A method for implementing intelligent agricultural planting decision support based on big data analysis

Abstract

An implementation method of intelligent agricultural planting decision support based on big data analysis, including: equipping with diversified sensor equipment and remote sensing technology, monitoring various indicators of farmland environment, integrating external relevant information; using the entropy-based weighted principal component analysis (PCA) method to clean data, and introducing the construction of spatiotemporal sequence features under the geographic space coordinate system; applying statistical big data analysis technology to construct an environment-crop response model and resource demand prediction model based on big data analysis; using the particle swarm optimization algorithm (PSO) based on adaptive learning weights and multimodal search to perform global optimization search and formulate planting decision strategies. The implementation method of intelligent agricultural planting decision support based on big data analysis realizes the intelligence of the entire process from data collection, processing, analysis to decision generation, and provides scientific, accurate, and real-time planting decision support for agricultural producers, with significant application value and broad market prospects.

Description

Technical Field

[0001] The present invention belongs to the fields of agricultural informatization, big data technology, artificial intelligence application and precision agriculture technology, and specifically is an intelligent agricultural planting decision support implementation method based on big data analysis. Background Art

[0002] With the growth of the global population, the limited land resources and the challenges brought by climate change to agricultural production, improving agricultural production efficiency, ensuring food security and achieving sustainable agricultural development have become major issues of concern to the international community. At the same time, the rapid development of science and technology, especially innovations in information technology, Internet of Things technology, big data analysis, artificial intelligence and other fields, has provided strong technical support for solving these problems.

[0003] Traditional agricultural planting management mainly relies on farmers' experience, local climate conditions and soil characteristics, and has limitations, such as insufficient information acquisition: insufficient real-time monitoring of farmland environment (such as soil moisture, nutrient content, pest and disease conditions, etc.) and climate conditions, resulting in decision-making based mainly on experience judgment and rough weather forecasts, lacking accurate data support; non-quantitative and non-personalized decision-making process: planting decisions are often based on general agronomic principles and regional average data, and fail to fully consider the microenvironmental differences between plots, individual differences in crops, and market changes, resulting in planting plans that lack pertinence and flexibility; low resource utilization efficiency: irrigation, fertilization and other agricultural operations often adopt a "one-size-fits-all" approach, and fail to make precise adjustments based on actual crop needs and environmental conditions, resulting in water resource waste, soil nutrient imbalance and environmental pollution; delayed pest and disease control: relying on regular inspections and visual observations to identify pests and diseases, with slow response speed, it is difficult to achieve early warning and precise application of pesticides, increasing prevention and control costs and easily leading to excessive pesticide residues.

[0004] Although some intelligent agricultural decision support systems have been put into use, the following problems are common: low data integration: the system often focuses on a specific link (such as irrigation decision-making, pest and disease warning), lacking in-depth integration and analysis of agricultural data throughout the chain; balance between the versatility and applicability of decision models: existing decision models may be too simplified and unable to fully reflect the diversity of complex agricultural production systems; or too complex and poorly adaptable to specific regions and specific crops; insufficient user-friendliness and ease of operation: some systems have complex interfaces and cumbersome operations, which are not conducive to farmers, especially those who lack information technology backgrounds, to quickly understand and adopt them.

[0005] Although the application of information technology in the agricultural field has made significant progress, the existing agricultural planting decision support system still has many shortcomings. It is necessary to further develop an intelligent agricultural planting decision support system that can fully integrate IoT data, deeply apply big data analysis and artificial intelligence technology, provide all-round and personalized planting decision support, and take into account user friendliness and data privacy protection. The present invention is aimed at the problems existing in the above background technology and proposes an innovative system architecture and implementation method.

[0006] Compared with the prior art, the present invention has the following differences:

[0007] Technical comparison with patent 202311803430.2 "A planting auxiliary decision system based on crop cultivation management model";

[0008] The optimal cultivation management model for crop growth in patent 202311803430.2 relies on the capabilities of the PAI and AIFLOW algorithm platforms, combines crop growth mechanism data, environmental IoT sensing monitoring data, crop growth data, and weather forecast data, builds an agricultural A1 algorithm market, and constructs a digital model for crop production management in different growth periods. The intelligent agricultural planting decision support implementation method based on big data analysis of the present invention adopts statistical big data analysis technology to discover hidden laws and correlations from massive data; constructs an environment-crop response model and a resource demand prediction model based on big data analysis, and an environment-crop response model: uses a recurrent neural network (RNN) model based on a deep attention mechanism and a multi-scale time window.

[0009] Patent 202311803430.2 is that the information platform used by the present invention can integrate multiple heterogeneous databases, such as climate and soil, crop growth, pests and diseases, and agricultural knowledge databases, and combine the optimal management and regulation model of crop growth, use the calculation of the cloud platform, perform simulation and analysis, and integrate into a full-process planting auxiliary decision support system, which assists in decision-making for environmental regulation, water and fertilizer management, pest and disease prediction and early warning, and agricultural early warning. The present invention is a method for implementing intelligent agricultural planting decision support based on big data analysis, which mainly uses the entropy-based weighted principal component analysis (PCA) method for data cleaning, and introduces the construction of spatiotemporal sequence features under the geographic space coordinate system; applies statistical big data analysis technology to construct an environment-crop response model and a resource demand prediction model based on big data analysis; uses a particle swarm optimization algorithm (PSO) based on adaptive learning weights and multimodal search to perform global optimization search and formulate planting decision strategies.

[0010] Technical comparison with patent 201510619678.2 "Planting plan optimization system and method based on agricultural Internet of Things and big data analysis"

[0011] Patent 201510619678.2 is based on the Internet of Things and supported by information technology to build a unified business information management platform to achieve information management of various agricultural production links such as scientific research, production, logistics, and sales, and to achieve reasonable planning, efficient production, scientific prediction, accurate sales, and data decision-making for agricultural product planting and production. The present invention uses a particle swarm optimization algorithm (PSO) based on adaptive learning weights and multimodal search to perform global optimization search and formulate planting decision strategies. The intelligent agricultural planting decision support implementation method based on big data analysis realizes the intelligence of the entire process from data collection, processing, analysis to decision generation, and provides scientific, accurate, and real-time planting decision support for agricultural producers.

[0012] Technical comparison with patent 201910639119.6 "Smart greenhouse control system and method based on big data analysis"

[0013] Patent 201910639119.6 The cloud server determines the crop growth requirements based on the collected crop growth pictures, information processing decision system and convolutional neural network model. The present invention uses a recurrent neural network (RNN) based on deep attention mechanism and multi-scale time window to establish an environment-crop response model, which is mainly used for the next step of optimizing decision-making and scheduling modules.

[0014] Patent 201910639119.6 analyzes the growth environment requirements of the crops planted based on big data, and automatically controls and adjusts the environmental parameters according to the growth environment requirements of the crops, which greatly saves labor costs and increases crop yields. The intelligent agricultural planting decision support implementation method based on big data analysis of the present invention is based on adaptive learning weights and multi-modal search particle swarm optimization algorithm (PSO) to perform global optimization search, formulate planting decision strategies, and realize the full process intelligence from data collection, processing, analysis to decision generation, providing scientific, accurate and real-time planting decision support for agricultural producers.

[0015] Technical comparison with patent 202310443136.9 "An agricultural blockchain big data management device and management method"

[0016] Patent 202310443136.9 Agricultural data processing uses Hadoop's attribute reduction algorithm based on inter-class discrimination to classify agricultural data, and based on the agricultural data classification results, the corresponding agricultural data is delivered and distributed to the established online blockchain website. The present invention is a method for implementing intelligent agricultural planting decision support based on big data analysis, which uses a weighted principal component analysis (PCA) method based on entropy value to clean data, reduce dimension, extract features, and remove invalid, erroneous and missing data.

[0017] Patent 202310443136.9 mines corresponding agricultural information in online blockchain websites through decision tree mining classification algorithm and K-Means clustering algorithm, and acquires target data based on preset target data calculation model, which is helpful for fast query and more effective data download, and helps to better understand agricultural data. The intelligent agricultural planting decision support implementation method based on big data analysis of the present invention is based on adaptive learning weights and multimodal search particle swarm optimization algorithm (PSO) for global optimization search, formulates planting decision strategies, realizes the whole process from data collection, processing, analysis to decision generation, and provides scientific, accurate and real-time planting decision support for agricultural producers. Summary of the invention

[0018] Aiming at the limitations of existing agricultural planting management, such as insufficient information acquisition, non-quantitative and non-personalized decision-making, low resource utilization efficiency, lagging pest and disease prevention and control, high labor intensity and low efficiency, as well as the problems existing in the existing intelligent agricultural decision support system in data integration, versatility and applicability of decision models, system update, etc., the present invention provides an intelligent agricultural planting decision support implementation method based on big data analysis, which uses a weighted principal component analysis (PCA) method based on entropy to perform data cleaning, dimensionality reduction, feature extraction, remove invalid, erroneous and missing data, and introduces the construction of spatiotemporal sequence features under the geographic space coordinate system to form a high-quality data set suitable for deep learning models; statistical big data analysis technology is used to discover hidden laws and correlations from massive data, and an environment-crop response model and a resource demand prediction model based on big data analysis are constructed; a particle swarm optimization algorithm (PSO) based on adaptive learning weights and multimodal search is used to perform global optimization search, formulate planting decision strategies, and maximize the benefit function; agricultural practitioners perform farming operations according to the decision suggestions put forward by the system, and feed back the implementation results to the system, and the system continuously adjusts and optimizes model parameters according to actual effects. It has realized the intelligence of the entire process from data collection, processing, analysis to decision-making, which will help improve agricultural production efficiency, resource utilization efficiency, agricultural product quality and market competitiveness, and promote the sustainable development of agriculture.

[0019] To achieve the above object, the technical solution adopted by the present invention is:

[0020] A method for implementing intelligent agricultural planting decision support based on big data analysis, comprising the following steps:

[0021] S10: Data collection;

[0022] Equipped with a variety of sensor equipment and remote sensing technology, it monitors various indicators of farmland environment in real time, including soil temperature, humidity, pH value, nutrient content, and climate factors including rainfall, light intensity, and wind speed; collects biophysical information during crop growth and development, plant height, chlorophyll content, and images of pest and disease symptoms; and integrates external relevant information such as market data, policies and regulations, and historical yield data;

[0023] S20: data preprocessing and fusion;

[0024] The entropy-based weighted principal component analysis method is used to clean the data, reduce the dimension, extract features, and remove invalid, erroneous, and missing data. The formula is:

[0025]

[0026] And introduce the construction of spatiotemporal sequence features under the geographic space coordinate system to form a high-quality data set suitable for deep learning models;

[0027] S30: Big Data Analysis and Modeling;

[0028] Apply statistical big data analysis technology to discover hidden patterns and correlations from massive data; build an environment-crop response model and resource demand prediction model based on big data analysis. The environment-crop response model: use a recurrent neural network model based on a deep attention mechanism and a multi-scale time window, and the output layer is y=FC(c)=w o c+b0, where w o and b0 are the weight matrix and bias term of the FC layer; Resource demand prediction model: The convolution operation formula is C t =f(W*X t:t+k +b c ), where C t is the output of the convolution layer, W is the convolution kernel, X t:t+k is a subwindow of the input sequence, b c is the bias term, f is the activation function, and combined with the attention mechanism α t =softmax(W a p t +b a ) to weight the features, and finally fuse the feature vector Z = ∑ t α t p t And build a resource demand prediction model by predicting resource demand through the fully connected layer;

[0029] S40: Optimization decision and scheduling module;

[0030] Combined with the output of the above model, a global optimization search is performed using a particle swarm optimization algorithm based on adaptive learning weights and multimodal search to formulate a planting decision strategy and maximize the benefit function.

[0031] S50: Decision implementation and feedback optimization;

[0032] Agricultural practitioners perform agricultural operations based on the decision suggestions put forward by the system and feed back the implementation results to the system; the system continuously adjusts and optimizes model parameters based on actual results to form a data-driven closed-loop control system.

[0033] As a further improvement of the present invention, in step S10, various sensor devices and remote sensing technologies are equipped to monitor various indicators of the farmland environment in real time, including soil temperature, humidity, pH value, nutrient content, and climate factors including rainfall, light intensity, and wind speed; specifically as follows;

[0034] Equipment deployment and layout: evenly or specifically deploy soil sensor networks in the farmland area, including soil temperature sensors, humidity sensors, pH sensors, and soil nutrient sensors, buried at different depths to monitor different levels of soil environment; set up meteorological monitoring stations equipped with rain gauges, light sensors, and anemometers to record climate factors in the farmland environment in real time;

[0035] Remote sensing data acquisition uses drones equipped with high-resolution cameras and multispectral sensors to conduct regular aerial scans of farmland to obtain remote sensing image data of vegetation coverage, normalized vegetation index, and soil moisture for large areas of farmland; remote sensing image data are preprocessed, including radiation correction and geometric correction, to ensure the data is accurate.

[0036] Real-time data transmission: All sensor devices transmit the collected real-time data to edge computing nodes or local data collectors through wireless communication technology to ensure real-time and efficient data transmission;

[0037] Spatiotemporal data fusion integrates sensor data from different time periods and depths of the same plot to construct a dynamic spatiotemporal change picture of the farmland environment, providing a continuous and complete data set for subsequent big data analysis.

[0038] As a further improvement of the present invention, in step S10, biophysical information of crops during growth and development, plant height, chlorophyll content, and images of pests and diseases symptoms are collected; specifically as follows;

[0039] Use height sensors to regularly record crop height data to reflect crop growth rate and development stage;

[0040] Through the use of implanted chlorophyll meters, the chlorophyll content of crop leaves can be obtained in real time, and the photosynthesis efficiency and nutritional status of crops can be indirectly evaluated. By deploying intelligent camera devices in the field, combined with computer vision and deep learning technologies, high-definition images of crop disease and pest symptoms can be automatically captured and identified, achieving early warning of diseases and pests, and formulating effective prevention and control strategies accordingly.

[0041] As a further improvement of the present invention, in step S20, using the weighted principal component analysis method based on entropy value includes the following steps:

[0042] Variables closely related to planting decisions are screened out and combined into the original data matrix X, where X = [X1, X2, …, X n ], X n is the nth eigenvector;

[0043] First, the weight of each feature is calculated by the entropy weight method, W = [w1, w2, ..., wn], where wn represents the weight of the nth feature;

[0044] Next, the weighted data matrix X based on the entropy weights w It is expressed as:

[0045] X w =X×diag(W)

[0046] Where diag(W) is a diagonal matrix, the elements on the diagonal are the elements of W, and the off-diagonal elements are 0, so that the original data is weighted according to the importance of the features;

[0047] Then, the weighted data matrix X w Calculate the covariance matrix and get the weighted covariance matrix C w :

[0048]

[0049] Finally, for C w Perform eigenvalue decomposition, select the largest k eigenvectors as the principal components after dimensionality reduction, and complete the PCA process based on entropy weighting.

[0050] As a further improvement of the present invention, in step S30, statistical big data analysis technology is applied to discover hidden laws and correlations from massive data; the steps of constructing an environment-crop response model and a resource demand prediction model based on big data analysis are as follows:

[0051] Environment-crop response model input layer:

[0052] The original input data is X={X1,X2,…,X T}, where X T Represents the feature vector at time step T, which contains environmental factors and crop historical status information. The environmental factors include temperature, humidity, and light. In order to adapt to multi-scale time windows, the data is divided into K time windows of different lengths. The size of each window is w1, w2, …, w k , the data within the window is organized into subsequences to form a multi-scale input dataset in Represents the feature set in the i-th window;

[0053] For each time window k, apply the RNN (GRU) unit to the data in the window Processing to generate hidden state sequence The update equation of RNN is as follows:

[0054]

[0055] in, and They are update gate and reset gate respectively. is a candidate hidden state, σ is the activation function, and ⊙ represents element-wise multiplication;

[0056] The hidden state H at all scales is H = {H (1) ,H (2) ,…,H (k)}, assigning weights to features of different time steps and scales to enhance the model’s ability to capture key information. The attention weight A is calculated using the following formula:

[0057]

[0058] Among them, Q, K, and V are query, key, and value vectors, respectively, which are obtained from the hidden state H through linear transformation. The Attention function uses different forms such as additive attention or dot product attention, and Softmax is used to normalize the attention weight;

[0059] Environment-crop response model output layer:

[0060] The context vector C obtained by integrating the attention mechanism is used to predict the crop response indicators through the fully connected layer. The crop response indicators include growth rate and yield estimation. The output dimension is o, and the output y calculation formula is:

[0061] y=FC(c)=w o ·c+b0

[0062] Among them, w o and b0 are the weight matrix and bias term of the FC layer;

[0063] Resource demand prediction model collects historical resource input sequence P, environment and crop growth status sequence H, and corresponding future resource demand Q; organizes the data in time series, and converts the sequence data into one-dimensional data according to the needs of CNN; defines a one-dimensional convolution layer, and extracts features from the input sequence through the convolution kernel. The convolution operation formula is:

[0064] C t =f(W*X t：t+k +b c )

[0065] Among them C t is the output of the convolution layer, W is the convolution kernel, X t：t+k is a subwindow of the input sequence, b c is the bias term, f is the activation function;

[0066] Adding max pooling or average pooling layers to reduce the data dimension and retain important features;

[0067] p t =Pool(C t )

[0068] Design an attention weight calculation function to calculate the attention weight of each time step based on the feature vector extracted by CNN. The attention weight calculation formula is:

[0069] α t =softmax(W a p t +b a )

[0070] Where W a and b a are weight and bias terms, α t is the attention weight at time step t;

[0071] The features extracted by CNN are weighted and summed using the attention weights to obtain a comprehensive feature vector. The weighted fusion formula is:

[0072]

[0073] The fused feature vector is mapped to the resource demand prediction space through the fully connected layer. The output formula of the fully connected layer is:

[0074] U = g(W f Z+b f )

[0075] Among them, U is the output of the middle layer, W f and b f are the weights and biases of the fully connected layer, and g is the activation function;

[0076] The output layer predicts future resource demand based on the output of the fully connected layer. The formula is:

[0077]

[0078] in, is the predicted future resource demand, W y and b y are the weights and biases of the final output layer, and f is the deep learning prediction function, which takes into account the influence of time interval.

[0079] As a further improvement of the present invention, in step S40, the steps of performing global optimization search by combining the above model output and using the particle swarm optimization algorithm based on adaptive learning weights and multi-modal search are as follows:

[0080] Initialization: According to the output of the environment-crop response model, the output of the resource demand prediction model, the economic and market data, and the environmental monitoring data, the input data is a particle swarm, including each particle representing a planting decision strategy to set the initial position, i.e., the planting configuration, and the speed, i.e., the strategy adjustment direction;

[0081] Adaptive learning weight adjustment: Dynamically adjust cognitive weight c1(t) and social weight c2(t):

[0082]

[0083] c1(t) and c2(t) represent the cognitive weight and social weight at the tth iteration, respectively, which change dynamically with the number of iterations; t represents the current number of iterations, which starts from the initial iteration (1) and gradually increases as the algorithm executes; T is the total number of iterations, which is the stopping condition predetermined by the algorithm. The algorithm will terminate after reaching this number of iterations; c 1min and c 1max are the minimum and maximum values ​​of cognitive weight, respectively, which determine the lower and upper limits of weight adjustment; C 2min and c 2max The same principle applies to social weights, setting their range of variation;

[0084] Multimodal search: Maintain m cluster centers (μ1, μ2, …, μ m ), representing different optimal solution modes; particles are classified according to their similarity with the cluster center to promote multimodal search;

[0085] For each particle's decision plan, the benefit is calculated according to the given formula:

[0086]

[0087] Ensure that each decision plan meets resource constraints and market demand constraints:

[0088]

[0089] Update the velocity and position of each particle based on its current position, personal optimal position, and global optimal position, as well as inertia weight, cognitive and social learning factors:

[0090]

[0091] Repeat the above steps until the preset maximum number of iterations is reached or the profit function converges to a certain threshold;

[0092] Obtaining the global optimal solution, that is, the planting decision plan with the greatest benefit, is the final planting decision strategy.

[0093] Compared with the prior art, the present invention has the following beneficial effects:

[0094] The present invention is a kind of intelligent agricultural planting decision support implementation method based on big data analysis. By integrating multi-source agricultural data and applying big data analysis technology, the system can accurately reveal the intrinsic connection between crop growth and environmental factors, eliminate the subjectivity and uncertainty in the traditional decision-making process, and ensure the objectivity and accuracy of the decision-making basis; real-time monitoring of farmland environment and crop status, combined with market dynamic information, real-time update of decision suggestions, help farmers quickly respond to emergencies such as climate change and pest outbreaks, reduce losses, and improve production flexibility; use the particle swarm optimization algorithm (PSO) based on adaptive learning weights and multimodal search to perform global optimization search, formulate planting decision strategies, and maximize the benefit function; agricultural practitioners perform agricultural operations according to the decision suggestions proposed by the system, and feed back the implementation results to the system, and the system continuously adjusts and optimizes the model parameters according to the actual effect. The intelligent agricultural planting decision support implementation method of the present invention realizes the scientific, precise, personalized and dynamic agricultural planting decision-making through the application of big data analysis technology, effectively improves agricultural production efficiency, resource utilization efficiency, agricultural product quality and market competitiveness, reduces the labor intensity of farmers, enhances agricultural risk resistance, and promotes agricultural sustainable development. It has great significance and wide application value for the transformation and upgrading of modern agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1 It is a flowchart of an implementation method of intelligent agricultural planting decision support based on big data analysis;

[0096] Figure 2 Schematic diagram of deploying various sensors, remote sensing equipment and data transmission equipment for farmland;

[0097] Figure 3 It is a schematic diagram of remote sensing equipment and data transmission equipment;

[0098] Figure 4 It is a flow chart of the entropy-based weighted principal component analysis (PCA) method;

[0099] Figure 5 Schematic diagram of the process of particle swarm optimization algorithm (PSO) based on adaptive learning weights and multimodal search. DETAILED DESCRIPTION

[0100] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments:

[0101] like Figure 1 The method for implementing intelligent agricultural planting decision support based on big data analysis is shown, comprising the following steps:

[0102] Step 1: Data collection, equipped with a variety of sensor equipment and remote sensing technology, buried at different depths to monitor different levels of soil environment, set up meteorological monitoring stations, equipped with rain gauges, light sensors, anemometers and other equipment to record climate factors in the farmland environment in real time; collect biophysical information during crop growth and development, plant height, chlorophyll content, and images of pest and disease symptoms; integrate market data, policies and regulations, and external related information on historical production data.

[0103] Step 2: Data preprocessing and fusion, using the entropy-based weighted principal component analysis (PCA) method to clean the data, reduce the dimension, extract features, and remove invalid, erroneous, and missing data; and bind various farmland environmental indicators to their geographic spatial coordinates to form spatiotemporal data points; sort the data according to the time series, and combine the geographic coordinate information to construct a spatiotemporal series that reflects the changing trend of the farmland environment; combine the basic features processed by PCA with the spatiotemporal sequence features to construct a rich and multidimensional spatiotemporal feature vector; integrate the processed feature data to form a structured, standardized, and high-quality data set rich in spatiotemporal dimension information, which is conducive to the learning and training of deep learning models.

[0104] Step 3: Big data analysis and modeling. Apply statistical big data analysis techniques to discover hidden patterns and correlations from massive data; build an environment-crop response model and resource demand prediction model based on big data analysis. The environment-crop response model: use a recurrent neural network (RNN) model based on a deep attention mechanism and a multi-scale time window, and the output layer is y = FC (c) = w o ·c+b0, where, where, w o and b0 are the weight matrix and bias term of the FC layer; Resource demand prediction model: The convolution operation formula is C t =f(W*X t:t+k +b c ), where C t is the output of the convolution layer, W is the convolution kernel, Xt:t+k is the subwindow of the input sequence, b c is the bias term, f is the activation function (ReLU), and combined with the attention mechanism α t =softmax(W a p t +b a ) to weight the features, and finally fuse the feature vector Z = ∑ t α t p t And build a resource demand prediction model by predicting resource demand through the fully connected layer.

[0105] Step 4: Optimize decision-making and scheduling module. Combined with the above model output, the particle swarm optimization algorithm (PSO) based on adaptive learning weights and multimodal search is used to perform global optimization search, formulate planting decision strategies, and maximize the benefit function.

[0106] Initialization: According to the output of the environment-crop response model, the output of the resource demand prediction model, the economic and market data, and the environmental monitoring data, the input data is the particle swarm (each particle represents a planting decision strategy) to set the initial position (planting configuration) and speed (strategy adjustment direction).

[0107] Adaptive learning weight adjustment: Dynamically adjust cognitive weight c1(t) and social weight c2(t):

[0108]

[0109] c1(t) and c2(t) represent the cognitive weight and social weight at the tth iteration, respectively. They will change dynamically with the number of iterations. t represents the current number of iterations, which starts from the initial iteration (1) and gradually increases as the algorithm executes. T is the total number of iterations, which is the stopping condition predetermined by the algorithm. The algorithm will terminate after reaching this number of iterations. 1min and c 1max are the minimum and maximum values ​​of cognitive weight, respectively, which determine the lower and upper limits of weight adjustment; c 2min and c 2max The same principle is applied to social weights to set their range of variation.

[0110] Multimodal search: Maintain m cluster centers (μ1,μ2,…,μ m ), representing different optimal solution modes; particles are classified according to their similarity with the cluster center to promote multimodal search.

[0111] For each particle (decision plan), the benefit is calculated according to the given formula:

[0112]

[0113] Ensure that each decision plan meets resource constraints and market demand constraints:

[0114]

[0115] Update the velocity and position of each particle based on its current position, personal optimal position, and global optimal position, as well as inertia weight, cognitive and social learning factors:

[0116]

[0117] Repeat the above steps until the preset maximum number of iterations is reached or the profit function converges to a certain threshold.

[0118] Obtaining the global optimal solution (i.e. the planting decision plan with the greatest benefit) is the final planting decision strategy.

[0119] Step 5: Decision implementation and feedback optimization. Agricultural practitioners perform agricultural operations according to the decision suggestions put forward by the system and feed back the implementation results to the system; the system continuously adjusts and optimizes the model parameters according to the actual results, forming a data-driven closed-loop control system to improve the accuracy and adaptability of decision-making.

[0120] Embodiment:

[0121] In the step 1, various sensors, remote sensing equipment and data transmission equipment are deployed in the target farmland, and data acquisition equipment is deployed to obtain various multi-dimensional data, including:

[0122] a1. Figure 2 , 3 As shown, various sensors, remote sensing equipment and data transmission equipment are deployed in the target farmland, and soil sensor networks are deployed evenly or targetedly in the farmland area to record climate factors in the farmland environment in real time.

[0123] a2. All sensor devices transmit the collected real-time data to the edge computing node or local data collector through wireless communication technology to ensure real-time and efficient data transmission.

[0124] like Figure 4 As shown, the data preprocessing and fusion described in step 2, using the entropy-based weighted principal component analysis (PCA) method, specifically includes:

[0125] a1. Filter out the variables that are closely related to planting decisions and combine them into the original data matrix X, where X = [X1, X2, …, X n ], X n is the nth feature vector. First, the weight of each feature is calculated by the entropy weight method, W = [w1, w2, ..., wn], where wn represents the weight of the nth feature.

[0126] a2. Weighted data matrix X based on entropy weight w It can be expressed as:

[0127] X w =X×diag(W)

[0128] Where diag(W) is a diagonal matrix, the elements on the diagonal are the elements of W, and the off-diagonal elements are 0. This achieves the weighting of the original data according to the importance of the features.

[0129] a3. Weighted data matrix X w Calculate the covariance matrix and get the weighted covariance matrix C w :

[0130]

[0131] Finally, for C w Perform eigenvalue decomposition, select the largest k eigenvectors as the principal components after dimensionality reduction, and complete the PCA process based on entropy weighting.

[0132] In step 3, the construction of the environment-crop response model and resource demand prediction model based on big data analysis includes:

[0133] a1. Input layer of environment-crop response model:

[0134] The original input data is X={X1,X2,…,X T}, where X T Represents the feature vector at time step T, which contains environmental factors (temperature, humidity, light) and crop historical status information. In order to adapt to multi-scale time windows, the data is divided into K time windows of different lengths, and the size of each window is w1, w2, …, w k , the data within the window is organized into subsequences to form a multi-scale input dataset in Represents the feature set within the i-th window.

[0135] a2. For each time window k, apply the RNN (GRU) unit to the data in the window Processing to generate hidden state sequence The update equation of RNN is as follows:

[0136]

[0137] in, and They are update gate and reset gate respectively. is a candidate hidden state, σ is the activation function (sigmoid), and ⊙ represents element-wise multiplication.

[0138] a3. The hidden state H at all scales = {H (1) ,H (2) ,…,H (k)}, assigning weights to features at different time steps and scales to enhance the model's ability to capture key information. The attention weight A is calculated using the following formula:

[0139]

[0140] Among them, Q, K, and V are query, key, and value vectors, respectively, which are obtained from the hidden state H through linear transformation. The Attention function can adopt different forms such as additive attention or dot product attention, and Softmax is used to normalize the attention weight.

[0141] a4. Environment-crop response model output layer:

[0142] The context vector C obtained by integrating the attention mechanism is used to predict crop response indicators (growth rate, yield estimation) through the fully connected layer (FC). The output dimension is o, and the output y calculation formula is:

[0143] y=FC(c)=w o ·c+b0

[0144] Among them, w o and b0 are the weight matrix and bias term of the FC layer.

[0145] a5. Resource demand prediction model, collect historical resource input sequence P, environment and crop growth status sequence H, and corresponding future resource demand Q; organize the data in time series, and convert the sequence data into one-dimensional data according to the needs of CNN; define a one-dimensional convolution layer, and extract features from the input sequence through the convolution kernel. The convolution operation formula is:

[0146] C t =f(W*X t:t+k +b c )

[0147] Among them C t is the output of the convolution layer, W is the convolution kernel, X t:t+k is a subwindow of the input sequence, b c is the bias term, and f is the activation function (ReLU).

[0148] a6. Add a maximum pooling or average pooling layer to reduce the data dimension and retain important features.

[0149] p t=Pool(C t )

[0150] a7. Design an attention weight calculation function to calculate the attention weight of each time step based on the feature vector extracted by CNN. The attention weight calculation formula is:

[0151] α t =softmax(W a p t +b a )

[0152] Where W a and b a are weight and bias terms, α t is the attention weight at time step t.

[0153] The fused feature vector is mapped to the resource demand prediction space through the fully connected layer. The output formula of the fully connected layer is:

[0154] U = g(W f Z+b f )

[0155] Among them, U is the output of the middle layer, W f and b f are the weights and biases of the fully connected layer, and g is the activation function.

[0156] a8. The output layer predicts future resource demand based on the output of the fully connected layer. The formula is:

[0157]

[0158] in, is the predicted future resource demand, W y and b y are the weights and biases of the final output layer, and f is the deep learning prediction function, which takes into account the influence of time interval.

[0159] like Figure 5 As shown, step 4 uses the particle swarm optimization algorithm (PS0) based on adaptive learning weights and multimodal search to perform global optimization search and formulate a planting decision strategy. The maximization benefit function includes:

[0160] a1. Initialization: According to the output of the environment-crop response model, the output of the resource demand prediction model, economic and market data, and environmental monitoring data, input data is the particle swarm N, and each particle X is initialized. j (j=1, 2, ..., N) position (decision variable value) and velocity V i .

[0161] a2. Calculate each particle X jThe corresponding profit function value f(X j ), as its fitness.

[0162] a3. Update the personal best position. If the current particle X j The fitness is better than its historical best position pbest j The fitness of pbest is updated j For the current X j , while retaining the corresponding benefit function value

[0163] a4. Update the global best position, compare the historical best positions of all particles, and find the global best position gbest and its corresponding benefit function value.

[0164] a5. Update the particle speed and position. Update the speed and position of each particle according to the following formula:

[0165]

[0166] Where t represents the current iteration number, W is the inertia weight, c1 and c2 are the cognitive coefficient and social coefficient (usually positive values), and r1 and r2 are random numbers between [0, 1].

[0167] a6. Iteration termination condition: when the preset maximum number of iterations is reached or the change of the global optimal solution is less than a certain threshold, the iteration is stopped and the global best position gbest is output as the optimal planting decision strategy.

[0168] The above description is only a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any modification or equivalent change made based on the technical essence of the present invention still falls within the scope of protection required by the present invention.

Claims

1. A method for implementing intelligent agricultural planting decision support based on big data analysis, characterized in that: The following steps are involved: S10: Data collection: real-time monitoring of various indicators of farmland environment, collection of biophysical information during crop growth and development, integration of external related information such as market data, policies and regulations, and historical yield data; S20: Data preprocessing and fusion: Use the entropy-based weighted principal component analysis method to clean the data, reduce the dimension, extract features, remove invalid, erroneous and missing data, and introduce the spatiotemporal sequence feature construction under the geographic spatial coordinate system to form a high-quality data set suitable for deep learning models; S30: Big data analysis and modeling: Constructing environment-crop response model and resource demand prediction model. The steps are as follows: Environment-crop response model input layer: The original input data is X={X1,X2,…,X T }, where X T Represents the feature vector at time step T, which contains environmental factors and crop historical status information. The environmental factors include temperature, humidity, and light. The data is divided into K time windows of different lengths, and the size of each window is w1, w2, …, w k , the data within the window is organized into subsequences to form a multi-scale input dataset in Represents the feature set in the i-th window; For each time window k, apply the GRU unit of the RNN to the data in the window Processing to generate hidden state sequence The hidden state H at all scales is H = {H (1) ,H (2) ,…,H (k) }The multi-head self-attention mechanism is applied to assign weights to features of different time steps and scales to enhance the model's ability to capture key information; Environment-crop response model output layer: The context vector C obtained by integrating the attention mechanism is used to predict the crop response indicators through the fully connected layer. The crop response indicators include growth rate and yield estimation. The output dimension is o and the output is y. Resource demand prediction model: collect historical resource input sequences, environmental and crop growth status sequences, and corresponding future resource demand; organize the data in time series, and convert the sequence data into one-dimensional data according to the needs of CNN; define a one-dimensional convolution layer, extract features from the input sequence through the convolution kernel, and perform convolution operations using the convolution operation formula; Adding max pooling or average pooling layers to reduce the data dimension and retain important features; Design an attention weight calculation function to calculate the attention weight of each time step based on the feature vector extracted by CNN; Use the attention weights to perform weighted summation on the features extracted by CNN to obtain a comprehensive feature vector; The fused feature vector is mapped to the resource demand prediction space through a fully connected layer; The output layer predicts future resource demand based on the output of the fully connected layer; S40: Optimization decision and scheduling module: Combine the output of the environment-crop response model, the output of the resource demand prediction model, economic and market data, and environmental monitoring data, and use the particle swarm optimization algorithm based on adaptive learning weights and multimodal search to perform global optimization search and formulate planting decision strategies; obtain the global optimal solution, that is, the planting decision plan with the greatest benefit, which is the final planting decision strategy; S50: Decision implementation and feedback optimization: Agricultural practitioners perform agricultural operations according to the decision suggestions put forward by the system and feed back the implementation results to the system; the system continuously adjusts and optimizes model parameters according to actual results to form a data-driven closed-loop control system.

2. According to claim 1, a method for implementing intelligent agricultural planting decision support based on big data analysis is characterized in that: Step S10 includes: Equipped with a variety of sensor equipment and remote sensing technology, it can monitor various indicators of farmland environment in real time, including soil temperature, humidity, pH value, nutrient content, and climate factors including rainfall, light intensity, and wind speed; collect biophysical information during crop growth and development, plant height, chlorophyll content, and images of pest and disease symptoms; and integrate external relevant information such as market data, policies and regulations, and historical yield data; Equipment deployment and layout: evenly or specifically deploy soil sensor networks in the farmland area, including soil temperature sensors, humidity sensors, pH sensors, and soil nutrient sensors, buried at different depths to monitor different levels of soil environment; set up meteorological monitoring stations equipped with rain gauges, light sensors, and anemometers to record climate factors in the farmland environment in real time; Remote sensing data acquisition: using drones equipped with high-resolution cameras and multispectral sensors to regularly scan farmland from the air to obtain remote sensing image data of vegetation coverage, normalized vegetation index, and soil moisture for large areas of farmland; pre-processing remote sensing image data, including radiation correction and geometric correction, to ensure data accuracy; Real-time data transmission: All sensor devices transmit the collected real-time data to edge computing nodes or local data collectors through wireless communication technology to ensure real-time and efficient data transmission; Spatiotemporal data fusion integrates sensor data from different time periods and depths of the same plot to construct a dynamic spatiotemporal change picture of the farmland environment, providing a continuous and complete data set for subsequent big data analysis.

3. The method for implementing intelligent agricultural planting decision support based on big data analysis according to claim 1 is characterized in that: Step S10 includes: Height sensors are used to regularly record crop height data to reflect crop growth rate and development stage. An implanted chlorophyll meter is used to obtain real-time chlorophyll content in crop leaves to indirectly evaluate crop photosynthesis efficiency and nutritional status. Intelligent cameras are deployed in the field, combined with computer vision and deep learning technologies to automatically capture and identify high-definition images of crop disease and pest symptoms, achieve early warning of disease and pests, and formulate effective prevention and control strategies accordingly.

4. The method for implementing intelligent agricultural planting decision support based on big data analysis according to claim 1 is characterized in that: Step S20 includes: Variables closely related to planting decisions are screened out and combined into the original data matrix Y, where Y = [Y1,Y2,…,Y n ],Y n is the nth eigenvector; First, the weight of each feature is calculated by the entropy weight method, W = [w1, w2, ..., wn], where wn represents the weight of the nth feature; Next, the weighted data matrix X based on the entropy weights w It is expressed as: X w =Y×diag(W) Where diag(W) is a diagonal matrix, the elements on the diagonal are the elements of W, and the off-diagonal elements are 0, so that the original data is weighted according to the importance of the features; Then, the weighted data matrix X w Calculate the covariance matrix and get the weighted covariance matrix C w : Finally, for C w Perform eigenvalue decomposition, select the largest k eigenvectors as the principal components after dimensionality reduction, and complete the PCA process based on entropy weighting.

5. The method for implementing intelligent agricultural planting decision support based on big data analysis according to claim 1 is characterized in that: In step S30, the update equation of the RNN is as follows: in, and They are update gate and reset gate respectively. is a candidate hidden state, σ is the activation function, and ⊙ represents element-wise multiplication; In step S30, the attention weight A is calculated by the following formula: Among them, the query, key, and value vectors are obtained from the hidden state H through linear transformation, the Attention function uses additive attention or dot product attention, and Softmax is used to normalize the attention weight; In step S30, the output y calculation formula is: y=FC(c)=w o ·c+b0 Among them, w o and b0 are the weight matrix and bias term of the FC layer; In step S30, the convolution calculation formula is: C t =f jh (W jjh *X t:t+k +b c ) Among them C t is the output of the convolutional layer, W jjh is the convolution kernel, X t:t+k is the subwindow of the input sequence, b c is the bias term, f jh is the activation function; In step S30, after the convolution calculation is completed, the following steps are also included: Adding max pooling or average pooling layers to reduce the data dimension and retain important features; p t =Pool(C t ) Design an attention weight calculation function to calculate the attention weight of each time step based on the feature vector extracted by CNN. The attention weight calculation formula is: a t =softmax(W a p t +b a ) Where W a and b a are weight and bias terms, α t is the attention weight at time step t; The features extracted by CNN are weighted and summed using the attention weights to obtain a comprehensive feature vector. The weighted fusion formula is: The fused feature vector is mapped to the resource demand prediction space through the fully connected layer. The output formula of the fully connected layer is: U = g(W f Z+b f ) Among them, U is the output of the middle layer, W f and b f are the weights and biases of the fully connected layer, and g is the activation function; The output layer predicts future resource demand based on the output of the fully connected layer. The formula is: in, is the predicted future resource demand, W y and b y are the weights and biases of the final output layer, and f is the deep learning prediction function, which takes into account the influence of time interval.

6. The method for implementing intelligent agricultural planting decision support based on big data analysis according to claim 1 is characterized in that: In step S40, combined with the above model output, the global optimization search steps are performed using the particle swarm optimization algorithm based on adaptive learning weights and multi-modal search as follows: Initialization: According to the output of the environment-crop response model, the output of the resource demand prediction model, the economic and market data, and the environmental monitoring data, the input data is a particle swarm, including each particle representing a planting decision strategy to set the initial position, i.e., the planting configuration, and the speed, i.e., the strategy adjustment direction; Adaptive learning weight adjustment: Dynamically adjust cognitive weight c1(t) and social weight c2(t): c1(t) and c2(t) represent the cognitive weight and social weight at the tth iteration, respectively, which change dynamically with the number of iterations; t dd represents the current iteration number, starting from the initial iteration (1) and increasing gradually as the algorithm executes; T dd is the total number of iterations, i.e. the algorithm's predetermined stopping condition. The algorithm will terminate after reaching this number of iterations; c 1min and c 1max are the minimum and maximum values ​​of cognitive weight, respectively, which determine the lower and upper limits of weight adjustment; c 2min and c 2max The same principle applies to social weights, setting their range of variation; Multimodal search: Maintain m cluster centers (μ1,μ2,…,μ m ), representing different optimal solution modes; particles are classified according to their similarity with the cluster center to promote multimodal search; For each particle's decision plan, the benefit is calculated according to the given formula: Ensure that each decision plan meets resource constraints and market demand constraints: Update the velocity and position of each particle based on its current position, personal optimal position, and global optimal position, as well as inertia weight, cognitive and social learning factors: Repeat the above steps until the preset maximum number of iterations is reached or the profit function converges to a certain threshold; Obtaining the global optimal solution, that is, the planting decision plan with the greatest benefit, is the final planting decision strategy.

Images