A coordinated optimization method for food security combining regional coordinated development analysis

Abstract

The application discloses a kind of food security coordination optimization methods combined with regional coordinated development analysis, comprising: obtaining several different regional coordinated development level evaluation examples, classifying each regional coordinated development level evaluation example, and constructing regional coordinated development level evaluation example dataset;Based on the regional coordinated development level evaluation example dataset, food security evaluation index and regional coordinated development level evaluation index are respectively screened, and food security evaluation dataset and regional coordinated development level evaluation dataset are constructed;Each evaluation index parameter of each region of target region is collected, food security prediction and regional coordination degree evaluation are carried out, and food reserve allocation scheme is formulated to push out. To overcome the limitations of traditional index screening, provide scientific decision-making basis and guidance for regional food security.

Description

Technical Field

[0001] The present invention relates to the technical field of food security coordination optimization, and in particular to a food security coordination optimization method combined with regional coordinated development analysis. Background Art

[0002] In the traditional field of food security coordination and optimization, existing technologies often rely on single-dimensional yield forecasting models, employing time series analysis or regression methods to model production factors like climate and soil in isolation. These methods lack quantitative consideration of dynamic correlations such as inter-regional resource flows and industrial chain collaboration. Furthermore, regional coordination evaluation systems generally employ static weight allocation, making it difficult to capture the spatiotemporal evolution of cross-regional factors like infrastructure connectivity and ecological compensation mechanisms. This results in an inaccurate match between coordination analysis results and food distribution needs.

[0003] When developing reserve allocation plans, traditional linear programming methods can handle optimization problems under fixed constraints, but they fail to fully consider the impact of dynamic changes in regional coordination levels on transportation route reliability. Therefore, there is an urgent need to develop a food security coordination optimization method that incorporates regional coordinated development analysis to address the core pain points of existing technologies, such as insufficient forecasting accuracy, significant decision-making lags, and inefficient cross-regional resource allocation. Summary of the Invention

[0004] The present invention overcomes the defects of the prior art and provides a food security coordination optimization method combined with regional coordinated development analysis.

[0005] To achieve the above objectives, the present invention provides a first aspect of a food security coordination optimization method combined with regional coordinated development analysis, comprising:

[0006] Obtain several different regional coordinated development level evaluation instances, classify the regional coordinated development level evaluation instances into categories, and construct a regional coordinated development level evaluation instance dataset;

[0007] Based on the regional coordinated development level evaluation example dataset, food security evaluation indicators and regional coordinated development level evaluation indicators are screened respectively, and a food security evaluation dataset and a regional coordinated development level evaluation dataset are constructed.

[0008] In this solution, the acquisition of several different regional coordinated development level evaluation instances, classification of the regional coordinated development level evaluation instances, and construction of a regional coordinated development level evaluation instance dataset specifically includes:

[0009] Acquire a number of evaluation instances of coordinated development levels of different regions based on big data retrieval, wherein the obtained evaluation instances of coordinated development levels of different regions include relevant text descriptions and evaluation parameters of the evaluation of coordinated development levels of different regions;

[0010] Extract relevant text descriptions of each evaluation instance from several different regional coordinated development level evaluation instances, import them into the pre-trained BERT model, and use the attention mechanism to obtain the global context embedding of different words in the relevant text descriptions;

[0011] Build word vectors and keyword vectors based on the relevant text description of each evaluation instance, calculate the semantic relative distance between different word vectors and keyword vectors, and select words with a semantic relative distance less than a preset threshold as local context embeddings;

[0012] The global context embedding and the local context embedding are concatenated to obtain semantic features. Based on the obtained semantic features, evaluation labels for the corresponding regional coordinated development evaluation instances are generated. The Euclidean distance algorithm is used to calculate the Euclidean distance between the evaluation labels.

[0013] The calculated Euclidean distance value is compared with a preset Euclidean distance threshold, and regional coordinated development evaluation instances with a value greater than the preset Euclidean distance threshold are defined as the same category, thereby obtaining several evaluation instance categories;

[0014] The dynamic time warping algorithm is used to align the evaluation parameters of each evaluation instance in the regional coordinated development level evaluation, and outliers are eliminated and missing values ​​are supplemented. After completing data preprocessing, a regional coordinated development level evaluation instance dataset is constructed in combination with the divided evaluation instance categories.

[0015] This solution is characterized in that the method of screening food security evaluation indicators and regional coordinated development level evaluation indicators based on the regional coordinated development level evaluation instance dataset, and constructing a food security evaluation dataset and a regional coordinated development level evaluation dataset specifically includes:

[0016] Obtaining a regional coordinated development level evaluation instance dataset, extracting evaluation indicators related to food security and corresponding evaluation parameters from the regional coordinated development level evaluation instance dataset, calibrating the selected indicators as first screening indicators, and generating a primary screening indicator dataset;

[0017] The primary screening indicator dataset is imported as input into the LASSO regression model for preliminary dimensionality reduction, the regularization strength parameter is preset, the L1 regularization constraint is used to impose a sparsity penalty on the regression coefficient, and the primary screening indicators with non-zero regression coefficients are screened to generate a secondary screening indicator dataset;

[0018] The regularization strength parameter is obtained by dividing the initial screening index data set into several training subsets and validation subsets, constructing a preset logarithmic space using the validation subsets, traversing and calculating the regularization strength parameter in the preset logarithmic space, and selecting the regularization strength parameter with the smallest validation error as the optimal solution;

[0019] A random forest algorithm is introduced to re-screen the secondary screening indicators, the secondary screening indicator set and the evaluation parameters corresponding to the secondary screening indicators are initialized as the initial input data of the algorithm, and an initial weight is set for each secondary screening indicator;

[0020] The decision tree is constructed by generating an indicator feature set through the initial input data. A number of sample samples are obtained by probability sampling based on the preset initial weights. The hierarchical structure of the decision tree is constructed through the obtained sample samples and the Gini index is used as the basis for node segmentation. After the segmentation is completed, the random forest is output;

[0021] After outputting the random forest, extract the out-of-bag samples of each tree, randomly perturb the random forest based on the extracted out-of-bag samples and calculate the error increment. After traversing all trees in the forest, obtain the error increment of the target indicator on all trees and take the average as the global importance score;

[0022] Based on the global importance score of each indicator, an importance score ranking table is generated. The indicators within the preset range are output through the importance score ranking table as food security evaluation indicators and combined with the corresponding historical evaluation parameters to form a food security evaluation dataset.

[0023] In this solution, the method of screening food security evaluation indicators and regional coordinated development level evaluation indicators based on the regional coordinated development level evaluation example dataset, and constructing a food security evaluation dataset and a regional coordinated development level evaluation dataset, further includes:

[0024] Obtaining a regional coordinated development level evaluation instance dataset, extracting evaluation indicators related to regional coordinated development from the regional coordinated development level evaluation instance dataset, and calibrating them as second screening evaluation indicators;

[0025] The principal component analysis method is introduced to perform preliminary dimensionality reduction on the selected second screening evaluation index. The second screening evaluation index is used as input data to calculate the covariance matrix. The calculated covariance matrix is ​​subjected to eigenvalue decomposition to obtain eigenvalues ​​and corresponding eigenvectors, and several principal components are obtained.

[0026] Sorting the principal components based on the eigenvalues ​​of the principal components, selecting the principal components within a preset range as target principal components based on the sorting results, and obtaining a principal component index set; using the principal component index set as input, screening the independent principal component indices using the independent principal component analysis method;

[0027] The input principal component index set is centralized, and the separation matrix is ​​generated based on the initialization of the processed principal component index set. The orthogonal basis vectors of the principal component direction are used as the initial estimate to iteratively optimize the separation matrix.

[0028] The final separation matrix is ​​output through continuous iterative optimization analysis until the preset stopping criterion is reached. The independent principal component index is obtained by inverse transformation according to the final separation matrix, which is used as the evaluation index of the regional coordinated development level. The historical evaluation parameters of the corresponding indicators are obtained using the regional coordinated development level evaluation instance data set to form the regional coordinated development level evaluation data set.

[0029] In this plan, the evaluation index parameters of each region in the target area are collected to conduct food security forecasts and regional coordination evaluations, and a food reserve allocation plan is formulated and pushed, specifically including:

[0030] Obtain a food security evaluation data set, introduce a Markov algorithm, and calculate the state transition probability of historical evaluation parameters corresponding to each food security evaluation indicator in the food security evaluation data set under different time series characteristics through the Markov algorithm;

[0031] Constructing a food security prediction model based on a generative adversarial network, establishing a training dataset based on a food security evaluation dataset to train the food security prediction model, and introducing the calculated state transition probability into the model training process;

[0032] When the training data set is input into the food security prediction model, the generator reconstructs the sequence according to the input training data sequence, and introduces the state transition probability in the reconstruction process to correct the reconstructed sequence;

[0033] The discriminator’s discrimination rule is set based on the calculated state transition probability. The reconstructed sequence generated by the generator is input into the discriminator. If it meets the discrimination rule, the current reconstructed sequence is accepted and output. If it does not meet the discrimination rule, a reconstruction penalty is applied and the next reconstructed sequence is obtained for discrimination.

[0034] Through repeated iterative learning, a food security prediction model that meets the preset verification standards is obtained. Based on the screened food security indicators, the evaluation index parameters of various regions in the target area are collected in real time and input into the trained food security prediction model to obtain food security prediction information.

[0035] In this plan, the evaluation index parameters of each region in the target area are collected to make food security forecasts and regional coordination evaluations, and a food reserve allocation plan is formulated and pushed. The plan also includes:

[0036] Acquire a regional coordinated development level evaluation data set, and construct a regional coordinated development level evaluation system based on a regional coordinated development level evaluation indicator set and corresponding evaluation parameters in the regional coordinated development level evaluation data set;

[0037] Monitor the target area to obtain the real-time parameters corresponding to the evaluation indicators of the coordinated development level of each region. Combined with the constructed regional coordinated development level evaluation system, analyze the regional coordination level of each region in the target area to obtain regional coordination level analysis information;

[0038] Obtaining food security forecast information, importing the food security forecast information into a pre-trained Bayesian inference network to obtain a transient posterior distribution, and inferring the probability of a food reserve gap in each region within the target area based on the obtained transient posteriori to obtain gap probability inference information;

[0039] The regional coordination degree analysis information is compared with a preset threshold, and areas within the target area where the regional coordination degree is less than the preset threshold are marked as risk areas, and areas where the regional coordination degree is greater than the preset threshold are marked as safe areas. The gap probability inference information is combined to generate a food reserve risk heat map;

[0040] Using the regional geographic system, feasible paths within the target area and the geographical locations of risk areas and safe areas are calibrated in the food reserve risk heat map to generate a node topology map, wherein the safe area is calibrated as the starting node and the risk area is the ending node;

[0041] The ant colony algorithm is introduced to formulate the food reserve allocation plan. The existence of three optimal paths at the end node is set as the node path optimization stop condition. The parameters are initialized in the node topology graph to obtain a virtual ant colony.

[0042] The virtual ant colony is driven by the preset path selection probability to perform path optimization, and iterative updates are performed until the termination conditions are met to output the optimal food allocation path for each risk area. After a feasibility assessment, a food reserve allocation plan is generated.

[0043] The second aspect of the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium includes a food security coordination optimization method program combined with regional coordinated development analysis. When the food security coordination optimization method program suitable for combining regional coordinated development analysis is executed by a processor, the steps of the food security coordination optimization method combined with regional coordinated development analysis as described in any one of the above items are implemented.

[0044] The application discloses a grain safety coordination optimization method combined with regional coordination development analysis, and comprises the following steps: acquiring a plurality of regional coordination development level evaluation instances, classifying the regional coordination development level evaluation instances, and constructing a regional coordination development level evaluation instance dataset; screening grain safety evaluation indexes and regional coordination development level evaluation indexes based on the regional coordination development level evaluation instance dataset, and constructing a grain safety evaluation dataset and a regional coordination development level evaluation dataset; collecting evaluation index parameters of each region in a target region, performing grain safety prediction and regional coordination degree evaluation, and formulating a grain reserve allocation scheme for pushing. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments or examples of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or examples. Obviously, the drawings in the following description only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor.

[0046] Figure 1 A flow chart of a grain safety coordination optimization method combined with regional coordination development analysis is provided for an embodiment of the present application.

[0047] Figure 2 A flow chart of a grain safety evaluation index screening method is provided for an embodiment of the present application.

[0048] Figure 3 A flow chart of a regional coordination development level evaluation index screening method is provided for an embodiment of the present application.

[0049] The implementation, functional features and advantages of the present application will be further described with reference to the drawings in combination with the embodiments. DETAILED DESCRIPTION

[0050] In order to more clearly illustrate the technical solutions in the embodiments or examples of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or examples. Obviously, the drawings in the following description only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor.

[0051] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, however, the present application can also be implemented in other ways different from those described herein, therefore, the protection scope of the present application is not limited by the specific embodiments disclosed below.

[0052] Figure 1A grain safety coordination optimization method combined with regional coordinated development analysis is provided for an embodiment of the present application, and a flowchart thereof is shown in the figure.

[0053] As shown in the figure, the present application provides a grain safety coordination optimization method combined with regional coordinated development analysis, comprising: Figure 1

[0054] S102, a plurality of different regional coordinated development level evaluation instances are obtained, each regional coordinated development level evaluation instance is classified, and a regional coordinated development level evaluation instance dataset is constructed;

[0055] S104, based on the regional coordinated development level evaluation instance dataset, a grain safety evaluation index and a regional coordinated development level evaluation index are respectively screened, and a grain safety evaluation dataset and a regional coordinated development level evaluation dataset are constructed;

[0056] Further, in a preferred embodiment of the present application, the plurality of different regional coordinated development level evaluation instances are obtained, each regional coordinated development level evaluation instance is classified, and a regional coordinated development level evaluation instance dataset is constructed, which specifically comprises:

[0057] A plurality of different regional coordinated development level evaluation instances are obtained based on big data retrieval, and the obtained different regional coordinated development level evaluation instances include relevant text descriptions and evaluation parameters of regional coordinated development level evaluation of different regions;

[0058] The relevant text descriptions of each evaluation instance are extracted from the obtained plurality of different regional coordinated development level evaluation instances, and the global context embedding of different words in the relevant text descriptions is obtained by importing a pre-trained BERT model using an attention mechanism;

[0059] The word vectors and keyword vectors are constructed based on the relevant text descriptions of each evaluation instance, the semantic relative distance between different word vectors and keyword vectors is calculated, and the words less than a preset semantic relative distance threshold value are selected as local context embedding;

[0060] The global context embedding and the local context embedding are spliced to obtain semantic features, the evaluation labels of the corresponding regional coordinated development evaluation instances are generated based on the obtained semantic features, and the Euclidean distance values between the evaluation labels are calculated using the Euclidean distance algorithm;

[0061] The calculated Euclidean distance values are compared with a preset Euclidean distance threshold value, the regional coordinated development evaluation instances greater than the preset Euclidean distance threshold value are defined as the same category, and a plurality of evaluation instance categories are obtained;

[0062] ​The evaluation parameters of the regional coordinated development level evaluation of each evaluation example are time-aligned using a dynamic time warping algorithm, and outliers are removed and missing values are supplemented. After data preprocessing, the regional coordinated development level evaluation example dataset is constructed in combination with the divided evaluation example categories.

[0063] It should be noted that first, the network crawler and API interface are integrated with multi-source data (such as government work reports, academic literature, statistical yearbooks, etc.), and evaluation example data covering different regions and different years are collected, each example containing unstructured text description (such as "a certain region realizes grain production and marketing balance through industrial chain coordination") and structured evaluation parameters. After preprocessing such as sentence division, stop word removal, and stem extraction, the text description part is input into the pre-trained BERT model (such as bert-base-uncased), and the context-sensitive embedding vector of each word is extracted through multiple layers of bidirectional Transformer encoder. The self-attention mechanism gives the model the ability to capture long-distance semantic dependencies, for example, in the sentence "the efficiency of fiscal transfer payments between grain-producing and grain-consuming areas has been significantly improved", the correlation between "fiscal transfer payments" and "efficiency" is quantified through attention weights. Subsequently, based on the TF-IDF algorithm, keywords (such as "industrial chain coordination" and "ecological compensation") are extracted from the text to construct keyword vectors, and Word2Vec or GloVe is used to generate word vectors. The semantic distance between word vectors and keyword vectors is calculated by cosine similarity, and words with a similarity higher than a predetermined threshold are selected as local context embeddings. The global and local embedding vectors are aligned in dimension and concatenated into fused semantic features, which are input into a fully connected layer to map to evaluation labels (such as "high coordination degree" and "medium coordination degree"), and the similarity between labels is calculated based on Euclidean distance: if the distance between the label vectors of two examples exceeds a predetermined threshold, they are determined to be different categories, otherwise they are classified into the same category.

[0064] It should be noted that for structured evaluation parameters, the dynamic time warping (DTW) algorithm is used to align parameter sequences of different regions and different time granularities (such as aligning quarterly data to annual scale), eliminate temporal phase differences through local stretching and compression, and remove outliers (such as data points exceeding 3 times the interquartile range) and fill in missing values (such as using the mean value of the same indicator of adjacent regions for interpolation) using box plots and KNN interpolation. Finally, the text-driven semantic category labels and preprocessed parameter sequences are integrated to construct a spatio-temporally aligned, multi-modal fusion regional coordinated development level evaluation example dataset, providing high-quality input for subsequent indicator selection.

[0065] Further, in a preferred embodiment of the present scheme, the collection of each evaluation indicator parameter of each region in the target region, the grain safety prediction and regional coordination degree evaluation, and the development of a grain reserve allocation scheme for pushing, specifically includes:

[0066] The food security evaluation dataset is obtained, a Markov algorithm is introduced, and state transition probabilities of historical evaluation parameters corresponding to each food security evaluation index in the food security evaluation dataset under different time sequence characteristics are calculated by the Markov algorithm;

[0067] A food security prediction model is constructed based on a generative adversarial network, a training dataset is established according to the food security evaluation dataset, the food security prediction model is trained, and the state transition probabilities calculated are introduced in the model training process;

[0068] When the training dataset is input into the food security prediction model, the generator performs sequence reconstruction according to the input training data sequence, and the state transition probability is introduced in the reconstruction process to correct the reconstructed sequence;

[0069] The state transition probability calculated is used to set the discriminator judgment rule, the reconstructed sequence generated by the generator is input into the discriminator, if it meets the judgment rule, the current reconstructed sequence is accepted and output, if it does not meet the judgment rule, reconstruction punishment is performed and the next reconstructed sequence is obtained for judgment;

[0070] The food security prediction model meeting the preset verification standard is obtained through repeated iterative learning, the evaluation index parameters of each region in the target area are collected in real time based on the screened food security indexes, and are input into the trained food security prediction model to obtain food security prediction information.

[0071] It should be noted that first, the food security evaluation dataset is obtained, covering core indexes such as agricultural production conditions, resource environment carrying capacity, and market circulation efficiency. The Markov algorithm focuses on mining the evolution law of the evaluation parameters in the time dimension, and quantifies the dynamic change mode of each index under different time slices by constructing a state transition probability matrix. Specifically, the time sequence observation value of each evaluation index is discretized into a finite state space, the state transition probability between adjacent time steps is calculated using the maximum likelihood estimation method, and a random process model reflecting the internal evolution mechanism of food security is formed. This probability matrix not only describes the time sequence dependence characteristics of a single index, but also reveals the cooperative evolution relationship between multiple indexes through joint transition probability analysis, for example, the conditional transition probability between soil fertility decline and irrigation efficiency improvement may affect the stability of food production capacity.

[0072] Subsequently, a food security prediction model is constructed based on a generative adversarial network, and the state transition probability is deeply integrated into the training process. The input layer of the generator receives the standardized time series training data sequence, the hidden layer captures long-term dependence features through a gated recurrent unit, and a Markov state transition matrix is introduced as a priori constraint in the sequence reconstruction phase. Specifically, in the reconstruction value calculation at each time step, the probability distribution of the previous state is combined for weighted correction to ensure that the generated sequence meets both the data distribution characteristics and the statistical laws of state transition. The discriminator is constructed using a convolutional neural network to extract local time series features, and its discrimination rule is based on the threshold condition set by the stationary distribution theory of Markov chain: for the reconstructed sequence output by the generator, the probability cumulative value of its state transition path is calculated and compared with the state transition expectation obtained from historical statistics. If the path probability is lower than the preset confidence interval, the reconstruction penalty mechanism is triggered to force the generator to adjust the parameters to generate a sequence that better meets the Markov characteristics.

[0073] It is worth mentioning that the model training adopts an adversarial alternating optimization strategy. The generator improves the prediction accuracy by minimizing the joint loss function of reconstruction error and state transition deviation, and the discriminator dynamically adjusts the discrimination threshold to balance the authenticity and diversity of the generated sequence. In the iteration process, as the state transition probability matrix is continuously updated, the system gradually learns the nonlinear evolution law of the food security indicators under the action of complex factors. When the model reaches the preset performance indicators through cross-validation, the deployment stage collects dynamic monitoring data of the target area in real time, processes the features homologous to the training data, and inputs them into the model. The generator predicts the evolution trajectory of the indicators within the future time window based on the current observation sequence and state transition law, and outputs the food security prediction results with confidence evaluation after probability checking by the discriminator, providing quantitative basis for decision support.

[0074] Further, in a preferred embodiment of the present scheme, the collection of evaluation indicator parameters of each region in the target area, the food security prediction and regional coordination degree evaluation, and the development of a food reserve allocation scheme for pushing, further include:

[0075] Obtain a regional coordinated development level evaluation dataset, and construct a regional coordinated development level evaluation system based on a set of regional coordinated development level evaluation indicators and corresponding evaluation parameters in the regional coordinated development level evaluation dataset;

[0076] Monitor the target area to obtain real-time parameters corresponding to each regional coordinated development level evaluation indicator, analyze the regional coordination degree of each region in the target area by combining the constructed regional coordinated development level evaluation system, and obtain regional coordination degree analysis information;

[0077] Obtaining food security prediction information, importing the food security prediction information into a pre-trained Bayesian inference network to obtain a transient posterior distribution, respectively inferring food reserve gap probability of each region in a target area based on the obtained transient posterior, and obtaining gap probability inference information;

[0078] Judging the regional coordination degree analysis information and a preset threshold, marking a region whose regional coordination degree is less than the preset threshold as a risk region and a region whose regional coordination degree is greater than the preset threshold as a safe region in the target area, and generating a food reserve risk heat map in combination with the gap probability inference information;

[0079] Using a regional geographic system to mark a feasible path in the target area and a geographic location of the risk region and the safe region in the food reserve risk heat map to generate a node topology map, wherein the safe region is marked as a starting node and the risk region is marked as a terminal node;

[0080] Introducing an ant colony algorithm to develop a food reserve deployment scheme, setting three optimal paths existing at the terminal node as a node path optimization stop condition, and initializing parameters in the node topology map to obtain a virtual ant colony;

[0081] Driving the virtual ant colony to perform path optimization through a preset path selection probability, iteratively updating until the termination condition is met to output an optimal path of food deployment for each risk region, and generating a food reserve deployment scheme after feasibility evaluation.

[0082] It should be noted that first, the regional coordinated development level evaluation dataset is obtained, the regional coordinated development level evaluation system is constructed based on the regional coordinated development level evaluation index set and the corresponding evaluation parameters in the regional coordinated development level evaluation dataset, the real-time parameters corresponding to each regional coordinated development level evaluation index are obtained, the regional coordination degree of each region in the target region is analyzed combined with the constructed regional coordinated development level evaluation system, and regional coordination degree analysis information is obtained. Subsequently, the food security prediction information is obtained, the food security prediction information is input into the pre-trained Bayesian inference network to obtain the transient posterior distribution, and the food reserve gap probability of each region in the target region is inferred based on the obtained transient posterior, and gap probability inference information is obtained. The pre-trained Bayesian network takes food production capacity fluctuation rate and consumption demand growth rate as parent nodes, and storage capacity turnover rate and cross-border transportation capacity as child nodes, and a multi-layer network structure containing a conditional probability table is constructed. When the real-time prediction data is input, the Markov Chain Monte Carlo (MCMC) sampling method is used to quickly update the posterior distribution, and the confidence interval of the reserve gap probability is derived through KL divergence calculation. The regional coordination degree analysis information is judged with the preset threshold, the regions in the target region whose regional coordination degree is less than the preset threshold are marked as risk regions, and the regions whose regional coordination degree is greater than the preset threshold are marked as safe regions, and the food reserve risk heat map is generated combined with the gap probability inference information. The discrete risk points are spatially smoothed by using the kernel density estimation method, and a risk visualization layer with gradual characteristics is constructed through HSV color space mapping. The regional geographic system is used to mark the feasible path in the food reserve risk heat map and the geographic location of the risk region and the safe region to generate a node topology graph, wherein the safe region is marked as the starting node and the risk region is marked as the terminal node. The construction of the node topology graph introduces a multi-layer graph structure, in addition to the basic road network data, three special edges are additionally defined: emergency access (weighted based on historical disaster relief path data), potential alternative route (considering the accessibility analysis of terrain undulation and bridge carrying capacity), and virtual connection line (used to break through the administrative boundary constraint of cross-regional coordination path).

[0083] Further, the ant colony algorithm is introduced to develop the food reserve allocation scheme. The movement decision of the ant individual introduces the coordination degree attenuation factor to correct the path weight, so that the transportation route in the high coordination degree region obtains the selection preference. After each iteration, the non-inferior solution is screened through the Pareto frontier analysis, and when each risk region obtains three candidate paths that meet the multi-objective constraints (minimum transportation cost, highest path reliability, and optimal regional coordination impact), the termination condition is triggered. In the feasibility evaluation process, if the number of terminal nodes connected to a starting node exceeds the preset number, the food reserve prediction quantity of the target starting node is calculated to determine whether it can bear the corresponding allocation quantity, thereby determining whether it is feasible. If it is not feasible, the corresponding connected terminal node is removed and the removed node is re-optimized, thereby obtaining the final food reserve allocation scheme.

[0084] Figure 2 A grain safety evaluation index screening method flow chart is provided for an embodiment of the present application;

[0085] As shown in Figure 2 , the present application provides a grain safety evaluation index screening method flow chart, comprising:

[0086] S202, obtaining a regional coordinated development level evaluation instance data set, extracting evaluation indexes related to grain safety and corresponding evaluation parameters from the regional coordinated development level evaluation instance data set, marking the selected indexes as first screening indexes, and generating a preliminary screening index data set;

[0087] S204, importing the preliminary screening index data set as input into a LASSO regression model for preliminary dimension reduction, presetting a regularization strength parameter, using L1 regularization constraint to impose sparsification penalty on the regression coefficient, and screening the preliminary screening indexes with non-zero regression coefficients to generate a secondary screening index data set;

[0088] S206, wherein the regularization strength parameter is obtained by dividing the preliminary screening index data set into several training subsets and validation subsets, using the validation subset to construct a preset logarithmic space, calculating the regularization strength parameter in the preset logarithmic space, and selecting the regularization strength parameter with the smallest validation error as the optimal solution;

[0089] S208, introducing a random forest algorithm to further screen the secondary screening indexes, initializing the secondary screening index data set and the evaluation parameters corresponding to the secondary screening indexes as initial input data of the algorithm, and setting initial weights for each secondary screening index;

[0090] S210, generating an index feature set through the initial input data to construct a decision tree, performing probability sampling based on the preset initial weights to obtain a plurality of sampling samples, constructing a hierarchical structure of the decision tree through the obtained sampling samples, and taking the Gini index as the node splitting basis, and outputting the random forest after splitting is completed;

[0091] S212, extracting out-of-bag samples of each tree after outputting the random forest, performing random disturbance on the random forest based on the extracted out-of-bag samples and calculating the error increment, obtaining the error increment of the target index on all trees after traversing all trees, and taking the mean value as the global importance score;

[0092] S214, generating an importance score ranking table based on the global importance score of each index, outputting the indexes within the preset range as the grain safety evaluation indexes through the importance score ranking table, and combining the corresponding historical evaluation parameters to form a grain safety evaluation data set.

[0093] It should be noted that during the indicator screening process for the regional coordinated development level evaluation dataset, evaluation indicators directly or indirectly related to food security (such as "grain reserve coverage rate," "grain logistics timeliness," and "arable land non-agriculturalization rate") were first extracted from the dataset and calibrated as a preliminary screening indicator set. These indicators were then associated with corresponding historical evaluation parameters (such as the annual value of reserve coverage rate and the monthly average of logistics timeliness), forming a preliminary screening dataset containing multidimensional indicators and spatiotemporal parameters. Subsequently, this dataset was input into the LASSO regression model for preliminary dimensionality reduction. By setting the regularization intensity parameter to control the sparsity of the model, a coordinate descent algorithm was used to solve the linear regression problem with L1 regularization constraints, compressing some regression coefficients to zero. This allowed the selection of non-zero coefficient indicators that significantly impacted the target variables of regional coordination (such as the "rural-urban income gap reduction rate").

[0094] To further optimize indicator importance, a random forest algorithm was introduced. This algorithm takes the secondary screening indicator set and its parameters as input and constructs a decision tree ensemble model. Each tree generates a training subset through bootstrap sampling, and the optimal split point is selected at the node splitting point, minimizing the Gini index. After model training, feature importance is calculated based on out-of-bag (OOB) samples. The values ​​of the target indicator (e.g., "cultivated land quality") in the OOB samples of each tree are randomly perturbed (e.g., permutation or noise injection). The predictions are then re-calculated and the mean squared error increment is calculated. The mean squared error increment across all trees is then averaged as the global importance score. For example, the error increment for "grain reserve coverage" is 0.15 (high importance), while the error increment for "agricultural science and technology investment intensity" is only 0.03 (low importance). Finally, an indicator importance ranking table is generated. Core indicators, such as reserve coverage, logistics timeliness, cultivated land quality, total grain output, grain yield per unit area, and grain sown area, are selected based on preset thresholds (e.g., importance > 0.1). These indicators, combined with their historical parameters, are used to construct a food security assessment dataset. By quantitatively characterizing food production and circulation capacity, accurate and representative food evaluation indicators can be screened.

[0095] Figure 3 A flow chart of a method for screening regional coordinated development level evaluation indicators provided by one embodiment of the present invention;

[0096] like Figure 3 As shown, the present invention provides a flow chart of a method for screening regional coordinated development level evaluation indicators, including:

[0097] S302, obtaining a regional coordinated development level evaluation instance dataset, extracting evaluation indicators related to regional coordinated development from the regional coordinated development level evaluation instance dataset, and marking them as second screening evaluation indicators;

[0098] S304, introducing principal component analysis to the selected second screening evaluation index for preliminary dimension reduction, taking the second screening evaluation index as the input data to calculate the covariance matrix, performing eigenvalue decomposition on the calculated covariance matrix to obtain eigenvalues and corresponding eigenvectors, obtaining a plurality of principal components;

[0099] S306, sorting the principal components based on the eigenvalues of each principal component, selecting the principal components within the preset range as target principal components through the sorting results, obtaining a principal component index set; taking the principal component index set as input, and using independent principal component analysis to screen independent principal component indexes;

[0100] S308, centering the input principal component index set, initializing the generation of a separation matrix based on the processed principal component index set, and taking the orthogonal basis vector of the principal component direction as the initial estimate to iteratively optimize the separation matrix;

[0101] S310, continuously iteratively optimizing and analyzing until a preset stopping criterion is reached to output a final separation matrix, and obtaining independent principal component indexes by inverse transformation based on the final separation matrix, as regional coordinated development level evaluation indexes, and using a regional coordinated development level evaluation example data set to obtain historical evaluation parameters of the corresponding indexes to form a regional coordinated development level evaluation data set.

[0102] It should be noted that first, the regional coordinated development level evaluation example data set extracts specific indexes in the dimensions of economy, society, environment, infrastructure, etc., such as regional gross domestic product (GDP) annual growth rate, industrial added value above designated size, per capita disposable income of urban residents, per capita disposable income ratio of rural residents, PM2.5 annual concentration reduction rate, railway freight volume annual growth rate, night light index (NTL), 5G base station coverage rate, etc., which are calibrated as the second screening evaluation index set. Through principal component analysis (PCA) to preliminarily reduce the dimension of the index: calculate the covariance matrix of the standardized data, then perform eigenvalue decomposition on the covariance matrix, extract the eigenvalues and eigenvectors, and select the first k principal components with cumulative variance contribution rate ≥85% after sorting the eigenvalues. For example, the first principal component has an eigenvalue of 8.2, which is composed of high load indexes such as GDP growth rate, industrial added value, and night light index, reflecting economic activity; the second principal component has an eigenvalue of 5.1, which is dominated by PM2.5 concentration reduction rate and ecological protection and development intensity, representing environmental governance level; the third principal component has an eigenvalue of 3.3, which is composed of rural resident income ratio and 5G base station coverage rate, reflecting urban and rural digital development.

[0103] Subsequently, the scores of the first three principal components were input into independent component analysis (ICA). The data was first centered (mean zero), and the separating matrix was initialized with random orthogonal basis vectors. The FastICA algorithm was used to maximize non-Gaussianity. The separating matrix was iteratively optimized until convergence, ultimately yielding statistically independent components. For example, the independent component "Economy-Energy Synergy" was driven by a nonlinear combination of "GDP Growth Rate" and "Nighttime Light Index." An inverse transformation was used to map the independent components back to the original indicator space. Specific indicators with absolute loadings > 0.6 were selected, ultimately retaining core indicators such as regional GDP growth rate, industrial added value above designated size, nighttime light index, and railway freight volume. Redundant indicators (e.g., "General Public Budget Revenue" was filtered out due to its strong correlation with GDP) were removed. A dataset for evaluating the level of regional coordinated development was constructed by combining historical parameters. This two-stage screening process, using PCA to extract common trends and ICA to analyze independent drivers, not only preserves the economic explanatory power of key indicators (e.g., the nighttime light index intuitively reflects the intensity of economic activity) but also highlights the independent influences of environmental and urban-rural dimensions, providing a high-information quantitative basis for the level of regional coordinated development.

[0104] Furthermore, in a preferred embodiment of the present invention, the association rules between food security evaluation indicators and regional coordinated development level evaluation indicators are analyzed, and a cross-dimensional association rule library is constructed, specifically including:

[0105] Obtaining a food security evaluation dataset and a regional coordinated development level evaluation dataset, and discretizing historical evaluation parameters associated with food security evaluation indicators stored in the food security evaluation dataset into a number of food security states;

[0106] Discretizing historical evaluation parameters associated with the regional coordinated development level evaluation indicators stored in the regional coordinated development level evaluation data set into a plurality of regional coordination states, and constructing a state combination transaction set by combining the discretized food security state and the regional coordination state;

[0107] The state combination transaction set is used as input to perform cross-dimensional frequent item set mining using the Apriori algorithm, and the occurrence frequency of all single indicator states in the state combination transaction set is scanned and counted as the support of the corresponding indicator state;

[0108] Select the indicator state greater than the preset minimum support to construct the initial frequent itemset, generate candidate frequent itemsets based on the initial frequent itemset, introduce time continuity constraints to optimize the candidate frequent itemset, eliminate candidates that do not meet the time continuity constraints, and output the final frequent itemset through repeated iterative mining and optimization;

[0109] The final frequent item set is converted into a rule-instance matrix, each rule corresponds to a Boolean feature, and a sparse principal component analysis is used to reduce the dimension of each rule-instance matrix, and the principal component scores representing the rule coordination mode are extracted;

[0110] The principal component scores are used as input layer features, and the regional coordination state is used as output layer features, the radial basis function hidden layer is used to fit the nonlinear relationship between the rule cluster and the coordination level, and finally the asymmetric association rule dataset is output, and the cross-dimensional association rule library is constructed combined with the final frequent item set.

[0111] It should be noted that first, the historical data of the two evaluation systems of food security and regional coordinated development need to be integrated, the continuous parameters (such as food reserve coverage rate, farmland quality index, etc.) in the food security evaluation dataset are discretized and graded, and are converted into “high, medium, low” three-level state labels according to industry standards or quantile division method, and the evaluation parameters (such as economic coordination index, ecological compensation compliance rate, etc.) in the regional coordination dataset are discretized into “excellent, good, poor” levels according to the policy target threshold or historical distribution, and a state combination transaction set containing space-time labels (year, region code) is formed, each transaction corresponds to the discrete state combination of each index in a specific space-time unit.

[0112] Subsequently, based on the state combination transaction set, the Apriori algorithm is used for cross-dimensional frequent item set mining: the first round of scanning the transaction set and counting the support of single index state (such as “high farmland quality” frequency is 18%), and screening the initial frequent item set exceeding the preset minimum support; then generate candidate frequent item sets through layer-by-layer connection, and impose time continuity constraints, requiring candidate item sets to appear stably in at least 3 consecutive time windows, and output cross-dimensional frequent item sets (such as “reserve coverage rate medium-emergency coordination good” support 0.18). The frequent item set is converted into a rule-instance matrix, the matrix row represents a regional instance, the column corresponds to a rule, and the element is a Boolean value (1 if the rule is met, otherwise 0), the high-dimensional matrix is reduced by sparse principal component analysis (SPCA), and the principal components with cumulative variance contribution rate ≥80% are extracted, each principal component represents a rule coordination mode. The principal component scores are input into the generalized regression neural network (GRNN), the radial basis function hidden layer is used to fit the nonlinear relationship between the rule cluster intensity and the regional coordination state, and the asymmetric rule and intensity are output (such as “when the production- circulation coordination cluster intensity >0.7, the probability of coordination level improving to excellent is 83%”). Finally, a cross-dimensional association rule library with space-time labels is constructed, forming a full-process closed loop from data discretization to nonlinear association modeling.

[0113] Further, in a preferred embodiment of the present application, a regional coordination state comprehensive evaluation model is constructed through the cross-dimensional association rule library, specifically comprising:

[0114] Obtain a cross-dimension association rule library, define a grain security index state as a first node, a regional coordination state as a second node, and establish a directed edge description connecting the first node and the second node according to a plurality of rules stored in the cross-dimension association rule library to build a grain security state-regional coordination state topology graph;

[0115] Obtain an adjacency matrix through the grain security state-regional coordination state topology graph, construct a regional coordination state comprehensive evaluation model by using a graph neural network, and train the model by using the obtained adjacency matrix;

[0116] When the output comprehensive evaluation result meets a preset standard, output the trained regional coordination state comprehensive evaluation model.

[0117] It should be noted that firstly, all association rules of "grain security index_state X → regional coordination index_state Y" are extracted from the cross-dimension association rule library, each grain security index state (such as "reserve coverage rate medium") is defined as a first type of node in the topology graph, a regional coordination index state (such as "emergency coordination good") and a corresponding coordination score (such as 0.72) are defined as a second type of node, and a directed edge weight is constructed based on a rule confidence. For example, the confidence 0.82 and the causal intensity 0.75 of the rule "high quality of arable land → optimal ecological compensation" are converted into a directed edge from the high quality of arable land node to the optimal ecological compensation node, and the weight of the edge is calculated by the support and confidence of the rule. The generation of the adjacency matrix needs to integrate the weights of the association rules between all nodes, and if two nodes are not directly connected by a rule, the weight is 0, and finally a sparse weighted adjacency matrix is formed. For complex association rules (such as multi-index combination rules), a virtual node is introduced to represent the composite state, and a time sequence effect is processed by an edge weight decay factor. The finally generated topology graph contains a dynamic adjacency matrix, and the matrix elements are filled by the weighted values of the directed edges between nodes, and the weight between nodes not directly connected is set to 0.

[0118] Subsequently, based on the adjacency matrix and node features (such as the historical frequency of food security state and the time sequence average of regional coordination score), a graph neural network is used to construct a regional coordination state comprehensive evaluation model. The input layer receives the node feature vector, aggregates the neighbor node information through multi-layer graph convolution operation, the hidden layer captures the complex interaction relationship through the nonlinear activation function ReLU, and the output layer is mapped to the regional coordination comprehensive score. During the training process, the historical coordination score true value is used as the supervision signal, the mean square error (MSE) loss function is combined with L1 regularization to prevent overfitting, and the network parameters are optimized through back propagation. In the model verification stage, the preset standards include the average absolute error of the predicted score and the actual value of the test set, the causal intensity retention rate (> 85%) of the key rule path, etc. When the model output meets the threshold value, the trained GNN model and the topological structure visualization map are output, realizing the end-to-end mapping from single indicator state association to comprehensive score, for example, inputting the "high quality of cultivated land" state node, the model can output the score improvement prediction value along the "high quality of cultivated land → ecological compensation_ optimal → regional coordination total score" path, providing an interpretable path deduction for policy intervention, and providing differentiated regulation strategy guidance for different regions.

[0119] Further, in the food security coordination optimization method provided by the present application combined with regional coordination development analysis, further comprising:

[0120] Based on the food security evaluation index, the food security related data of the target region is obtained, and the regional coordination development related data of the target region is obtained through the regional coordination development level evaluation index, which is input into the regional coordination state comprehensive evaluation model for analysis;

[0121] The food security related data input into the model is used to evaluate the food security of the target region to obtain the food security evaluation result, and the target region food security evaluation portrait is generated combined with the food security related data;

[0122] Cosine similarity is used as the basis for rule matching, and cross-dimensional association rule matching is performed in the preset search space according to the target region food security evaluation portrait, obtaining a plurality of matched cross-dimensional association rules;

[0123] The historical evaluation data of the regional coordination development level of each cross-dimensional association rule is obtained through the matched cross-dimensional association rule, and the difference value is obtained by difference calculation between the regional coordination development related data of the target region;

[0124] The calculated difference value is compared with the preset evaluation rule, and the regional coordination development level evaluation result of the target region is output according to the judgment result, and the regional coordination state comprehensive evaluation information is generated combined with the food security evaluation result.

[0125] It should be noted that in the regional coordination state comprehensive evaluation process, first, based on the screened food security evaluation indexes (such as "food reserve coverage rate", "cultivated land quality index", "cold chain logistics timeliness"), the historical and real-time data of the target region (such as the reserve coverage rate of a certain region in 2023 is 75%, the cultivated land quality index is 0.82, and the average timeliness of cold chain logistics is 24 hours) are obtained, and the data corresponding to the regional coordinated development level evaluation indexes (such as "GDP growth rate", "night light index", "PM2.5 concentration reduction rate") (such as GDP growth of 6.2%, night light index increased by 8% compared with last year, and PM2.5 concentration decreased by 5%) are extracted, and the two types of data are input into the pre-trained regional coordination state comprehensive evaluation model. The model calculates the food security score of the target region through the food security data, and generates a multi-dimensional portrait. Subsequently, based on the portrait feature vector, cosine similarity matching is performed in the cross-dimension association rule base. Each rule in the rule base includes the mapping relationship between the food security state combination and the regional coordination state, and the rules with similarity exceeding the threshold are selected as the effective matching rules. The difference between the historical regional coordination data associated with the matching rules (such as the average GDP growth rate of the region corresponding to rule A is 6.5%) and the current data of the target region (such as the actual GDP growth rate is 6.2%) is calculated, and the preset evaluation rule (such as the absolute value of the difference is less than 0.5%) is used to judge the deviation degree of the coordination level. If the difference exceeds the threshold, it is marked (such as "GDP growth rate does not meet the expected value of the associated rule"). Finally, the food security score, coordination level deviation analysis and rule matching conclusion are integrated to generate a comprehensive evaluation report (such as the coordination level "good", and the GDP growth rate needs to be improved), so as to show the short board of the region and provide a quantitative basis for dynamic optimization, and provide guidance for regional coordinated development.

[0126] The application further provides a computer readable storage medium, characterized in that the computer readable storage medium comprises a food security coordination optimization method program combined with regional coordinated development analysis.

[0127] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only illustrative. For example, the division of the units is only a logical functional division, and actual implementation can have another division mode. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed components can be through some interface, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0128] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; they can be located in one place or distributed on multiple network units; and some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0129] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be realized in the form of hardware or in the form of hardware plus software functional unit.

[0130] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware, and the aforementioned program can be stored in a computer readable storage medium, and the program executes the steps including the above-mentioned method embodiments when executed; and the aforementioned storage medium includes mobile storage device, read-only memory (ROM), random access memory (RAM), magnetic disc or optical disc and various storage program codes.

[0131] Alternatively, the integrated unit of the present application, if implemented in the form of a software function module and sold or used as an independent product, can also be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present application can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the embodiments of the present application. The aforementioned storage medium includes mobile storage device, ROM, RAM, magnetic disc or optical disc and various storage program codes.

[0132] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A food security coordination optimization method combined with regional coordinated development analysis, characterized by: include: Obtain several different regional coordinated development level evaluation instances, classify the regional coordinated development level evaluation instances into categories, and construct a regional coordinated development level evaluation instance dataset; Based on the regional coordinated development level evaluation example dataset, food security evaluation indicators and regional coordinated development level evaluation indicators are respectively selected, and a food security evaluation dataset and a regional coordinated development level evaluation dataset are constructed; Based on the selected food security indicators and regional coordinated development level evaluation indicators, the evaluation indicator parameters of each region in the target area are collected to conduct food security forecasts and regional coordination evaluations, and a food reserve allocation plan is formulated and pushed; The aforementioned collection of evaluation index parameters for each region in the target area, the conduct of food security forecasts and regional coordination evaluations, and the formulation and delivery of food reserve allocation plans specifically include: Obtain a food security evaluation data set, introduce a Markov algorithm, and calculate the state transition probability of historical evaluation parameters corresponding to each food security evaluation indicator in the food security evaluation data set under different time series characteristics through the Markov algorithm; Constructing a food security prediction model based on a generative adversarial network, establishing a training dataset based on a food security evaluation dataset to train the food security prediction model, and introducing the calculated state transition probability into the model training process; When the training data set is input into the food security prediction model, the generator reconstructs the sequence according to the input training data sequence, and introduces the state transition probability in the reconstruction process to correct the reconstructed sequence; The discriminator’s discrimination rule is set based on the calculated state transition probability. The reconstructed sequence generated by the generator is input into the discriminator. If it meets the discrimination rule, the current reconstructed sequence is accepted and output. If it does not meet the discrimination rule, a reconstruction penalty is applied and the next reconstructed sequence is obtained for discrimination. Through repeated iterative learning, a food security prediction model that meets the preset verification standards is obtained. Based on the screened food security indicators, evaluation index parameters of various regions in the target area are collected in real time and input into the trained food security prediction model to obtain food security prediction information. The aforementioned collection of evaluation index parameters of various regions in the target area, conducting food security forecasts and regional coordination evaluations, and formulating and delivering food reserve allocation plans also includes: Acquire a regional coordinated development level evaluation data set, and construct a regional coordinated development level evaluation system based on a regional coordinated development level evaluation indicator set and corresponding evaluation parameters in the regional coordinated development level evaluation data set; Monitor the target area to obtain the real-time parameters corresponding to the evaluation indicators of the coordinated development level of each region. Combined with the constructed regional coordinated development level evaluation system, analyze the regional coordination level of each region in the target area to obtain regional coordination level analysis information; Obtaining food security forecast information, importing the food security forecast information into a pre-trained Bayesian inference network to obtain a transient posterior distribution, and inferring the probability of a food reserve gap in each region within the target area based on the obtained transient posteriori to obtain gap probability inference information; The regional coordination degree analysis information is compared with a preset threshold, and areas within the target area where the regional coordination degree is less than the preset threshold are marked as risk areas, and areas where the regional coordination degree is greater than the preset threshold are marked as safe areas. The gap probability inference information is combined to generate a food reserve risk heat map; Using the regional geographic system, feasible paths within the target area and the geographical locations of risk areas and safe areas are calibrated in the food reserve risk heat map to generate a node topology map, wherein the safe area is calibrated as the starting node and the risk area is the ending node; The ant colony algorithm is introduced to formulate the food reserve allocation plan. The existence of three optimal paths at the end node is set as the node path optimization stop condition. The parameters are initialized in the node topology graph to obtain a virtual ant colony. The virtual ant colony is driven by the preset path selection probability to perform path optimization, and iterative updates are performed until the termination conditions are met to output the optimal food allocation path for each risk area. After a feasibility assessment, a food reserve allocation plan is generated.

2. A food security coordination optimization method combined with regional coordinated development analysis according to claim 1, characterized in that: The step of obtaining a number of different regional coordinated development level evaluation instances, classifying the regional coordinated development level evaluation instances into categories, and constructing a regional coordinated development level evaluation instance dataset specifically includes: Acquire a number of evaluation instances of coordinated development levels of different regions based on big data retrieval, wherein the obtained evaluation instances of coordinated development levels of different regions include relevant text descriptions and evaluation parameters of the evaluation of coordinated development levels of different regions; Extract relevant text descriptions of each evaluation instance from several different regional coordinated development level evaluation instances, import them into the pre-trained BERT model, and use the attention mechanism to obtain the global context embedding of different words in the relevant text descriptions; Build word vectors and keyword vectors based on the relevant text description of each evaluation instance, calculate the semantic relative distance between different word vectors and keyword vectors, and select words with a semantic relative distance less than a preset threshold as local context embeddings; The global context embedding and the local context embedding are concatenated to obtain semantic features. Based on the obtained semantic features, evaluation labels for the corresponding regional coordinated development evaluation instances are generated. The Euclidean distance algorithm is used to calculate the Euclidean distance between the evaluation labels. The calculated Euclidean distance value is compared with a preset Euclidean distance threshold, and regional coordinated development evaluation instances with a value greater than the preset Euclidean distance threshold are defined as the same category, thereby obtaining several evaluation instance categories; The dynamic time warping algorithm is used to align the evaluation parameters of each evaluation instance in the regional coordinated development level evaluation, and outliers are eliminated and missing values ​​are supplemented. After completing data preprocessing, a regional coordinated development level evaluation instance dataset is constructed in combination with the divided evaluation instance categories.

3. The method for food security coordination optimization combined with regional coordinated development analysis according to claim 1 is characterized in that: The step of selecting food security evaluation indicators and regional coordinated development level evaluation indicators based on the regional coordinated development level evaluation example dataset, and constructing a food security evaluation dataset and a regional coordinated development level evaluation dataset, specifically includes: Obtaining a regional coordinated development level evaluation instance dataset, extracting evaluation indicators related to food security and corresponding evaluation parameters from the regional coordinated development level evaluation instance dataset, calibrating the selected indicators as first screening indicators, and generating a primary screening indicator dataset; The primary screening indicator dataset is imported as input into the LASSO regression model for preliminary dimensionality reduction, the regularization strength parameter is preset, the L1 regularization constraint is used to impose a sparsity penalty on the regression coefficient, and the primary screening indicators with non-zero regression coefficients are screened to generate a secondary screening indicator dataset; The regularization strength parameter is obtained by dividing the initial screening index data set into several training subsets and validation subsets, constructing a preset logarithmic space using the validation subsets, traversing and calculating the regularization strength parameter in the preset logarithmic space, and selecting the regularization strength parameter with the smallest validation error as the optimal solution; A random forest algorithm is introduced to re-screen the secondary screening indicators, and the secondary screening indicator data set and the evaluation parameters corresponding to the secondary screening indicators are used as the initial input data of the algorithm to initialize and set the initial weight for each secondary screening indicator; The decision tree is constructed by generating an indicator feature set through the initial input data. A number of sample samples are obtained by probability sampling based on the preset initial weights. The hierarchical structure of the decision tree is constructed through the obtained sample samples and the Gini index is used as the basis for node segmentation. After the segmentation is completed, the random forest is output; After outputting the random forest, extract the out-of-bag samples of each tree, randomly perturb the random forest based on the extracted out-of-bag samples and calculate the error increment. After traversing all trees in the forest, obtain the error increment of the target indicator on all trees and take the average as the global importance score; Based on the global importance score of each indicator, an importance score ranking table is generated. The indicators within the preset range are output through the importance score ranking table as food security evaluation indicators and combined with the corresponding historical evaluation parameters to form a food security evaluation dataset.

4. The method for food security coordination optimization combined with regional coordinated development analysis according to claim 1 is characterized in that: The method of screening food security evaluation indicators and regional coordinated development level evaluation indicators based on the regional coordinated development level evaluation example dataset, and constructing a food security evaluation dataset and a regional coordinated development level evaluation dataset, further includes: Obtaining a regional coordinated development level evaluation instance dataset, extracting evaluation indicators related to regional coordinated development from the regional coordinated development level evaluation instance dataset, and calibrating them as second screening evaluation indicators; The principal component analysis method is introduced to perform preliminary dimensionality reduction on the selected second screening evaluation index. The second screening evaluation index is used as input data to calculate the covariance matrix. The calculated covariance matrix is ​​subjected to eigenvalue decomposition to obtain eigenvalues ​​and corresponding eigenvectors, and several principal components are obtained. Sorting the principal components based on the eigenvalues ​​of the principal components, selecting the principal components within a preset range as target principal components based on the sorting results, and obtaining a principal component index set; using the principal component index set as input, screening the independent principal component indices using the independent principal component analysis method; The input principal component index set is centralized, and the separation matrix is ​​generated based on the initialization of the processed principal component index set. The orthogonal basis vectors of the principal component direction are used as the initial estimate to iteratively optimize the separation matrix. The final separation matrix is ​​output through continuous iterative optimization analysis until the preset stopping criterion is reached. The independent principal component index is obtained by inverse transformation according to the final separation matrix, which is used as the evaluation index of the regional coordinated development level. The historical evaluation parameters of the corresponding indicators are obtained using the regional coordinated development level evaluation instance data set to form the regional coordinated development level evaluation data set.

5. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a food security coordination optimization method program combined with regional coordinated development analysis, which is suitable for implementing the steps of the food security coordination optimization method combined with regional coordinated development analysis as described in any one of claims 1 to 4 when the food security coordination optimization method program combined with regional coordinated development analysis is executed by a processor.

Images