-source heterogeneous data; inputting the aligned multi-source heterogeneous data into a multi-layer feature extraction network for hierarchical feature learning and correlation analysis to obtain equipment status feature data; inputting the equipment status feature data into an adaptive prediction model for trend prediction and anomaly detection to obtain equipment performance prediction data; performing multi-objective optimization analysis and parameter sensitivity evaluation on the equipment performance prediction data to obtain parameter optimization scheme data; converting the optimization scheme into control commands and executing them; and achieving closed-loop optimization and dynamic adjustment through performance feedback and reinforcement learning strategies. This method can improve the intelligence level and overall operational efficiency of supporting equipment in planting.

Description

Technical Field

[0001] This invention belongs to the field of agricultural planting technology, and in particular relates to a method, system, equipment and medium for intelligent improvement of supporting equipment for planting. Background Technology

[0002] With the rapid development of precision agriculture and smart planting technologies, the automation and informatization levels of supporting equipment in the planting industry are increasing day by day. In order to achieve more efficient agricultural production, the industry has generally begun to try to optimize and adjust the operating parameters of supporting equipment such as irrigation, fertilization, and plant protection.

[0003] In traditional technologies, improvements to planting equipment often rely on manual experience or simple rules based on a single data source. For example, irrigation might be performed solely according to a preset schedule, or water pump power might be adjusted based solely on soil moisture sensor readings. Equipment maintenance and parameter optimization are often reactive, only addressed after problems occur, or routine checks are conducted at fixed intervals, lacking foresight and a systematic approach.

[0004] However, current traditional methods have significant problems. First, due to the limited data sources and insufficient dimensions, they cannot comprehensively and accurately reflect the complex coupling relationship between crop growth environment and equipment operation status. Second, methods relying on human experience or static rules lack self-learning and adaptive capabilities, and cannot cope with dynamic changes such as sudden weather changes and spatial heterogeneity of soil conditions. Finally, adjustments to equipment operating parameters often focus only on a single objective (such as water conservation), making it difficult to achieve dynamic balance and overall optimization among multiple interdependent objectives such as equipment energy efficiency, operational quality, and resource consumption. This results in low overall equipment operating efficiency and hinders further improvement in the level of intelligent agriculture. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, system, equipment, and medium for intelligent improvement of supporting equipment for planting industry, addressing the aforementioned technical problems.

[0006] Firstly, this application provides a method for intelligent improvement of agricultural equipment, including:

[0007] S1. Obtain multi-source heterogeneous data of the target planting area, and perform data cleaning and spatiotemporal alignment on the multi-source heterogeneous data to obtain aligned multi-source heterogeneous data.

[0008] S2. Align the multi-source heterogeneous data and input it into a multi-layer feature extraction network to perform hierarchical feature learning and correlation analysis to obtain equipment status feature data; the equipment status feature data is used to characterize the spatiotemporal correlation features of the equipment operating status.

[0009] S3. Input the equipment status feature data into the adaptive prediction model to perform trend prediction and anomaly detection, and obtain equipment performance prediction data. The adaptive prediction model includes a trend prediction branch and an anomaly detection branch. The trend prediction branch is constructed based on a gated recurrent unit network, and the anomaly detection branch is constructed based on a support vector data description algorithm.

[0010] S4. Perform multi-objective optimization analysis and parameter sensitivity assessment on the equipment performance prediction data to obtain parameter optimization scheme data.

[0011] In one embodiment, S1 includes:

[0012] S11. Obtain the original multi-source heterogeneous data of the target planting area;

[0013] S12. Perform missing value imputation, noise filtering and outlier removal on the original multi-source heterogeneous data to obtain cleaned multi-source heterogeneous data.

[0014] S13. After cleaning, the multi-source heterogeneous data is uniformly converted to the preset spatiotemporal reference coordinate system and then meshed and aligned to obtain aligned multi-source heterogeneous data.

[0015] In one embodiment, S2 includes:

[0016] S21. Align the multi-source heterogeneous data and input it into the convolutional neural network layer of the multi-layer feature extraction network to extract local spatial features and obtain the primary spatial feature map.

[0017] S22. Input the primary spatial feature map into the bidirectional long short-term memory network layer of the multi-layer feature extraction network to extract time-dependent features and obtain the spatiotemporal coupled feature sequence.

[0018] S23. Input the spatiotemporal coupled feature sequence into the self-attention network layer of the multi-layer feature extraction network to perform cross-modal correlation analysis and calculate the correlation weight matrix between features;

[0019] S24. Based on the correlation weight matrix, the spatiotemporal coupling feature sequences are weighted and fused to generate equipment status feature data.

[0020] In one embodiment, S3 includes:

[0021] S31. Input the equipment status characteristic data into the trend prediction branch to perform rolling time window prediction and obtain the equipment performance trend data for the next N time steps.

[0022] S32. Input the equipment status feature data into the anomaly detection branch, calculate the deviation of the equipment status feature data from the historical normal data distribution, and obtain the real-time anomaly score.

[0023] S33. Combine and merge the equipment performance trend data and real-time anomaly scores to generate equipment performance prediction data. When the real-time anomaly score exceeds a preset threshold, mark the anomaly point in the equipment performance trend data.

[0024] In one embodiment, S4 includes:

[0025] S41. Establish a multi-objective optimization function, which includes the objectives of maximizing equipment energy efficiency, optimizing work quality, and minimizing resource consumption.

[0026] The multi-objective optimization function is expressed as follows:

[0027]

[0028]

[0029]

[0030] In the formula, To maximize the energy efficiency of the equipment, For the effective workload of the equipment, For input power, To optimize the objective function value for job quality, For the first Item of work quality indicators, For the first The corresponding weights of each task quality indicator To minimize the value of the objective function for resource consumption, For the first Consumption of various resources For the first The unit cost of resource consumption;

[0031] S42. Iteratively solve the multi-objective optimization function to obtain the Pareto solution set;

[0032] S43. Perform parameter sensitivity evaluation on the Pareto solution set and obtain the parameter sensitivity evaluation results;

[0033] S44. Based on the parameter sensitivity evaluation results, select the optimal solution from the Pareto solution set and generate parameter optimization scheme data.

[0034] In one embodiment, the method further includes:

[0035] S51. Generate control instructions from the parameter optimization scheme data to obtain the equipment control instruction sequence;

[0036] S52. Based on the equipment control command sequence, real-time parameter adjustment is performed through the PID controller to obtain real-time control output;

[0037] S53. Monitor the performance indicators of the real-time control output and obtain feedback evaluation results;

[0038] S54. Optimize the strategy based on the feedback evaluation results to obtain equipment optimization data.

[0039] In one embodiment, S54 includes:

[0040] S61. Construct the state space from the feedback evaluation results to obtain the environmental state representation;

[0041] S62. Based on the environmental state representation, obtain the training sample set;

[0042] S63. Train a Q-value network on the training sample set to obtain the optimal action value function; wherein, the loss function expression of the Q-value network is:

[0043]

[0044] In the formula, For loss function, For the current network parameters, For the target network parameters, As a discount factor, For experience replay buffer, For empirical quadruplets, This is the current environment state vector. In the state The action to be performed To perform the action The instant reward value obtained afterwards The next state to transition to after performing an action. In the state Actions in the set of all possible actions. For the current Q-network state-action pairs The value estimate, For the target Q-network, state-action pairs The value estimate, This indicates the buffer from the experience replay. Calculate the expectation of empirical quadruples from random sampling.

[0045] S64. Select actions based on the optimal action value function to obtain equipment optimization data.

[0046] Secondly, this application also provides an intelligent improvement system for agricultural equipment, including:

[0047] The multi-source data acquisition module is used to acquire multi-source heterogeneous data of the target planting area, and to perform data cleaning and spatiotemporal alignment on the multi-source heterogeneous data to obtain aligned multi-source heterogeneous data.

[0048] The feature association analysis module is used to input aligned multi-source heterogeneous data into a multi-layer feature extraction network, perform hierarchical feature learning and association analysis, and obtain equipment status feature data; the equipment status feature data is used to characterize the spatiotemporal association features of equipment operating status;

[0049] The trend prediction and anomaly detection module is used to input equipment status feature data into the adaptive prediction model to perform trend prediction and anomaly detection, and obtain equipment performance prediction data. The adaptive prediction model includes a trend prediction branch and an anomaly detection branch. The trend prediction branch is built based on a gated recurrent unit network, and the anomaly detection branch is built based on a support vector data description algorithm.

[0050] The parameter optimization scheme generation module is used to perform multi-objective optimization analysis and parameter sensitivity assessment on equipment performance prediction data to obtain parameter optimization scheme data.

[0051] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0052] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0053] The aforementioned intelligent improvement method, system, equipment, and medium for planting equipment comprehensively capture the complex correlation between crop growth environment and equipment operating status by cleaning and spatiotemporally aligning multi-source heterogeneous data; it achieves accurate prediction and early warning of equipment performance by utilizing multi-layer feature extraction and adaptive prediction models, and can flexibly respond to dynamic environmental changes; through multi-objective optimization and parameter sensitivity assessment, it achieves a dynamic balance between equipment energy efficiency, operation quality, and resource consumption, and forms a closed-loop improvement mechanism by combining feedback iterative optimization, thereby enhancing the foresight and systematic nature of equipment improvement, effectively improving the intelligence level and comprehensive operating efficiency of planting equipment, and adapting to the actual needs of intelligent improvement of planting equipment. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart illustrating a method for intelligent improvement of agricultural equipment in one embodiment.

[0056] Figure 2 This is a schematic diagram of the structure of an intelligent improvement system for agricultural equipment in one embodiment. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0058] In one embodiment, reference Figure 1 The document presents a flowchart illustrating a method for intelligent improvement of agricultural equipment provided in this application. This embodiment uses the application of this method to an agricultural equipment improvement terminal (hereinafter referred to as the terminal) as an example. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0059] S1. Obtain multi-source heterogeneous data of the target planting area, and perform data cleaning and spatiotemporal alignment on the multi-source heterogeneous data to obtain aligned multi-source heterogeneous data.

[0060] The multi-source heterogeneous data acquisition sources cover various data related to the operation of planting equipment and crop growth within the target planting area, including but not limited to meteorological data, soil data, operating data of planting equipment, and crop growth status data. Meteorological data includes light, precipitation, and temperature; soil data includes soil texture, nutrient content, and humidity; operating data of planting equipment includes the start / stop status of irrigation equipment, the discharge rate of fertilizer equipment, and the operating trajectory of plant protection equipment; and crop growth status data includes plant height and leaf growth.

[0061] For example, the planting equipment improvement terminal establishes communication connections with various data acquisition devices to collect multi-source heterogeneous data in real time. Simultaneously, it can obtain historical data stored on third-party platforms through interface calls, achieving comprehensive collection of multi-dimensional data. Data cleaning removes invalid information from the original data to ensure accuracy. Missing value imputation uses interpolation methods adapted to agricultural data characteristics, supplementing missing information based on the temporal or spatial correlation of the data. Noise filtering removes redundant noise caused by equipment or environmental interference during data acquisition. Outlier removal uses statistical analysis methods to identify and remove extreme values ​​deviating from the normal data distribution, resulting in cleaned multi-source heterogeneous data. During spatiotemporal alignment, the planting equipment improvement terminal presets a unified spatiotemporal reference coordinate system, uniformly transforming the cleaned multi-source heterogeneous data to this coordinate system. Through gridding, spatiotemporal synchronization of data from different sources and dimensions is achieved, eliminating deviations in time scale and spatial location, resulting in spatiotemporally consistent aligned multi-source heterogeneous data.

[0062] S2. Align the multi-source heterogeneous data and input it into the multi-layer feature extraction network to perform hierarchical feature learning and correlation analysis to obtain equipment status feature data.

[0063] Among them, the multi-layer feature extraction network is a feature extraction model built on deep learning algorithms. It consists of a convolutional neural network layer, a bidirectional long short-term memory network layer, and a self-attention network layer. Each layer works together to achieve hierarchical feature learning and cross-modal correlation analysis.

[0064] The improved planting equipment terminal aligns multi-source heterogeneous data into a multi-layer feature extraction network, then extracts spatial features through a convolutional neural network layer. Utilizing the sliding operation of the convolutional kernel, it captures local spatial features in the data and outputs a primary spatial feature map. These local spatial features include spatial distribution differences of equipment operating parameters in different regions, and spatial correlation features between soil conditions and equipment operating status.

[0065] The improved planting equipment terminal inputs the primary spatial feature map into a bidirectional long short-term memory network layer. This network layer can simultaneously capture the positive and negative time dependencies of the data, effectively mining the patterns of equipment operation status changes over time, as well as the temporal correlation features between crop growth environment changes and equipment operation status. It integrates spatial and temporal features to output a spatiotemporal coupled feature sequence.

[0066] The planting equipment improvement terminal inputs the spatiotemporal coupled feature sequence into the self-attention network layer. This network layer calculates the correlation weights between different features to achieve cross-modal and cross-spatiotemporal correlation analysis. It focuses on mining the intrinsic correlation between equipment operation data and meteorological data, soil data, and crop growth data to obtain the correlation weight matrix between features. The planting equipment improvement terminal performs weighted fusion of the spatiotemporal coupled feature sequence according to the weight matrix to generate equipment status feature data for comprehensively characterizing the spatiotemporal correlation features of equipment operation status.

[0067] S3. Input the equipment status characteristic data into the adaptive prediction model to perform trend prediction and anomaly detection, and obtain equipment performance prediction data.

[0068] The adaptive prediction model employs a dual-branch structure of trend prediction and anomaly detection, enabling simultaneous prediction of equipment performance trends and anomaly detection. This enhances the comprehensiveness and accuracy of predictions and adapts to the dynamic changes in equipment operating status during agricultural production. The trend prediction branch is built upon a gated recurrent unit network (GRN), an improved recurrent neural network that effectively captures the temporal variation patterns of equipment status characteristic data through the synergistic effect of reset and update gates. This allows for rapid adaptation to dynamic data changes, avoids gradient problems in long-sequence data processing, and is suitable for long-term trend prediction of equipment performance.

[0069] The anomaly detection branch is built based on the Support Vector Data Description (SVM) algorithm. This algorithm learns the distribution patterns of historical normal equipment status feature data to construct feature boundaries for normal data. When new equipment status feature data is input, its deviation from these feature boundaries is calculated to determine whether there are any anomalies in equipment operation. The planting equipment improvement terminal simultaneously inputs the equipment status feature data into both branches. The trend prediction branch outputs the future trend of equipment performance changes, while the anomaly detection branch outputs the real-time anomalies in equipment operation. The planting equipment improvement terminal fuses the outputs of the two branches to obtain equipment performance prediction data.

[0070] S4. Perform multi-objective optimization analysis and parameter sensitivity assessment on the equipment performance prediction data to obtain parameter optimization scheme data.

[0071] For example, the planting equipment improvement terminal performs multi-objective optimization analysis and parameter sensitivity assessment on equipment performance prediction data to obtain parameter optimization scheme data. The multi-objective optimization analysis seeks a dynamic balance and overall optimality among three mutually constraining objectives: equipment energy efficiency, operational quality, and resource consumption. This overcomes the limitations of traditional single-objective optimization and adapts to the diversified needs of precision agriculture. The parameter sensitivity assessment aims to analyze the impact of changes in various operating parameters of the planting equipment on equipment performance, identify key parameters that significantly affect equipment performance, and provide a basis for formulating parameter optimization schemes.

[0072] The improved planting equipment terminal first establishes a multi-objective optimization function that includes maximizing equipment energy efficiency, optimizing operation quality, and minimizing resource consumption. It clarifies the calculation method and constraints of each optimization objective. Then, iteratively solves the multi-objective optimization function through a suitable optimization algorithm to obtain a Pareto solution set that can achieve a balance of multiple objectives. Each solution in the Pareto solution set corresponds to a set of equipment operating parameters, and no single solution can be better than other solutions in all objectives at the same time.

[0073] The planting equipment improvement terminal performs parameter sensitivity assessment on each parameter combination in the Pareto solution set, analyzes the impact of parameter changes on equipment energy efficiency, operation quality, and resource consumption, screens out parameter combinations with low sensitivity, good stability, and the ability to achieve multi-objective optimization, and generates parameter optimization scheme data to guide the adjustment of operating parameters of planting equipment.

[0074] In the aforementioned intelligent improvement method for agricultural equipment, multi-source heterogeneous data is cleaned and spatiotemporally aligned to comprehensively capture the complex relationship between crop growth environment and equipment operating status, thus overcoming the limitations of a single data source. Through multi-layer feature extraction and adaptive prediction models, accurate characterization, trend prediction, and anomaly detection of equipment operating status are achieved, eliminating reliance on human experience and static rules and enabling adaptive responses to dynamic changes in agricultural scenarios. Through multi-objective optimization and feedback iterative optimization, the overall optimality of equipment energy efficiency, operational quality, and resource consumption is achieved, improving the comprehensive operational efficiency of the equipment. This method does not rely on complex operations, adapts to the needs of precision agriculture, and effectively enhances the intelligence level of agricultural equipment.

[0075] In an optional embodiment, S1 includes:

[0076] S11. Obtain the original multi-source heterogeneous data of the target planting area.

[0077] Optionally, the improved planting equipment terminal establishes bidirectional communication with various sensors, controllers of planting equipment, and crop growth monitoring equipment deployed within the target planting area through an integrated communication module, collecting raw data in real time. Sensors include, but are not limited to, soil sensors and meteorological sensors. Simultaneously, the terminal accesses an agricultural big data platform via a network interface to retrieve historical multi-source data from the target planting area, including historical meteorological data, historical soil data, and historical equipment operation data. This combines real-time raw data acquisition with historical data retrieval, ensuring the comprehensiveness and continuity of the raw data. The raw multi-source heterogeneous data encompasses both structured and unstructured data. Structured data includes equipment operating parameters and real-time sensor readings, while unstructured data includes crop growth images and equipment operation status videos.

[0078] S12. Perform missing value imputation, noise filtering, and outlier removal on the original multi-source heterogeneous data to obtain cleaned multi-source heterogeneous data.

[0079] Optionally, during the missing value imputation process, the planting equipment improvement terminal selects an appropriate imputation method based on the type of the original data. For data with temporal continuity, temporal interpolation is used to imput missing values, ensuring the temporal correlation of the data. For spatially distributed data, spatial interpolation is used to imput missing information based on the distribution characteristics of surrounding valid data, avoiding data bias caused by a single imputation method. During the noise filtering process, corresponding filtering algorithms are selected for different types of data based on their noise characteristics to filter out random noise and system noise generated during data acquisition, retaining the core features of the data and ensuring data stability. During the outlier removal process, the planting equipment improvement terminal determines the normal distribution range of the data through statistical analysis, identifies outlier data that exceeds this range, and, combined with the actual agricultural production scenario, distinguishes between outlier data and reasonable extreme data. Data confirmed as invalid outliers is removed to avoid interference from outlier data with subsequent analysis results, resulting in cleaned multi-source heterogeneous data.

[0080] S13. After cleaning, the multi-source heterogeneous data is uniformly converted to the preset spatiotemporal reference coordinate system and then meshed and aligned to obtain aligned multi-source heterogeneous data.

[0081] Optionally, the planting equipment improvement terminal uses a built-in preset spatiotemporal reference coordinate system to set a spatial coordinate benchmark based on the geographical location and planting layout of the target planting area, and sets a time coordinate benchmark at fixed time intervals to ensure the uniformity and rationality of the spatiotemporal reference. The terminal performs coordinate transformation on the cleaned multi-source heterogeneous data, mapping the spatiotemporal coordinates corresponding to different acquisition devices and data types to a unified preset coordinate system, eliminating spatiotemporal deviations between different devices. During the grid alignment process, the terminal divides the target planting area into uniform grids according to preset rules, distributes the transformed multi-source data to each grid cell, and simultaneously performs time synchronization calibration on the multi-dimensional data within the same grid cell, ensuring that multi-source data under the same grid and time node can match each other, achieving spatiotemporal alignment of the data and obtaining aligned multi-source heterogeneous data.

[0082] In an optional embodiment, S2 includes:

[0083] S21. Align the multi-source heterogeneous data and input it into the convolutional neural network layer of the multi-layer feature extraction network to extract local spatial features and obtain the primary spatial feature map.

[0084] Optionally, the improved planting equipment terminal aligns the heterogeneous data from multiple sources and inputs it into a convolutional neural network layer of a multi-layer feature extraction network. This layer consists of multiple convolutional kernels, each corresponding to a different spatial feature extraction dimension. Through sliding convolution operations on the data matrix, the terminal captures local spatial features in the data, including the distribution differences of equipment operating parameters at different spatial locations, the local correlation between soil nutrient distribution and irrigation equipment operating status, and the spatial heterogeneity of meteorological conditions within the target area. During the convolution operation, the improved planting equipment terminal performs non-linear transformation on the convolution results using an activation function to enhance the expressive power of the features. Simultaneously, it employs pooling operations to reduce the dimensionality of the convolutional features, preserving core spatial features and reducing computational load. After processing by the convolutional neural network layer, the local spatial features from the aligned heterogeneous data are extracted and integrated, outputting a primary spatial feature map reflecting the spatial distribution characteristics of the data.

[0085] S22. Input the primary spatial feature map into the bidirectional long short-term memory network layer of the multi-layer feature extraction network to extract time-dependent features and obtain the spatiotemporal coupled feature sequence.

[0086] Optionally, the improved planting equipment terminal inputs the primary spatial feature map into the bidirectional long short-term memory (LSTM) layer of the multi-layer feature extraction network to extract time-dependent features, thus obtaining a spatiotemporally coupled feature sequence. The bidirectional LSTM layer, by introducing forward and backward memory units, can simultaneously traverse the time series of the primary spatial feature map in both forward and backward directions, capturing the feature change patterns at different time points and the time dependencies between features.

[0087] For example, the temporal characteristics of irrigation equipment operating parameters changing with weather conditions, and the temporal correlation between fertilizer application equipment discharge rate and crop growth status are captured, allowing for deep fusion of spatial and temporal features. A bidirectional long short-term memory (LSTM) network layer uses a gating mechanism to selectively remember and forget temporal features, effectively solving the gradient vanishing or exploding problem in long-sequence data processing, and accurately mining the long-term temporal dependence features of equipment operating status. After processing by the bidirectional LSM network layer, the primary spatial feature map and temporal dependence features are integrated to generate a spatiotemporally coupled feature sequence that combines spatial and temporal features, achieving preliminary fusion of the spatiotemporal features of equipment operating status.

[0088] S23. Input the spatiotemporal coupled feature sequence into the self-attention network layer of the multi-layer feature extraction network to perform cross-modal correlation analysis and calculate the correlation weight matrix between features.

[0089] Optionally, the spatiotemporal coupled feature sequence includes multimodal spatiotemporal features such as equipment operation, weather, soil, and crop growth. Complex intrinsic relationships exist between these modal features. A self-attention network layer is used to mine these cross-modal and cross-spatiotemporal relationships. The planting equipment improvement terminal uses a self-attention mechanism to calculate the degree of correlation between each feature in the spatiotemporal coupled feature sequence and all other features, quantifying the correlation degree into correlation weights to generate a correlation weight matrix between features. During the calculation of correlation weights, the planting equipment improvement terminal performs linear transformations and dot product operations on the features to highlight features closely related to the equipment's operating status, such as the correlation between soil moisture and irrigation equipment operating parameters, and the correlation between meteorological precipitation and the start / stop status of irrigation equipment. It reduces the weights of features with weaker correlation to the equipment's operating status, achieving accurate identification of key correlated features and obtaining the correlation weight matrix between features.

[0090] S24. Based on the correlation weight matrix, the spatiotemporal coupling feature sequences are weighted and fused to generate equipment status feature data.

[0091] Optionally, the improved planting equipment terminal assigns a corresponding correlation weight to each feature component in the spatiotemporal coupled feature sequence based on the correlation weight matrix. The feature component with the higher the correlation weight has a greater proportion in the fusion process, thereby highlighting the role of key correlation features. During the weighted fusion process, the improved planting equipment terminal uses a linear fusion algorithm to perform corresponding operations on the spatiotemporal coupled feature sequence and the correlation weight matrix, integrating multimodal and cross-spatiotemporal key features, eliminating redundant and interfering features, and achieving optimized feature integration. The generated equipment status feature data is used to characterize the spatiotemporal correlation features of the equipment operating status, reflecting the intrinsic relationship between the equipment operating status and the crop growth environment and resource consumption.

[0092] In an optional embodiment, S3 includes:

[0093] S31. Input the equipment status characteristic data into the trend prediction branch to perform rolling time window prediction and obtain the equipment performance trend data for the next N time steps.

[0094] Optionally, after the gated recurrent unit network of the trend prediction branch is trained with historical equipment state feature data, it possesses time-series prediction capabilities. The rolling time window prediction method refers to using the current equipment state feature data as input to predict the equipment performance data for the next time step. This prediction result is then combined with the actual equipment state feature data for the next time step to continue predicting the performance data for that next time step, and so on, to achieve continuous prediction for the next N time steps. During the prediction process, the planting equipment improvement terminal selectively forgets historical feature information irrelevant to future performance prediction through the reset gate of the gated recurrent unit network, and retains key time-series features through the update gate, ensuring the accuracy of the prediction results. The equipment performance trend data for the next N time steps includes multiple dimensions such as equipment operating efficiency, resource consumption, and work quality, reflecting the future changes in equipment performance.

[0095] S32. Input the equipment status feature data into the anomaly detection branch, calculate the deviation of the equipment status feature data from the historical normal data distribution, and obtain the real-time anomaly score.

[0096] Optionally, the support vector data description algorithm of the anomaly detection branch constructs a minimum feature boundary that can enclose all normal data by learning historical normal equipment state feature data. This boundary can accurately characterize the feature distribution pattern of normal equipment operating status. After the planting equipment improvement terminal inputs the equipment state feature data into the anomaly detection branch, it calculates the distance from the input data to the feature boundary. This distance is the deviation of the input data from the historical normal data distribution. The greater the deviation, the more the equipment operating status deviates from the normal range. The planting equipment improvement terminal quantifies the deviation into a real-time anomaly score, which reflects the severity of the equipment operating anomaly.

[0097] S33. Combine and merge the equipment performance trend data and real-time anomaly scores to generate equipment performance prediction data. When the real-time anomaly score exceeds a preset threshold, mark the anomaly point in the equipment performance trend data.

[0098] Optionally, the planting equipment improvement terminal splices and merges equipment performance trend data and real-time anomaly scores. During the splicing and fusion process, the terminal correlates the equipment performance trend data and real-time anomaly scores along a time dimension, ensuring that each time step of the equipment performance trend data corresponds to a real-time anomaly score, thus generating equipment performance prediction data. A preset threshold is pre-set by the planting equipment improvement terminal based on actual agricultural production scenarios, equipment operating requirements, and historical anomaly data statistics. This threshold is used to determine the severity of equipment operational anomalies. When the real-time anomaly score exceeds this threshold, it indicates a significant equipment operational anomaly that may affect operational quality and resource consumption. The terminal then marks the anomaly point at the corresponding time step in the equipment performance trend data, clearly indicating the time and severity of the anomaly.

[0099] In an optional embodiment, S4 includes:

[0100] S41. Establish a multi-objective optimization function, which includes the objectives of maximizing equipment energy efficiency, optimizing work quality, and minimizing resource consumption.

[0101] The multi-objective optimization function is expressed as follows:

[0102]

[0103]

[0104]

[0105] In the formula, To maximize the energy efficiency of the equipment, For the effective workload of the equipment, For input power, To optimize the objective function value for job quality, For the first Item of work quality indicators, For the first The corresponding weights of each task quality indicator To minimize the value of the objective function for resource consumption, For the first Consumption of various resources For the first The unit cost of consuming a certain resource.

[0106] The objective optimization function includes the objectives of maximizing equipment energy efficiency, optimizing operational quality, and minimizing resource consumption. The improved planting equipment terminal clarifies the objectives and constraints for equipment parameter optimization by establishing a multi-objective optimization function.

[0107] In the above multi-objective optimization function expression, To maximize the energy efficiency of the equipment, The effective workload of the equipment refers to the total amount of work that meets the requirements of agricultural production and is completed by the equipment per unit of time. Input power refers to the total power consumed by the equipment during operation. The higher the value, the higher the energy utilization efficiency of the equipment; To optimize the objective function value for job quality, For the first The operational quality indicators cover metrics related to planting operation quality, such as irrigation uniformity, fertilization accuracy, and plant protection coverage. For the first Each operational quality indicator has a corresponding weight, which is set based on agricultural production needs and crop growth characteristics to reflect the importance of different operational quality indicators. The higher the value, the better the equipment operation quality; To minimize the value of the objective function for resource consumption, For the first Resource consumption includes various resources consumed during the planting process, such as water, fertilizer, and electricity. For the first The unit cost of resource consumption is used to quantify the cost of consuming different resources. The lower the value, the less resources are consumed and the lower the cost during equipment operation.

[0108] S42. Iteratively solve the multi-objective optimization function to obtain the Pareto solution set.

[0109] Optionally, the improved planting equipment terminal iteratively solves the multi-objective optimization function to obtain a Pareto solution set. During the iterative solution process, the terminal selects an optimization algorithm adapted to the multi-objective optimization problem. Based on equipment performance prediction data, it repeatedly calculates the multi-objective optimization function, continuously adjusting the combination of equipment operating parameters to seek a parameter combination that simultaneously maximizes equipment energy efficiency, optimizes operational quality, and minimizes resource consumption. During the iteration process, each parameter adjustment calculates the corresponding three objective function values. By comparing the objective function values ​​of different parameter combinations, non-dominated solutions are selected. The set of all non-dominated solutions constitutes the Pareto solution set. Each solution in the Pareto solution set corresponds to a unique combination of equipment operating parameters, each emphasizing different optimization objectives, thus meeting the needs of different agricultural production scenarios.

[0110] S43. Perform parameter sensitivity evaluation on the Pareto solution set and obtain the parameter sensitivity evaluation results.

[0111] Optionally, the planting equipment improvement terminal performs parameter sensitivity assessment on the Pareto solution set. During this assessment, the terminal uses the controlled variable method, fixing all other parameters in a specific parameter combination within the Pareto solution set and changing only the value of one parameter. The impact of this parameter's value change on the three objective function values ​​is observed, and the correlation between the parameter change and the objective function value change is calculated to quantify the parameter's sensitivity. The terminal then performs parameter sensitivity assessment on each operating parameter for all parameter combinations within the Pareto solution set, identifying parameters with high and low sensitivity to the objective function values. Simultaneously, it analyzes the impact of interactions between different parameters on equipment performance, generating parameter sensitivity assessment results that include the sensitivity levels of each parameter and the patterns of parameter interactions.

[0112] S44. Based on the parameter sensitivity evaluation results, select the optimal solution from the Pareto solution set and generate parameter optimization scheme data.

[0113] Optionally, the planting equipment improvement terminal selects the optimal solution from the Pareto solution set based on the parameter sensitivity assessment results. During this selection process, the terminal prioritizes parameter combinations with low sensitivity and high stability of key parameters, ensuring that these combinations effectively mitigate the significant impact of minor parameter changes on equipment performance and improve operational stability. Simultaneously, considering the actual needs of agricultural production and balancing equipment energy efficiency, operational quality, and resource consumption, the terminal selects the parameter combination that best meets production requirements from the Pareto solution set as the optimal solution. The parameter optimization scheme data, centered on the optimal solution, includes the specific adjustment direction and range for each operating parameter, as well as the expected equipment performance indicators after parameter adjustment. This generates a complete and executable parameter optimization scheme to guide the intelligent improvement and parameter adjustment of planting equipment.

[0114] In an optional embodiment, the method further includes:

[0115] S51. Generate control instructions from the parameter optimization scheme data to obtain the equipment control instruction sequence.

[0116] Optionally, the planting equipment improvement terminal uses a preset instruction generation algorithm to convert parameter optimization scheme data into control instructions that the equipment can recognize and execute. These control instructions cover various functions such as equipment start / stop control, operating parameter adjustment, and operating mode switching. Based on the adjustment sequence and requirements of each parameter in the parameter optimization scheme, the planting equipment improvement terminal arranges the control instructions in chronological order, generating a sequence of equipment control instructions. This ensures that the execution order of the control instructions is consistent with the requirements of the parameter optimization scheme, achieving orderly adjustment of equipment parameters. Operating parameter adjustments include irrigation flow, fertilizer application, and plant protection pressure.

[0117] S52. Based on the equipment control command sequence, real-time parameter adjustment is performed through the PID controller to obtain real-time control output.

[0118] Optionally, a PID controller is a commonly used closed-loop control device, characterized by high control accuracy, fast response speed, and good stability, making it suitable for the real-time parameter adjustment needs of planting equipment. The planting equipment improvement terminal uses each control command in the equipment control command sequence as the setpoint for the PID controller, while simultaneously collecting real-time operating parameters of the planting equipment as feedback values. The PID controller calculates the deviation between the setpoint and the feedback value, and, combining proportional, integral, and derivative control links, adjusts the deviation, outputting a real-time control signal to control the actuators of the equipment, thus achieving real-time adjustment of the equipment's operating parameters. The real-time control output is the control signal output by the PID controller, which can dynamically adjust according to the real-time operating status of the equipment, ensuring that the equipment's operating parameters quickly approach the setpoints in the parameter optimization scheme, improving the accuracy and stability of parameter adjustment. Real-time operating parameters include actual irrigation flow rate, actual fertilizer application, etc., and the actuators controlling the equipment include water pumps, valves, and sprinklers.

[0119] S53. Monitor the performance indicators of the real-time control output and obtain feedback evaluation results.

[0120] Optionally, during the execution of real-time control output, the planting equipment improvement terminal establishes real-time communication with the sensors and controllers of the planting equipment to continuously collect real-time operating performance indicators of the equipment, including indicators from multiple dimensions such as equipment energy efficiency, operation quality, and resource consumption. Simultaneously, it collects real-time data on crop growth status and environmental data as auxiliary data for feedback evaluation. The planting equipment improvement terminal compares and analyzes the collected real-time performance indicators with the preset expected performance indicators in the parameter optimization scheme to evaluate the execution effect of the real-time control output, analyze the improvement in equipment performance after parameter adjustment, whether the optimization target has been achieved, and identify potential problems during parameter adjustment. It integrates the evaluation results, comparative analysis data, and existing problems to generate feedback evaluation results. Potential problems include unstable equipment operation due to excessively rapid parameter adjustment and substandard operation quality due to parameter adjustment deviations.

[0121] S54. Optimize the strategy based on the feedback evaluation results to obtain equipment optimization data.

[0122] Optionally, the feedback evaluation results are used to reflect the execution effect and shortcomings of the parameter optimization scheme. The strategy optimization process is used to adjust and optimize the parameter optimization scheme, control command generation logic, PID control parameters, etc., to address these shortcomings and improve the equipment improvement effect. The planting equipment improvement terminal, combined with the feedback evaluation results, analyzes the unreasonable aspects of the parameter optimization scheme, and analyzes the causes of deviations in the control command execution process and the control effect of the PID controller. Based on these analysis results, the relevant strategies are iteratively optimized to correct the shortcomings of the parameter optimization scheme, optimize the control command generation algorithm, and adjust the control parameters of the PID controller, obtaining optimized equipment optimization data. This optimized data includes the optimized parameter adjustment scheme, control command sequence, PID control parameters, etc., and is used to guide the subsequent improvement and operation control of planting equipment, achieving continuous improvement of equipment performance. The unreasonable aspects include unreasonable parameter adjustment ranges and improper setting of key parameter weights.

[0123] In an optional embodiment, S54 includes:

[0124] S61. Construct the state space from the feedback evaluation results to obtain the environmental state representation.

[0125] Optionally, state space construction is used to standardize various data in the feedback evaluation results, converting them into state vectors that can characterize the current equipment operating environment and optimization execution status. The dimensions of the state space are determined by the data type and quantity of the feedback evaluation results, with each dimension corresponding to an evaluation index. The improved planting equipment terminal uses feature encoding methods to convert unstructured feedback evaluation data into structured feature vectors. Simultaneously, it normalizes the structured data to eliminate dimensional differences between different indices, ensuring the consistency and comparability of the state vectors, thus obtaining an environmental state representation that reflects the overall operating status of the current equipment, the execution effect of the parameter optimization scheme, and existing problems.

[0126] S62. Based on the environmental state representation, the training sample set is obtained.

[0127] Optionally, the training sample set is used to train the Q-value network, enabling it to optimize policies. The planting equipment improvement terminal uses the environmental state representation as its core, combining parameter adjustment actions, reward values ​​after action execution, and the next environmental state after action execution to construct each sample in the training sample set. Specifically, parameter adjustment actions refer to various parameter adjustment measures the terminal may take to address problems identified in the feedback evaluation results; reward values ​​are set based on the equipment performance improvement after the parameter adjustment actions and whether the optimization goal has been achieved—the more significant the performance improvement and the closer it is to the optimization goal, the higher the reward value; the next environmental state is the environmental state representation obtained after the terminal re-collects the equipment operating status and feedback evaluation data after the parameter adjustment actions are executed, and constructs the state space. The planting equipment improvement terminal generates the training sample set by accumulating multiple sets of environmental state representations, parameter adjustment actions, reward values, and next environmental state data to ensure the training effectiveness of the Q-value network.

[0128] S63. Train a Q-value network on the training sample set to obtain the optimal action value function; wherein, the loss function expression of the Q-value network is:

[0129]

[0130] In the formula, For loss function, For the current network parameters, For the target network parameters, As a discount factor, For experience replay buffer, For empirical quadruplets, This is the current environment state vector. In the state The action to be performed To perform the action The instant reward value obtained afterwards The next state to transition to after performing an action. In the state Actions in the set of all possible actions. For the current Q-network state-action pairs The value estimate, For the target Q-network, state-action pairs The value estimate, This indicates the buffer from the experience replay. Calculate the expectation of empirical quadruples randomly sampled from the data.

[0131] In the loss function expression of the Q-value network mentioned above, The loss function measures the deviation between the current Q-network's prediction and the target value. The smaller the deviation, the higher the prediction accuracy of the Q-network. These are the current network parameters, i.e., the weight parameters of the current Q-value network; These are the target network parameters, i.e., the weight parameters of the target Q-value network. The target Q-network has the same structure as the current Q-network and is used to provide stable target values ​​to avoid oscillations during training. This is a discount factor used to adjust the degree to which future reward values ​​influence the current action choice; its value ranges from 0 to 1. The closer it is to 1, the greater the impact of future reward values; It serves as an experience replay buffer, used to store experience quadruples in the training sample set, to avoid training bias caused by the correlation between samples during training. For empirical quadruplets, This is the current environment state vector, that is, the vector representing the environment state. In the state The action to be performed is the parameter adjustment action; To perform the action The instant reward value obtained afterwards The vector represents the next state after an action is performed, i.e., the next environment state. In the state Actions in the set of all possible actions. For the current Q-network state-action pairs The value estimate, that is, the prediction in state Next action The long-term reward value that can be obtained later; For the target Q-network, state-action pairs The value estimate, This indicates the buffer from the experience replay. Calculate the expectation of empirical quadruples randomly sampled from the data.

[0132] Optionally, the improved terminal of the planting equipment trains a Q-value network on the training sample set. During the training process, the terminal randomly samples empirical quadruplets from the empirical replay buffer, substitutes them into the loss function to calculate the loss value, updates the parameters of the current Q-network through the gradient descent algorithm, continuously reduces the loss value, and periodically copies the parameters of the current Q-network to the target Q-network to update the target network parameters, ensuring the stability of the target value. After multiple rounds of iterative training, when the loss value converges to a preset range, the training ends and the optimal action value function is obtained.

[0133] S64. Select actions based on the optimal action value function to obtain equipment optimization data.

[0134] Optionally, the optimal action value function can accurately predict the long-term reward value obtained by performing various parameter adjustment actions under different environmental conditions. During the action selection process, the planting equipment improvement terminal calculates the value estimate corresponding to all possible parameter adjustment actions based on the current environmental state representation through the optimal action value function, and selects the action with the highest value estimate as the optimal parameter adjustment action. This action can maximize the improvement of equipment performance and is closest to the multi-objective optimization goal. The planting equipment improvement terminal integrates the parameter adjustment scheme, control command logic, PID control parameters, etc., corresponding to the optimal parameter adjustment action, and generates equipment optimization data by combining the current environmental state and optimization goal. The equipment optimization data can specifically solve the problems existing in the feedback evaluation results, realize the further optimization of parameter optimization strategy and control strategy, provide scientific and accurate guidance for the continuous improvement of planting supporting equipment, and promote the continuous improvement of equipment operating efficiency.

[0135] In the aforementioned intelligent improvement method for supporting equipment in planting, the comprehensive collection, cleaning, and spatiotemporal alignment of multi-source heterogeneous data compensates for the shortcomings of traditional data dimensions, accurately reflecting the complex relationship between equipment operation and crop growth environment. The combination of multi-layer feature extraction and adaptive prediction models enables prediction of equipment performance trends and early detection of anomalies, breaking free from dependence on static rules and flexibly responding to dynamic environmental changes. Multi-objective optimization and parameter sensitivity assessment balance equipment energy efficiency, operational quality, and resource consumption, overcoming the limitations of single-objective optimization. A subsequent feedback optimization mechanism enables continuous strategy iteration, enhancing adaptive capabilities, improving the intelligence and overall operational efficiency of planting equipment, reducing passive maintenance, and raising the level of intelligent planting.

[0136] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0137] Based on the same inventive concept, this application also provides an intelligent improvement system for planting equipment to implement the above-mentioned intelligent improvement method for planting equipment. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the intelligent improvement system for planting equipment provided below can be found in the limitations of the intelligent improvement method for planting equipment described above, and will not be repeated here.

[0138] In one exemplary embodiment, such as Figure 2 As shown, a schematic diagram of a smart improvement system 10 for agricultural equipment is provided, comprising:

[0139] The multi-source data acquisition module 11 is used to acquire multi-source heterogeneous data of the target planting area, and to perform data cleaning and spatiotemporal alignment on the multi-source heterogeneous data to obtain aligned multi-source heterogeneous data.

[0140] The feature association analysis module 12 is used to input aligned multi-source heterogeneous data into a multi-layer feature extraction network, perform hierarchical feature learning and association analysis, and obtain equipment status feature data; the equipment status feature data is used to characterize the spatiotemporal association features of the equipment operating status;

[0141] The trend prediction and anomaly detection module 13 is used to input equipment status feature data into the adaptive prediction model to perform trend prediction and anomaly detection, and obtain equipment performance prediction data. The adaptive prediction model includes a trend prediction branch and an anomaly detection branch. The trend prediction branch is constructed based on a gated recurrent unit network, and the anomaly detection branch is constructed based on a support vector data description algorithm.

[0142] The parameter optimization scheme generation module 14 is used to perform multi-objective optimization analysis and parameter sensitivity assessment on equipment performance prediction data to obtain parameter optimization scheme data.

[0143] Furthermore, the multi-source data acquisition module 11 is also used for:

[0144] S11. Obtain the original multi-source heterogeneous data of the target planting area;

[0145] S12. Perform missing value imputation, noise filtering and outlier removal on the original multi-source heterogeneous data to obtain cleaned multi-source heterogeneous data.

[0146] S13. After cleaning, the multi-source heterogeneous data is uniformly converted to the preset spatiotemporal reference coordinate system and then meshed and aligned to obtain aligned multi-source heterogeneous data.

[0147] Furthermore, Feature Association Analysis Module 12 is also used for:

[0148] S21. Align the multi-source heterogeneous data and input it into the convolutional neural network layer of the multi-layer feature extraction network to extract local spatial features and obtain the primary spatial feature map.

[0149] S22. Input the primary spatial feature map into the bidirectional long short-term memory network layer of the multi-layer feature extraction network to extract time-dependent features and obtain the spatiotemporal coupled feature sequence.

[0150] S23. Input the spatiotemporal coupled feature sequence into the self-attention network layer of the multi-layer feature extraction network to perform cross-modal correlation analysis and calculate the correlation weight matrix between features;

[0151] S24. Based on the correlation weight matrix, the spatiotemporal coupling feature sequences are weighted and fused to generate equipment status feature data.

[0152] Furthermore, the trend prediction and anomaly detection module 13 is also used for:

[0153] S31. Input the equipment status characteristic data into the trend prediction branch to perform rolling time window prediction and obtain the equipment performance trend data for the next N time steps.

[0154] S32. Input the equipment status feature data into the anomaly detection branch, calculate the deviation of the equipment status feature data from the historical normal data distribution, and obtain the real-time anomaly score.

[0155] S33. Combine and merge the equipment performance trend data and real-time anomaly scores to generate equipment performance prediction data. When the real-time anomaly score exceeds a preset threshold, mark the anomaly point in the equipment performance trend data.

[0156] Furthermore, the parameter optimization scheme generation module 14 is also used for:

[0157] S41. Establish a multi-objective optimization function, which includes the objectives of maximizing equipment energy efficiency, optimizing work quality, and minimizing resource consumption.

[0158] The multi-objective optimization function is expressed as follows:

[0159]

[0160]

[0161]

[0162] In the formula, To maximize the energy efficiency of the equipment, For the effective workload of the equipment, For input power, To optimize the objective function value for job quality, For the first Item of work quality indicators, For the first The corresponding weights of each task quality indicator To minimize the value of the objective function for resource consumption, For the first Consumption of various resources For the first The unit cost of resource consumption;

[0163] S42. Iteratively solve the multi-objective optimization function to obtain the Pareto solution set;

[0164] S43. Perform parameter sensitivity evaluation on the Pareto solution set and obtain the parameter sensitivity evaluation results;

[0165] S44. Based on the parameter sensitivity evaluation results, select the optimal solution from the Pareto solution set and generate parameter optimization scheme data.

[0166] Furthermore, the system also includes a feedback optimization module 15, used for:

[0167] S51. Generate control instructions from the parameter optimization scheme data to obtain the equipment control instruction sequence;

[0168] S52. Based on the equipment control command sequence, real-time parameter adjustment is performed through the PID controller to obtain real-time control output;

[0169] S53. Monitor the performance indicators of the real-time control output and obtain feedback evaluation results;

[0170] S54. Optimize the strategy based on the feedback evaluation results to obtain equipment optimization data.

[0171] Furthermore, the feedback optimization module 15 is also used for:

[0172] S61. Construct the state space from the feedback evaluation results to obtain the environmental state representation;

[0173] S62. Based on the environmental state representation, obtain the training sample set;

[0174] S63. Train a Q-value network on the training sample set to obtain the optimal action value function; wherein, the loss function expression of the Q-value network is:

[0175]

[0176] In the formula, For loss function, For the current network parameters, For the target network parameters, As a discount factor, For experience replay buffer, For empirical quadruplets, This is the current environment state vector. In the state The action to be performed To perform the action The instant reward value obtained afterwards The next state to transition to after performing an action. In the state Actions in the set of all possible actions. For the current Q-network state-action pairs The value estimate, For the target Q-network, state-action pairs The value estimate, This indicates the buffer from the experience replay. Calculate the expectation of empirical quadruples from random sampling.

[0177] S64. Select actions based on the optimal action value function to obtain equipment optimization data.

[0178] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the intelligent improvement method for agricultural equipment as described above.

[0179] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0180] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0181] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for intelligent improvement of supporting equipment in planting industry, characterized in that, The method includes: S1. Obtain multi-source heterogeneous data of the target planting area, and perform data cleaning and spatiotemporal alignment on the multi-source heterogeneous data to obtain aligned multi-source heterogeneous data. S2. The aligned multi-source heterogeneous data is input into a multi-layer feature extraction network to perform hierarchical feature learning and correlation analysis to obtain equipment status feature data; the equipment status feature data is used to characterize the spatiotemporal correlation features of the equipment operating status. S3. Input the device status feature data into the adaptive prediction model for trend prediction and anomaly detection to obtain device performance prediction data; the adaptive prediction model includes a trend prediction branch and an anomaly detection branch, the trend prediction branch is constructed based on a gated recurrent unit network, and the anomaly detection branch is constructed based on a support vector data description algorithm; S4. Perform multi-objective optimization analysis and parameter sensitivity evaluation on the equipment performance prediction data to obtain parameter optimization scheme data.

2. The method according to claim 1, characterized in that, S1 includes: S11. Obtain the original multi-source heterogeneous data of the target planting area; S12. Perform missing value imputation, noise filtering and outlier removal on the original multi-source heterogeneous data to obtain cleaned multi-source heterogeneous data. S13. The cleaned multi-source heterogeneous data is uniformly converted to a preset spatiotemporal reference coordinate system and then meshed and aligned to obtain the aligned multi-source heterogeneous data.

3. The method according to claim 2, characterized in that, S2 includes: S21. Input the aligned multi-source heterogeneous data into the convolutional neural network layer of the multi-layer feature extraction network to extract local spatial features and obtain a primary spatial feature map. S22. Input the primary spatial feature map into the bidirectional long short-term memory network layer of the multi-layer feature extraction network to extract time-dependent features and obtain a spatiotemporal coupled feature sequence. S23. Input the spatiotemporal coupled feature sequence into the self-attention network layer of the multi-layer feature extraction network to perform cross-modal correlation analysis and calculate the correlation weight matrix between features; S24. Based on the correlation weight matrix, the spatiotemporal coupling feature sequence is weighted and fused to generate the device status feature data.

4. The method according to claim 1, characterized in that, S3 includes: S31. Input the device status feature data into the trend prediction branch to perform rolling time window prediction, and obtain the device performance trend data for the next N time steps. S32. Input the device status feature data into the anomaly detection branch, calculate the deviation of the device status feature data from the historical normal data distribution, and obtain the real-time anomaly score; S33. The device performance trend data and the real-time anomaly score are spliced ​​and fused to generate the device performance prediction data. When the real-time anomaly score exceeds a preset threshold, an anomaly point is marked in the device performance trend data.

5. The method according to claim 1, characterized in that, S4 includes: S41. Establish a multi-objective optimization function, wherein the objective optimization function includes the objective of maximizing equipment energy efficiency, the objective of optimizing work quality, and the objective of minimizing resource consumption; The expression for the multi-objective optimization function is as follows: In the formula, To maximize the energy efficiency of the equipment, For the effective workload of the equipment, For input power, To optimize the objective function value for job quality, For the first Item of work quality indicators For the first The corresponding weights of each task quality indicator To minimize the value of the objective function for resource consumption, For the first Consumption of various resources For the first The unit cost of resource consumption; S42. Iteratively solve the multi-objective optimization function to obtain the Pareto solution set; S43. Perform parameter sensitivity evaluation on the Pareto solution set to obtain the parameter sensitivity evaluation results; S44. Based on the parameter sensitivity evaluation results, select the optimal solution from the Pareto solution set and generate the parameter optimization scheme data.

6. The method according to claim 1, characterized in that, The method further includes: S51. Generate control instructions from the parameter optimization scheme data to obtain a sequence of equipment control instructions; S52. Based on the device control command sequence, real-time parameter adjustment is performed through a PID controller to obtain real-time control output; S53. Monitor the performance indicators of the real-time control output and obtain feedback evaluation results; S54. Optimize the strategy based on the feedback evaluation results to obtain equipment optimization data.

7. The method according to claim 6, characterized in that, S54 includes: S61. Construct a state space from the feedback evaluation results to obtain an environmental state representation; S62. Based on the environmental state representation, a training sample set is obtained; S63. Train the Q-value network on the training sample set to obtain the optimal action value function; wherein, the loss function expression of the Q-value network is: In the formula, For loss function, For the current network parameters, For the target network parameters, As a discount factor, For experience replay buffer, For empirical quadruplets, This is the current environment state vector. In the state The action to be performed To perform the action The instant reward value obtained afterwards The next state to transition to after performing an action. In the state Actions in the set of all possible actions. For the current Q-network state-action pairs The value estimate, For the target Q-network, state-action pairs The value estimate, This indicates the buffer from the experience replay. Calculate the expectation of empirical quadruples from random sampling. S64. Perform action selection on the optimal action value function to obtain the equipment optimization data.

8. An intelligent improvement system for agricultural equipment, characterized in that, The system includes: The multi-source data acquisition module is used to acquire multi-source heterogeneous data of the target planting area, and to perform data cleaning and spatiotemporal alignment on the multi-source heterogeneous data to obtain aligned multi-source heterogeneous data. The feature association analysis module is used to input the aligned multi-source heterogeneous data into a multi-layer feature extraction network, perform hierarchical feature learning and association analysis, and obtain equipment status feature data; the equipment status feature data is used to characterize the spatiotemporal association features of the equipment operating status; The trend prediction and anomaly detection module is used to input the equipment status feature data into the adaptive prediction model to perform trend prediction and anomaly detection, and obtain equipment performance prediction data. The adaptive prediction model includes a trend prediction branch and an anomaly detection branch. The trend prediction branch is constructed based on a gated recurrent unit network, and the anomaly detection branch is constructed based on a support vector data description algorithm. The parameter optimization scheme generation module is used to perform multi-objective optimization analysis and parameter sensitivity evaluation on the equipment performance prediction data to obtain parameter optimization scheme data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images