Method and device for evaluating efficiency of overall land remediation

Abstract

The application relates to the technical field of land resource management, and discloses a method and device for evaluating the efficiency of global land comprehensive regulation. The method for evaluating the efficiency of global land comprehensive regulation provided by the application realizes precise modeling of nonlinear coupling effect by constructing an evaluation index system and combining a neural network efficiency evaluation model, and after the result is discretized, a decision tree model is used for multidimensional interpretation, so that the problem that the evaluation model precision and result interpretability are difficult to be considered in the prior art is effectively solved, and the scientificity and practical guiding value of the efficiency evaluation of global land comprehensive regulation are improved.

Description

Technical Field

[0001] This application relates to the technical field of effectiveness evaluation methods for comprehensive land consolidation, specifically to an effectiveness evaluation method and apparatus for comprehensive land consolidation. Background Technology

[0002] The evaluation of the effectiveness of comprehensive land consolidation plays a decisive role in the scientific formulation of land use strategies. Current mainstream evaluation methods struggle to achieve breakthroughs. Some methods employ linear weighting mechanisms, which fail to effectively characterize the nonlinear coupling effects of multi-dimensional indicators during the consolidation process. For example, the dynamic correlation between increased farmland contiguousness and changes in ecological restoration area is simplified in the model. While some methods incorporate fuzzy mathematics theory, they remain constrained by linear superposition frameworks, making it difficult to capture the complex interaction between newly added high-standard farmland area and industrial investment intensity. Some methods use neural networks for evaluation, but the evaluation accuracy cannot meet practical needs. These shortcomings prevent the existing technical framework from achieving an effective balance among the three key dimensions of model accuracy, parameter optimization robustness, and result interpretability, hindering the guiding value of evaluation results for consolidation practices. Summary of the Invention

[0003] This application provides a method and apparatus for evaluating the effectiveness of comprehensive land consolidation across an entire region, in order to address the problem of difficulty in balancing the accuracy and interpretability of the evaluation model for comprehensive land consolidation across an entire region.

[0004] Firstly, this application provides a method for evaluating the effectiveness of comprehensive land consolidation across an entire region. The method includes: constructing an evaluation index system and determining the weight of each evaluation index in the system, wherein the evaluation index system includes multiple primary indicators, and each primary indicator includes multiple secondary indicators; evaluating the effectiveness of the input evaluation indicators based on a trained neural network effectiveness evaluation model to obtain effectiveness evaluation results, wherein the neural network effectiveness evaluation model includes a three-layer backpropagation neural network; discretizing the effectiveness evaluation results into multiple level categories and constructing a decision tree model using the evaluation indicators in the evaluation index system as splitting attributes; and scoring and interpreting the effectiveness evaluation results based on the decision tree model.

[0005] Based on the aforementioned technical means, this application constructs a multi-level evaluation index system and combines it with a trained three-layer backpropagation neural network for performance evaluation, effectively capturing the nonlinear coupling relationship between various indicators in the comprehensive land consolidation, thereby improving the accuracy and robustness of the evaluation.

[0006] In one optional implementation, determining the weight of each evaluation indicator in the evaluation indicator system includes: classifying the remediation project into multiple remediation models based on the content of the remediation implementation, wherein the remediation models include farmland improvement, village remediation, ecological restoration, and industry introduction, and the same remediation project includes at least one remediation model; determining the corresponding weight vector in the evaluation indicator system according to different remediation models; when there is no corresponding evaluation indicator in the remediation model, marking the corresponding evaluation indicator as inapplicable in the evaluation indicator system, and allocating the weight corresponding to the evaluation indicator to other evaluation indicators of the same level.

[0007] Based on the aforementioned technical means, this application achieves deep coupling between the evaluation model and the actual business scenario of the project by refining the evaluation index system and introducing a dynamic weight allocation mechanism based on the rectification model.

[0008] In one optional implementation, the corresponding weight vector is determined in the evaluation index system according to different remediation models, including: constructing a judgment matrix based on the evaluation indicators in the evaluation index system; calculating the weight vector of the constructed judgment matrix through expert opinions and performing a consistency check to obtain the hierarchical analysis weights; calculating the information entropy and difference coefficient of each evaluation indicator in the evaluation index system based on the data information recorded in the remediation project, and determining the entropy weight based on the information entropy and difference coefficient, wherein the difference coefficient is used to characterize the sample discrimination ability of the evaluation indicator, and the sample discrimination ability is positively correlated with the corresponding assigned weight; and weighting and fusing the hierarchical analysis weights and entropy weights according to a preset fusion coefficient to obtain the weight vector.

[0009] Based on the aforementioned technical means, this application effectively addresses the limitations of a single weighting method in evaluating the effectiveness of comprehensive land consolidation across the entire region by employing a combined weighting strategy that combines the analytic hierarchy process (AHP) with the entropy method.

[0010] In one optional implementation, the neural network performance evaluation model is constructed based on a three-layer backpropagation neural network, wherein the number of input nodes in the neural network performance evaluation model is equal to the number of evaluation metrics, and the number of hidden layer nodes in the neural network performance evaluation model is determined by the following expression: ,in The number of nodes in the input layer. This represents the number of nodes in the output layer. To adjust the parameters, the hidden layer activation function of the neural network performance evaluation model is: The training of the neural network performance evaluation model includes: using a genetic algorithm to globally optimize the initial weights and thresholds of the three-layer backpropagation neural network to obtain the trained neural network performance evaluation model.

[0011] Based on the aforementioned technical means, this solution enables the evaluation model to exhibit stronger robustness when facing data from different land consolidation projects, thereby improving the accuracy and stability of the comprehensive land consolidation effectiveness evaluation.

[0012] In one optional implementation, a genetic algorithm is used to globally optimize the initial weights and thresholds of the three-layer backpropagation neural network to obtain a trained neural network performance evaluation model. This includes: using real-number encoding, concatenating the weights and thresholds of the three-layer backpropagation neural network into a one-dimensional vector as a chromosome, where the chromosome length is n. i ×n h +n h ×n0+n h +n0, where n i n is the number of nodes in the input layer, n0 is the number of nodes in the output layer, and n h The number of hidden layer nodes is denoted by ; the inverse of the mean squared error of the training set is taken as the fitness function; after configuring the population size, maximum number of generations, crossover probability, mutation probability, and elite retention ratio, convergent training is performed. After convergence, the weights and thresholds corresponding to the optimal individuals representing the evaluation index are used as the initial parameters of the three-layer backpropagation neural network; the parameters are adjusted based on the backpropagation algorithm to obtain the trained neural network performance evaluation model.

[0013] Based on the above technical means, not only is the stability of the model improved, reducing the fluctuation of its training results under different initial conditions, but it also provides a more reliable and accurate basis for the subsequent discretization of performance evaluation results and the construction of decision trees, thereby enhancing the robustness and interpretability of the entire evaluation method.

[0014] In one optional implementation, the performance evaluation results include predicted scores. The performance evaluation results are discretized into multiple level categories, and a decision tree model is constructed using evaluation indicators from the evaluation index system as splitting attributes. This includes: dividing the predicted scores into multiple categories according to level thresholds; and using information gain as the node splitting index, where information gain characterizes the reduction in information entropy before and after level division. Information gain includes Gain(D,A) = H(D) - H(D|A), where... For the training sample set, For information entropy, express For feature attributes of the training set Information gain, A is the candidate feature attribute; starting from the root node, traverse all possible thresholds of evaluation indicators for each node, and select the attribute with the largest information gain and the corresponding threshold combination for splitting, recursively constructing subtrees, where the stopping conditions of recursion include the number of samples in the node being lower than the preset minimum value, the tree depth reaching the preset upper limit, and a preset number of samples in the node belonging to the same category.

[0015] Based on the aforementioned technical means, this solution maintains the high-precision evaluation capability of neural networks while giving the evaluation results logical transparency, enabling users to intuitively understand the basis for the evaluation results and clarify the specific impact of each evaluation indicator on the final rating, thereby effectively solving the problems of black box nature and lack of interpretability of neural network models.

[0016] In one optional implementation, the performance evaluation results are scored and interpreted based on a decision tree model, including: interpreting the judgment path of the performance evaluation results, wherein the judgment path represents the indicators, thresholds, and comparison results used in the complete path from the root node to the leaf node; interpreting the boundary analysis of the performance evaluation results, wherein the boundary analysis is used to calculate the distance between the current indicator value and the adjacent level threshold to identify the indicator closest to the previous level threshold and the gap; and interpreting the importance ranking of the indicators in the performance evaluation results, wherein the importance ranking of the indicators is determined based on the position of the first appearance of each attribute in the decision tree and the information gain value.

[0017] Based on the aforementioned technical means, this solution not only retains the high-precision assessment capability but also endows the model with transparent interpretability, significantly enhancing the practical value and decision support capability of the comprehensive land consolidation effectiveness evaluation method, making the assessment results easier for business personnel to understand and adopt.

[0018] In an optional implementation, the method further includes: deriving actionable optimization suggestions based on the rule structure of the decision tree; wherein the optimization suggestions include path gap quantification suggestions, multi-indicator collaborative improvement suggestions, and improvement priority ranking suggestions; the path gap quantification suggestions are used to characterize the difference between the current indicator value and the corresponding threshold of the second rating for the first-rated project, calculated node by node along its judgment path, wherein the second rating is better than the first rating; the multi-indicator collaborative improvement suggestions are used to characterize the constraints that require simultaneous improvement of multiple evaluation indicators to achieve a grade leap; the improvement priority ranking suggestions are used to give priority suggestions considering the size of the gap and the feasibility of improvement of each evaluation indicator.

[0019] Based on the aforementioned technical means, the evaluation system can not only identify the problems in the project, but also provide actionable path planning, thereby effectively solving the problem of the disconnect between evaluation results and actual remediation actions, realizing closed-loop optimization of the comprehensive land remediation effectiveness evaluation, and significantly improving remediation effectiveness.

[0020] In an optional implementation, the method further includes: collecting data on remediation projects that have been completed and accepted within a preset time period, adding the data to the training set after data verification and preprocessing, and triggering retraining of the neural network performance evaluation model; when the cumulative number of corrected samples reviewed by experts reaches a preset threshold, or when the consistency rate between the prediction results of the neural network performance evaluation model and the expert evaluation results within multiple preset time periods is lower than a preset level, retraining of the neural network performance evaluation model is triggered.

[0021] Based on the above technical means, not only is it ensured that the evaluation model can be continuously optimized over time, but the data-driven approach also enhances the model's adaptability to complex remediation environments, ensuring the reliability and authority of the evaluation results throughout the entire life cycle. This enables the neural network-based performance evaluation method to provide accurate and reliable evaluation results in a long-term and stable manner.

[0022] Secondly, this application provides a performance evaluation device for comprehensive land consolidation across an entire region. The device includes: a construction module for constructing an evaluation index system and determining the weight of each evaluation index in the system, wherein the evaluation index system includes multiple primary indicators, and each primary indicator includes multiple secondary indicators; an evaluation module for evaluating the performance of the input evaluation indicators based on a trained neural network performance evaluation model to obtain performance evaluation results, wherein the neural network performance evaluation model includes a three-layer backpropagation neural network; a partitioning module for discretizing the performance evaluation results into multiple level categories and constructing a decision tree model using the evaluation indicators in the evaluation index system as splitting attributes; and an interpretation module for scoring and interpreting the performance evaluation results based on the decision tree model.

[0023] Thirdly, this application provides an electronic device, including: a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the efficiency evaluation method for comprehensive land consolidation of the first aspect or any corresponding embodiment described above.

[0024] Fourthly, this application provides a computer-readable storage medium storing computer instructions for causing a computer to execute the efficiency evaluation method for comprehensive land consolidation of the first aspect or any corresponding embodiment described above.

[0025] Fifthly, this application provides a computer program product, including computer instructions, which are used to cause a computer to execute the efficiency evaluation method for comprehensive land consolidation of the first aspect or any corresponding embodiment described above. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0027] Figure 1 This is a schematic diagram of the first process of the effectiveness evaluation method for comprehensive land consolidation according to an embodiment of this application; Figure 2 This is a schematic diagram of the second process of the effectiveness evaluation method for comprehensive land consolidation according to an embodiment of this application; Figure 3 This is a schematic diagram of the third process of the effectiveness evaluation method for comprehensive land consolidation according to the embodiments of this application; Figure 4 This is a first example diagram of the effectiveness evaluation method for comprehensive land consolidation according to an embodiment of this application; Figure 5 This is a second example diagram of the effectiveness evaluation method for comprehensive land consolidation according to an embodiment of this application; Figure 6 This is a third example diagram of the effectiveness evaluation method for comprehensive land consolidation according to an embodiment of this application; Figure 7 This is a structural block diagram of the performance evaluation device for comprehensive land consolidation according to an embodiment of this application; Figure 8 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0029] It is understood that before using the technical solutions disclosed in the various embodiments of this application, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this application in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0030] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0031] This application proposes a method for evaluating the effectiveness of comprehensive land consolidation across an entire region, such as... Figure 1 As shown, the method includes: Step S101: Construct an evaluation index system and determine the weight of each evaluation index in the evaluation index system. The evaluation index system includes multiple primary indicators, and each primary indicator includes multiple secondary indicators.

[0032] Step S102: Based on the trained neural network performance evaluation model, the performance of the input evaluation index is evaluated to obtain the performance evaluation result. The neural network performance evaluation model includes a three-layer backpropagation neural network.

[0033] Step S103: Discretize the performance evaluation results into multiple level categories, and construct a decision tree model using the evaluation indicators in the evaluation indicator system as splitting attributes.

[0034] Step S104: The performance evaluation results are scored and interpreted based on the decision tree model.

[0035] For ease of understanding, the following explains some key terms in this embodiment: An evaluation index system refers to a structured set of indicators used to measure and evaluate the effectiveness of comprehensive land consolidation projects across an entire region. This system typically includes multiple levels, such as primary and secondary indicators, aiming to comprehensively and systematically reflect the multifaceted impacts and achievements of the consolidation projects. Its function is to provide a comprehensive and hierarchical foundation of input data for effectiveness assessment.

[0036] The neural network performance evaluation model refers to a computational model based on artificial neural networks. Its design aims to predict or evaluate the performance of comprehensive land consolidation projects by learning the complex nonlinear relationship between input data (i.e., evaluation indicators) and project performance. This model can capture complex interactive effects that are difficult for traditional linear models to handle.

[0037] A three-layer backpropagation neural network refers to a specific configuration of neural network, which includes an input layer, a hidden layer, and an output layer. This network is trained using the backpropagation algorithm, which adjusts the connection weights and biases within the network based on the output error to optimize the model's predictive performance.

[0038] A decision tree model is a predictive model that uses a tree-like structure to represent decision rules. This model progressively classifies or regresses input data (e.g., evaluation indicators) to a final outcome (e.g., performance level) through a series of judgment conditions (node ​​splitting). Its key feature is that it provides intuitive and interpretable decision paths and rules.

[0039] This embodiment provides a method for evaluating the effectiveness of comprehensive land consolidation across an entire region. The method includes the following steps: First, an evaluation indicator system is constructed, and the weight of each indicator within this system is determined. The evaluation indicator system comprises multiple primary indicators, each of which includes multiple secondary indicators. The construction of the evaluation indicator system can be undertaken by a team of experts in the field, based on policy guidance and practical experience in comprehensive land consolidation. This can be achieved through brainstorming or the Delphi method to initially identify primary indicators covering multiple dimensions, including ecological, economic, social, and spatial aspects, and then further refine the secondary indicators under each primary indicator. For example, in terms of ecological benefits, indicators such as vegetation coverage and biodiversity can be considered; in terms of economic benefits, indicators such as agricultural output and industrial investment can be considered. When determining the weight of each evaluation indicator, an expert scoring method can be used, where multiple experts subjectively assign weights to the importance of each indicator, and the average value is taken as the final weight.

[0040] Secondly, based on the trained neural network performance evaluation model, the performance of the input evaluation indicators is assessed to obtain the performance evaluation results. The neural network performance evaluation model consists of a three-layer backpropagation neural network. The performance evaluation can be implemented using a pre-built and trained three-layer backpropagation neural network performance evaluation model. This model receives pre-processed evaluation indicator data as input and performs comprehensive analysis on these indicators through its internal nonlinear mapping mechanism. For example, the input layer receives the quantified values ​​of each secondary indicator, which are processed by the hidden layers, and finally a continuous evaluation score representing the overall performance of the project is generated at the output layer. During the training phase, the neural network can use historical project data for supervised learning, continuously adjusting the connection weights and biases within the network to accurately predict the actual performance level of the project from the input indicators. After training, the model has the ability to evaluate the performance of newly input evaluation indicators and output a continuous performance evaluation result.

[0041] Next, the performance evaluation results are discretized into multiple level categories, and a decision tree model is constructed using the evaluation indicators in the evaluation index system as splitting attributes. To make the performance evaluation results more readable and interpretable, the continuous performance evaluation results output by the neural network can be discretized. For example, a series of fixed score thresholds can be set to divide the evaluation results from 0 to 100 into multiple level categories such as "Excellent," "Good," "Average," and "Needs Improvement." Subsequently, using these discretized level categories as target variables and utilizing the various evaluation indicators in the evaluation index system as potential decision-making bases, a decision tree model is constructed. The construction process of the decision tree can start from the root node, analyze the influence of each evaluation indicator on the level category division, select the optimal indicator and its threshold for node splitting, and gradually form a tree structure. This tree structure can intuitively show the judgment path from the original indicators to the final level category.

[0042] Finally, the performance evaluation results are scored and interpreted based on the decision tree model. After the decision tree model is constructed, it can be used to score and interpret specific performance evaluation results. When an evaluation indicator for a project is input into the model and a performance evaluation result and its corresponding grade category are obtained, the decision path of the project in the decision tree can be traced from the root node to the final leaf node, clearly revealing which evaluation indicators, within what value range, jointly determined the project's final grade. For example, if a project is rated as "good," the decision tree interpretation can indicate that this is because its "vegetation coverage rate" is higher than a certain threshold, and its "agricultural output growth rate per acre" has also reached another threshold, thus making the project meet the criteria for a "good" grade. This interpretation method transforms the output of the black-box model into a series of understandable business rules, enhancing the transparency and credibility of the evaluation results.

[0043] It is understandable that this application, by constructing a multi-level evaluation index system and combining it with a trained three-layer backpropagation neural network for performance evaluation, effectively captures the nonlinear coupling relationship between various indicators in comprehensive land consolidation, thereby improving the accuracy and robustness of the evaluation. Simultaneously, by discretizing the evaluation results and constructing a decision tree model for scoring interpretation, the black-box evaluation results are transformed into understandable judgment paths. This solves the core technical problem of traditional methods struggling to balance model accuracy, parameter optimization robustness, and result interpretability, providing a clear decision-making basis for the optimization of land consolidation projects.

[0044] In some embodiments, this application further proposes a method for evaluating the effectiveness of comprehensive land consolidation, wherein primary indicators include ecological benefits, economic benefits, social benefits, and spatial structure optimization. Ecological benefits include vegetation coverage, biodiversity index, and the proportion of ecological restoration area. Economic benefits include the newly added area of ​​high-standard farmland, the growth rate of agricultural output per mu, and industrial investment intensity. Social benefits include the per capita net income of rural residents, the number of newly created jobs, and resident satisfaction. Spatial structure optimization includes the contiguousness of cultivated land, the intensity of construction land use, and the coverage rate of public service facilities. Figure 2 As shown, the method includes: Step S201: Construct an evaluation index system and determine the weight of each evaluation index in the system. Specifically, this includes: Step S2011: Based on the implementation content of the remediation, the remediation project is divided into multiple remediation models. The remediation models include farmland improvement, village remediation, ecological restoration, and industry introduction. The same remediation project shall include at least one remediation model. The implementation content of the remediation is the recorded content of the project establishment document.

[0045] Step S2012: Determine the corresponding weight vector in the evaluation index system according to different remediation models.

[0046] Step S2013: When there is no corresponding evaluation indicator in the rectification mode, mark the corresponding evaluation indicator as inapplicable in the evaluation indicator system, and allocate the weight of the evaluation indicator to other evaluation indicators of the same level.

[0047] Step S202: Based on the trained neural network performance evaluation model, the input evaluation indicators are evaluated to obtain the performance evaluation results. See details... Figure 1 Step S102 in the embodiment will not be described again here.

[0048] Step S203 involves discretizing the performance evaluation results into multiple level categories and constructing a decision tree model using the evaluation indicators in the evaluation indicator system as splitting attributes. See details... Figure 1 Step S103 in the embodiment will not be described again here.

[0049] Step S204: The performance evaluation results are scored and interpreted based on the decision tree model. See details in [link / reference]. Figure 1 Step S104 in the embodiment will not be described again here.

[0050] Specifically, the refinement of primary indicators aims to comprehensively measure the effectiveness of integrated land consolidation projects across the entire region from multiple dimensions. Ecological benefits focus on the project's improvement of the natural environment; for example, vegetation coverage quantifies the degree of greening, the biodiversity index reflects the health of the ecosystem, and the proportion of ecological restoration area directly reflects the scale of ecological restoration. Economic benefits emphasize the economic growth and efficiency improvement brought about by the project; for example, the newly added area of ​​high-standard farmland directly relates to agricultural production capacity, the growth rate of agricultural output per mu reflects land use efficiency, and industrial investment intensity reflects the scale and potential of industrial introduction. Social benefits focus on the project's positive impact on residents' living standards and community development; per capita net income of rural residents is an important indicator for measuring farmers' living standards, the number of newly created jobs reflects employment-creating capacity, and resident satisfaction directly reflects the public's recognition of the consolidation results. Spatial structure optimization focuses on the rationality and intensification of land use patterns; the contiguousness of cultivated land reflects the level of farmland standardization and scale, the intensification of construction land reflects the economical use of land resources, and the coverage rate of public service facilities measures the degree of infrastructure improvement. The setting of these indicators enables the evaluation system to comprehensively and meticulously capture the multifaceted impacts of consolidation projects.

[0051] The remediation projects are categorized into various models based on their implementation content, aiming to classify them according to their actual goals and priorities. The implementation content is typically recorded in the project initiation document, clearly defining the project's core tasks and expected outcomes. For example, the farmland improvement model primarily focuses on improving arable land quality and enhancing agricultural infrastructure; the village improvement model emphasizes improving the living environment and shaping the rural landscape; the ecological restoration model centers on ecosystem restoration and biodiversity conservation; and the industry introduction model aims to introduce new industries and promote economic transformation. This categorization ensures that subsequent performance evaluations align with the project's actual objectives.

[0052] Determining the corresponding weight vectors within the evaluation indicator system based on different remediation models is a crucial step in achieving dynamic weight allocation. Since different remediation models have different focuses, their contribution and importance to various evaluation indicators also differ. For example, for farmland improvement projects, the contiguousness of cultivated land in economic benefits and spatial structure optimization may have higher weights; while for ecological restoration projects, the weights of various ecological benefit indicators will significantly increase. There are various methods for determining the weight vectors. For instance, expert scoring can be used, inviting experts in relevant fields to assign weights to various indicators based on the importance of different remediation models; alternatively, objective data analysis methods can be combined, such as the Analytic Hierarchy Process (AHP) combined with the entropy weight method, to comprehensively determine the weights through expert experience and data-driven approaches.

[0053] When a corresponding evaluation indicator does not exist in the remediation model, the relevant evaluation indicator is marked as inapplicable in the evaluation indicator system, and its weight is allocated to other evaluation indicators at the same level. This aims to improve the accuracy and rationality of the evaluation. Under certain specific remediation models, some evaluation indicators may indeed be inapplicable or for which valid data cannot be obtained. For example, in a purely ecological restoration project, the newly added area of ​​high-standard farmland may not be considered a core evaluation indicator. In this case, marking it as inapplicable and rationally allocating its original weight to other applicable indicators at the same level can avoid interference from ineffective indicators in the evaluation results, ensuring the logical rigor and relevance of the evaluation. The redistribution of weights can be done in various ways, such as proportional allocation or allocation based on the relative importance of the remaining indicators.

[0054] It is understandable that, through the aforementioned technical solution, this application achieves deep coupling between the evaluation model and the actual business scenario of the project by refining the evaluation index system and introducing a dynamic weight allocation mechanism based on the remediation model. Specifically, by refining the primary indicators into four dimensions—ecological benefits, economic benefits, social benefits, and spatial structure optimization—and further breaking them down into specific secondary indicators, a multi-level evaluation framework is constructed, ensuring the comprehensiveness of the evaluation dimensions. Based on this, according to the remediation implementation content in the project initiation documents, the projects are divided into various models such as farmland improvement, village remediation, ecological restoration, and industry introduction. This classification method can accurately capture the core remediation objectives of different types of projects. By determining corresponding weight vectors according to different remediation models, the evaluation system can automatically adjust the contribution of each indicator according to the project's focus, thereby avoiding evaluation bias caused by a "one-size-fits-all" approach. Furthermore, for evaluation indicators missing in specific models, marking them as inapplicable and redistributing their weights effectively eliminates the interference of invalid indicators on the evaluation results, ensuring the logical rigor of the evaluation process and the relevance of the results. This significantly improves the adaptability of the evaluation model in complex and ever-changing remediation scenarios, enabling the effectiveness evaluation results to more accurately reflect the true effects of different types of remediation projects.

[0055] In some embodiments, this application further proposes determining corresponding weight vectors in the evaluation index system according to different remediation models, including: Step a1: Construct a judgment matrix based on the evaluation indicators in the evaluation indicator system.

[0056] Step a2: Calculate the weight vector of the constructed judgment matrix using expert opinions and perform a consistency check to obtain the hierarchical analysis weights; calculate the information entropy and difference coefficient of each evaluation indicator in the evaluation indicator system based on the data information recorded in the remediation project, and determine the entropy weight based on the information entropy and difference coefficient. The difference coefficient is used to characterize the sample discrimination ability of the evaluation indicator, and the sample discrimination ability is positively correlated with the corresponding assigned weight.

[0057] Step a3: The hierarchical analysis weights and entropy weights are weighted and fused according to the preset fusion coefficients to obtain the weight vector.

[0058] Specifically, when constructing the judgment matrix, a core tool in the Analytic Hierarchy Process (AHP), its role is to quantify the relative importance of different evaluation indicators, decomposing complex, multi-dimensional evaluation problems into a series of simple pairwise comparisons, thus providing a structured framework for the subsequent quantification of expert opinions. The judgment matrix can be constructed in various ways. For example, multiple domain experts can be organized to conduct pairwise comparisons and scoring of all secondary indicators in the evaluation indicator system, using a 1-9 scale (1 representing equal importance, 9 representing extreme importance), and the expert scores can be summarized to construct the judgment matrix. Alternatively, a questionnaire containing pairwise comparison questions of indicators can be designed and distributed to relevant stakeholders or professionals to collect their preferences on the relative importance of indicators, and then the collected data can be statistically analyzed to generate the judgment matrix.

[0059] When calculating the weight vector and performing a consistency check on the constructed judgment matrix based on expert opinions to obtain the hierarchical analysis weights, the weight vector calculation refers to deriving the relative weights of each indicator from the judgment matrix, while the consistency check is the process of evaluating the logical consistency of the expert judgments, ensuring that the judgment matrix does not contain serious logical contradictions. This step aims to systematically integrate the knowledge and experience of domain experts into the weight allocation, ensuring that the weights reflect the business logic of the governance model and the expert consensus, while the consistency check guarantees the rationality of the expert judgments. For example, the geometric mean method can be used to calculate the geometric mean of each row of elements in the judgment matrix, and then the result can be normalized to obtain the weight vector. The consistency check can also be completed by calculating the consistency index (CI) and the consistency ratio (CR).

[0060] When calculating the information entropy and difference coefficient of each evaluation indicator in the evaluation indicator system based on the data recorded in the remediation project, and determining the entropy weight based on the information entropy and difference coefficient, information entropy is an indicator that measures the degree of uncertainty or disorder of information. The smaller the information entropy, the greater the amount of information provided by the indicator, and the higher its importance. The difference coefficient is an indicator that measures the degree of data dispersion and is used to characterize the ability of the evaluation indicator to distinguish between different samples. The larger the difference coefficient, the higher the indicator's distinguishability, and the greater its weight should be. This step aims to uncover the inherent importance of the indicators from the perspective of objective data and avoid subjective bias. Through information entropy and difference coefficient, the information contribution and distinguishing ability of the indicators in actual project data can be quantified, making the weight allocation more objective and data-supported. For example, the original data can be standardized first to eliminate the influence of dimensions. Then, the information entropy of each indicator can be calculated based on the standardized data, and the difference coefficient can be obtained by calculating the ratio of the standard deviation to the mean of the indicator. Finally, the entropy weight can be determined by combining the information entropy and the difference coefficient. Alternatively, in addition to information entropy and the difference coefficient, other data distribution characteristics can be considered to comprehensively evaluate the sample discrimination ability of the indicator, design a comprehensive indicator, integrate multiple data features, and then allocate the entropy weight according to the size of the comprehensive indicator.

[0061] When weighting and fusing the weights of the hierarchical analysis (AMA) and entropy values ​​according to a preset fusion coefficient to obtain a weight vector, weighted fusion refers to linearly combining weights from two or more different sources using a preset fusion coefficient to obtain the final comprehensive weight vector. The preset fusion coefficient is a parameter used to balance the influence of different weight sources, and is usually determined through experience, expert consultation, or optimization algorithms. This step is crucial for organically combining subjective experience with objective data. Through weighted fusion, the expert's business-oriented understanding of the governance model can be preserved, while data-driven objective corrections can be introduced. This ensures that the final weight vector is both business-reasonable and reflects the objective laws of the data itself, thereby improving the comprehensiveness and robustness of weight allocation. For example, a fixed fusion coefficient can be preset, in which case the final weight is the weighted average of the AMA weights and entropy weights; or, the fusion coefficient can be dynamically adjusted according to specific conditions. For instance, when the consistency test results of expert opinions are poor, the fusion coefficient of the AMA weights can be appropriately reduced, and the proportion of the entropy weights can be increased.

[0062] It is understandable that, through the aforementioned technical solution, this application effectively addresses the limitations of a single weighting method in evaluating the effectiveness of comprehensive land consolidation by employing a combined weighting strategy that integrates the Analytic Hierarchy Process (AHP) and the entropy method. By constructing a judgment matrix based on the evaluation index system and combining it with expert opinions for weight calculation and consistency verification, the business experience and expert knowledge from consolidation projects can be transformed into AHP weights, ensuring the rationality and professionalism of the evaluation system in terms of macro-logic. Simultaneously, by calculating the information entropy and difference coefficient of each evaluation indicator based on the data recorded in the consolidation projects, the sample discrimination ability of the indicators can be extracted from the objective data level. That is, the entropy weight is determined by utilizing the volatility of the data itself, thereby avoiding the bias caused by relying entirely on subjective judgment, and ensuring that the weight allocation reflects the differentiated characteristics of different consolidation projects in actual implementation. The AHP weights and entropy weights are weighted and fused according to a preset fusion coefficient, achieving an organic combination of subjective experience and objective data. The integration mechanism retains the experts' business-oriented understanding of the remediation model while introducing data-driven objective corrections, making the final weight vector more robust and scientific when facing complex and ever-changing remediation projects, thus providing a more accurate basis for indicator weights for subsequent performance evaluation.

[0063] In some embodiments, this application further proposes that the neural network performance evaluation model is constructed based on a three-layer backpropagation neural network, wherein the number of input nodes of the neural network performance evaluation model is equal to the number of evaluation indicators, and the number of hidden layer nodes of the neural network performance evaluation model is determined by the following expression: ,in The number of nodes in the input layer. This represents the number of nodes in the output layer. To adjust the parameters, the hidden layer activation function of the neural network performance evaluation model is: The training of the neural network performance evaluation model includes: Step b1: The initial weights and thresholds of the three-layer backpropagation neural network are globally optimized using a genetic algorithm to obtain the trained neural network performance evaluation model.

[0064] Specifically, the neural network performance evaluation model is based on a three-layer backpropagation neural network. A three-layer backpropagation neural network (BP neural network) is a multi-layer feedforward network whose core function is to adjust the network's weights and thresholds by backpropagating the error to minimize the output error. It typically consists of an input layer, a hidden layer, and an output layer. This network structure can learn and store a large number of input-output pattern mappings, possessing strong nonlinear mapping capabilities, making it suitable for handling complex, nonlinear performance evaluation problems. In practical construction, a standard three-layer structure can be used: an input layer, a hidden layer, and an output layer, with fully connected layers. Alternatively, skip connections can be introduced in the hidden layer, allowing the input layer to directly connect to the output layer, thus capturing simpler linear relationships while retaining nonlinear mapping capabilities.

[0065] The number of input nodes in a neural network performance evaluation model equals the number of evaluation indicators. Input nodes are the interfaces through which the neural network receives external data; each input node corresponds to an input feature or variable. Evaluation indicators are specific quantitative metrics used to measure the overall effectiveness of land consolidation across the entire region. This configuration ensures that all indicators used for evaluation can be fully received and processed by the model, providing the necessary data foundation for comprehensive performance assessment. In implementation, preprocessed evaluation indicator values ​​can be directly used as inputs to the nodes of the input layer; alternatively, the evaluation indicators can be standardized or normalized before input, and then the processed values ​​can be input to the corresponding input nodes.

[0066] The number of hidden layer nodes in a neural network performance evaluation model is determined by the following expression: ,in The number of nodes in the input layer. This represents the number of nodes in the output layer. To adjust parameters. The number of hidden layer nodes is a key parameter in neural network architecture design, directly affecting the network's learning and generalization abilities. This expression provides a method for dynamically determining the number of hidden layer nodes based on the number of input and output nodes, aiming to balance model complexity and fitting ability, effectively avoiding overfitting or underfitting. The parameter 'a' can be set based on empirical values, such as a range of 1 to 10, and the optimal value of 'a' can be selected through methods such as cross-validation; alternatively, the parameter 'a' can be dynamically adjusted based on the characteristics of the dataset or the expected performance requirements of the model. For example, a smaller value of 'a' can be chosen on small datasets to reduce model complexity, while a larger value of 'a' can be chosen on large datasets to enhance model fitting ability.

[0067] The hidden layer activation function of the neural network performance evaluation model is: The activation function introduces a nonlinear element, enabling neural networks to approximate arbitrarily complex nonlinear functions. The Sigmoid function maps input values ​​to the interval (0, 1). It non-linearly transforms the output of the hidden layer, enabling the neural network to capture the complex non-linear coupling relationships between evaluation indicators for comprehensive land consolidation, thus enhancing the model's expressive power. Besides directly using the Sigmoid function as the activation function for the hidden layer, other activation functions with non-linear characteristics can be used, such as the hyperbolic tangent function (Tanh), whose output range is (-1, 1), or the ReLU (Rectified Linear Unit) function, which is computationally simple and effectively alleviates the gradient vanishing problem.

[0068] The training of the neural network performance evaluation model involves using a genetic algorithm to globally optimize the initial weights and thresholds of a three-layer backpropagation neural network, resulting in the trained neural network performance evaluation model.

[0069] Specifically, a genetic algorithm (GA) is a global optimization search algorithm that simulates natural selection and genetic mechanisms. Through operations such as selection, crossover, and mutation, it searches for the optimal solution in the solution space. The initial weights and thresholds of a neural network have a significant impact on the training effect and convergence speed of the model. Utilizing the global search capability of the genetic algorithm, the initial weights and thresholds can be optimized before the neural network training begins, placing them in a better parameter space. This avoids the local optima problem that may occur with traditional random initialization, improving the training stability and prediction accuracy of the model. In practice, the weights and thresholds of the neural network can be encoded as chromosomes of the genetic algorithm, and a fitness function (e.g., based on the reciprocal of the training set error) can be defined. The optimal combination of initial parameters can be found through an iterative evolutionary process. Alternatively, a hybrid optimization strategy can be adopted. For example, a coarse-grained global optimization can be performed using the genetic algorithm to find a better range of initial parameters, and then other local optimization algorithms (such as gradient descent) can be combined for fine-tuning.

[0070] It is understood that, through the above technical solution, this application introduces a genetic algorithm to globally optimize the initial weights and thresholds of the three-layer backpropagation neural network, fundamentally solving the training instability problem caused by random initialization of the neural network. The global search characteristic of the genetic algorithm enables the neural network to obtain a better parameter starting point in the early stage of training, effectively avoiding the model from getting trapped in local minima, thereby significantly improving the stability of model convergence and the reliability of prediction results. At the same time, by setting the number of input nodes to be consistent with the number of evaluation indicators, it is ensured that the model can fully receive all evaluation dimensions of comprehensive land consolidation, laying a data foundation for subsequent performance evaluation. Using a specific formula for calculating the number of hidden layer nodes, combined with parameter adjustment, the network complexity can be dynamically determined according to the dimensions of input and output, ensuring the model's fitting ability while effectively avoiding the risk of overfitting due to excessive network structure complexity. The use of the Sigmoid activation function can map the input signal to a nonlinear space, thereby effectively capturing the complex nonlinear coupling relationship between consolidation indicators and overcoming the shortcomings of traditional linear evaluation models in handling interaction effects. Overall, this scheme makes the evaluation model more robust when faced with data from different remediation projects, and improves the accuracy and stability of the evaluation of the effectiveness of comprehensive land remediation across the entire region.

[0071] In some embodiments, this application further proposes a genetic algorithm to globally optimize the initial weights and thresholds of a three-layer backpropagation neural network, thereby obtaining a trained neural network performance evaluation model, including: Step c1 involves using real-number encoding to concatenate the weights and thresholds of the three-layer backpropagation neural network into a one-dimensional vector, which serves as the chromosome. The chromosome length is n. i ×n h +n h ×n0+n h +n0, where n i n is the number of nodes in the input layer, n0 is the number of nodes in the output layer, and n h This represents the number of hidden layer nodes.

[0072] Step c2: Take the reciprocal of the mean square error of the training set as the fitness function.

[0073] Step c3 involves configuring the population size, maximum number of generations, crossover probability, mutation probability, and elite retention ratio, and then performing convergence training. After convergence, the weights and thresholds corresponding to the optimal individuals representing the evaluation indicators are used as the initial parameters of the three-layer backpropagation neural network. Step c4: Adjust the parameters based on the backpropagation algorithm to obtain the trained neural network performance evaluation model.

[0074] Specifically, real-number encoding directly maps the continuous parameters (weights and thresholds) of the neural network to real values ​​on the chromosomes of the genetic algorithm. This effectively handles the high dimensionality and continuity of the neural network parameter space, avoiding the precision loss and encoding-decoding complexity that binary encoding might introduce. For example, each weight and threshold can be directly treated as a gene locus, or all weights and thresholds can be concatenated into a long real-number vector in a specific order (e.g., inter-layer weights, inter-layer thresholds). This allows the genetic algorithm to search directly in the continuous parameter space, thus approximating the optimal solution more accurately.

[0075] Taking the reciprocal of the mean squared error (MSE) of the training set as the fitness function quantifies the predictive performance of the neural network on the training set into the optimization objective of the genetic algorithm. MSE is a commonly used metric to measure the difference between model predictions and true values; a smaller MSE value indicates better model performance. By taking its reciprocal, the problem of minimizing error is transformed into a problem of maximizing fitness, which aligns with the fitness maximization principle typically used in genetic algorithms. Besides the reciprocal of MSE, the reciprocals of other error metrics can also be used, such as the reciprocal of the root mean squared error (RMSE) or the reciprocal of the mean absolute error (MAE), as long as they accurately reflect model performance and are consistent with the optimization direction of the genetic algorithm.

[0076] Using 75 comprehensive land consolidation projects implemented in a certain province between 2019 and 2023 as a sample, the projects were sorted by completion time. The first 60 projects were used for training, and the last 15 were used for testing. The specific example project is sample number 68, located in the test set. In the 75 samples, missing values ​​accounted for approximately 3.2% (mainly concentrated in the biodiversity index), and all were imputed using the mean of the same index from the city where the project was located. The fitness function was the reciprocal of the MSE of the training set, i.e., Fitness = 1 / MSE. In the formula, For the sample size, This is the actual value. This is a predicted value. In the early stages of the solution design, we attempted to add a business rule compliance term to the fitness function, but in the experiment, we found that this term was difficult to quantify uniformly and had limited guiding effect on the optimization direction. Therefore, we ultimately used the reciprocal of MSE as the sole fitness indicator.

[0077] Convergence training, after configuring population size, maximum number of generations, crossover probability, mutation probability, and elite retention ratio, refers to setting a series of key parameters before running the genetic algorithm to control its search behavior and efficiency. Population size determines the number of individuals in each generation, affecting the breadth of the search; the maximum number of generations sets the termination condition for the algorithm; the crossover probability controls the frequency of gene recombination, promoting the generation of new individuals; the mutation probability introduces random perturbations, helping to escape local optima; and the elite retention ratio ensures that the best individual in each generation can directly inherit to the next generation, accelerating convergence. Convergence training refers to continuously monitoring changes in the fitness of the population during the iteration process of the genetic algorithm. When the fitness no longer significantly increases or reaches the preset maximum number of generations, the algorithm is considered to have converged. At this point, the weights and thresholds represented by the individual with the highest fitness in the population (i.e., the optimal individual) are extracted and used as the initial parameters for the three-layer backpropagation neural network.

[0078] During convergence training, the parameters were determined based on genetic algorithm design experience and the problem size of this application (dimensionality of decision variables, evaluation sample size), and the values ​​are as follows:

[0079] The above-mentioned value range represents a conventional empirical range that balances convergence and diversity in optimization problems within this field. The termination condition is that the fitness change rate is below a preset threshold for several consecutive generations or reaches the maximum generation limit. In one embodiment: the population size is 100, the maximum number of generations is 1000, the selection operation uses roulette wheel selection, retaining the top 20% of elite individuals, the crossover probability is 0.8 (single-point crossover), and the mutation probability is 0.05 (applying a small-scale random perturbation to the selected gene). The termination condition is that the fitness change rate is <0.1% for 100 consecutive generations, and convergence actually occurs at generation 487. The comparison before and after GA optimization is as follows:

[0080] Convergence employs a dual criterion, terminating training upon satisfying either one: reaching the preset maximum number of generations or, within a continuous N_gen, if the relative improvement in optimal fitness is less than a threshold ε, then it is determined that "fitness no longer significantly improves," and the algorithm has converged. This criterion avoids invalid iterations, saves computational resources, and ensures the quality of the solution. In specific implementation, the termination condition is a fitness change rate of <0.1% for 100 consecutive generations; in practice, convergence occurred at generation 487.

[0081] Parameter tuning based on backpropagation refers to further refining the local training of the neural network using backpropagation after the genetic algorithm has provided a set of initially optimized weights and thresholds globally. Backpropagation is an efficient gradient descent optimization method that calculates the gradient of the loss function with respect to the network parameters and updates the parameters in the opposite direction of the gradient to gradually reduce prediction error. For example, variants of backpropagation such as stochastic gradient descent (SGD) and the Adam optimizer can be used, combined with appropriate learning rates and batch sizes, to iteratively train the neural network until the model performance reaches the expected level or converges.

[0082] It is understood that, through the above technical solution, this application transforms the complex parameters of the neural network into chromosomes that can be processed by the genetic algorithm through real-number encoding, effectively solving the optimization problem of the high-dimensional and nonlinear parameter space of the neural network, enabling the genetic algorithm to directly perform a global search on the network structure parameters. By using the reciprocal of the mean square error of the training set as the fitness function, a direct correlation is established between the model's predictive performance and the evolutionary direction of the genetic algorithm, ensuring that the population evolution always moves in the direction of reducing prediction errors. After configuring parameters such as population size, number of generations, and crossover and mutation probabilities, convergence training can effectively avoid the neural network from getting stuck in local optima in the early stages of training, thus providing the network with a high-quality set of initial weights and thresholds. Finally, based on the parameters optimized by the genetic algorithm as the starting point of the backpropagation algorithm, the local fine-tuning capability of the backpropagation algorithm is utilized to achieve an organic combination of global optimization and local convergence, significantly improving the training efficiency and final prediction accuracy of the neural network model in the evaluation of the effectiveness of comprehensive land consolidation across the entire region. This combination not only improves the stability of the model, reducing the fluctuation of its training results under different initial conditions, but also provides a more reliable and accurate foundation for the subsequent discretization of performance evaluation results and the construction of decision trees, thereby enhancing the robustness and interpretability of the entire evaluation method.

[0083] In some embodiments, this application further proposes that the performance evaluation results include predicted scores, the performance evaluation results are discretized into multiple level categories, and a decision tree model is constructed using evaluation indicators in the evaluation indicator system as splitting attributes, specifically including: Step d1 involves dividing the predicted scores into multiple categories based on the grade threshold.

[0084] The effectiveness evaluation results include a predicted score, which is a continuous numerical result output by the neural network effectiveness evaluation model after evaluating the input evaluation indicators. This score is used to quantify the overall effectiveness level of the comprehensive land consolidation project. It directly reflects the model's comprehensive judgment of the project's performance and forms the basis for subsequent discretization and interpretation. The predicted score can be a single floating-point value, such as 0.75; or it can be a normalized percentage score, such as 75.

[0085] The predicted scores are categorized into multiple classes based on a threshold, aiming to transform the continuous predicted scores output by the neural network model into discrete, business-meaning categories such as "Excellent," "Good," "Average," and "Poor." These categories are then used as target variables or classification labels when building a decision tree model, enabling the decision tree to learn how to classify items into different performance levels based on evaluation indicators. For example, a series of fixed thresholds can be preset, classifying 0-0.6 as "Poor," 0.6-0.75 as "Average," 0.75-0.9 as "Good," and 0.9-1.0 as "Excellent." Alternatively, unsupervised learning can be performed on historical predicted score data using cluster analysis (such as the K-means algorithm) to automatically discover natural cluster boundaries in the data, thereby determining the thresholds and categories. The indicators are then ranked according to their importance.

[0086] Step d2 uses information gain as the node splitting metric. Information gain characterizes the reduction in information entropy before and after the level split. Information gain is defined as Gain(D,A) = H(D) - H(D|A), where D is the training sample set, H(D) is the information entropy, and Gain represents the information gain of A as a feature attribute on the training set D, where A is a candidate feature attribute. The sample labels in the training sample set are determined by expert evaluation of previously accepted projects.

[0087] Information gain is a commonly used splitting criterion in decision trees. Its core idea is to select the attribute that minimizes dataset uncertainty for splitting. When constructing a decision tree, information gain quantifies the degree to which dataset purity improves after splitting using a specific evaluation metric, i.e., the degree to which classification uncertainty is reduced. Choosing the attribute with the highest information gain as the splitting node ensures that the decision tree maximizes classification efficiency and accuracy at each splitting step. Besides information gain, the information gain ratio can also be used as a splitting metric to avoid bias towards attributes with more values.

[0088] Step d3: Starting from the root node, traverse all possible thresholds for all evaluation metrics for each node, and select the attribute with the largest information gain and the corresponding threshold combination for splitting. Recursively construct subtrees. The stopping conditions for recursion include the number of samples in a node being lower than a preset minimum, the tree depth reaching a preset upper limit, and a preset number of samples in a node belonging to the same category. To prevent overfitting, cost complexity pruning is used, and the optimal complexity parameter is selected through cross-validation. The evaluation metric with the largest information gain is selected as the root node.

[0089] The construction of a decision tree is a top-down recursive process. Starting from the root node, the algorithm systematically examines all available evaluation metrics and their possible split thresholds, calculating the information gain for each split. The metric and threshold with the highest information gain are selected for splitting, dividing the dataset into subsets, and a new child node is created for each subset. This process continues recursively until a preset stopping condition is met, such as the node containing too few samples, the tree depth reaching a limit, or all samples in the node belonging to the same category; at this point, the node becomes a leaf node. To avoid the problem of decision tree models performing well on training data but having poor generalization ability on new data (i.e., overfitting), pruning techniques are used. Cost-complexity pruning balances the accuracy and complexity of the model by introducing a complexity parameter, and selects the optimal complexity parameter through cross-validation, resulting in a decision tree model with stronger generalization ability. When traversing the possible thresholds of the evaluation metrics, methods such as binary search or sorting and taking the median can be used to efficiently determine candidate thresholds. Besides cost-complexity pruning, a pre-pruning strategy can be used, which involves stopping splitting a node when its information gain falls below a preset threshold during decision tree growth. For cross-validation, K-fold cross-validation can be chosen to more robustly evaluate model performance and select optimal parameters.

[0090] It is understood that, through the above technical solution, this application first discretizes the continuous predicted scores output by the neural network model into level categories with clear business meanings through level thresholds, thereby transforming the complex numerical evaluation into a classification problem. Based on this, a decision tree model is constructed using evaluation indicators from the evaluation indicator system as splitting attributes and information gain as the node splitting indicator. Information gain quantifies the contribution of evaluation indicators to level division. By calculating the reduction in information entropy before and after level division, it ensures that the decision tree prioritizes the evaluation indicators with the greatest impact on the classification results as splitting attributes during construction, thus constructing a highly logically related decision path. Traversing the thresholds of evaluation indicators from the root node and recursively constructing subtrees decomposes the complex evaluation system into a series of intuitive judgment rules, transforming the evaluation process from a black box into a traceable path. Simultaneously, by setting recursion stopping conditions and cost complexity pruning, the model's fitting ability and generalization ability are effectively balanced, avoiding overfitting when the decision tree processes complex indicator data, and ensuring the stability and reliability of the model under different remediation project data. Overall, this solution maintains the high-precision evaluation capability of neural networks while providing logical transparency to the evaluation results. This allows users to intuitively understand the basis for the evaluation results and clarify the specific impact of each evaluation indicator on the final rating, thereby effectively solving the problems of black box nature and lack of interpretability of neural network models.

[0091] In some embodiments, this application further proposes a method for evaluating the effectiveness of comprehensive land consolidation, such as... Figure 3 As shown, the method includes: Step S301: Construct an evaluation index system and determine the weight of each evaluation index in the system. Specifically, this includes: Step S3011: Based on the implementation content of the remediation project, the project is divided into multiple remediation models. These models include farmland improvement, village remediation, ecological restoration, and industry introduction. Each remediation project must include at least one model. (See details below.) Figure 1 Step S2011 in the embodiment will not be described again here.

[0092] Step S3012: Determine the corresponding weight vector in the evaluation index system according to different remediation models. See details... Figure 1 Step S2012 in the embodiment will not be described again here.

[0093] Step S3013: When there is no corresponding evaluation indicator in the rectification model, mark the corresponding evaluation indicator as inapplicable in the evaluation indicator system, and allocate the weight of the evaluation indicator to other evaluation indicators of the same level. See details. Figure 1 Step S2013 in the embodiment will not be described again here.

[0094] Step S302: Based on the trained neural network performance evaluation model, the performance of the input evaluation indicators is assessed to obtain the performance evaluation results. See details... Figure 1 Step S102 in the embodiment will not be described again here.

[0095] Step S303 involves discretizing the performance evaluation results into multiple level categories and constructing a decision tree model using the evaluation indicators in the evaluation index system as splitting attributes. Specifically, this includes: Step S3031: Interpret the judgment path of the performance evaluation results, wherein the judgment path represents the indicators, thresholds and comparison results used for the complete path from the root node to the leaf node.

[0096] Step S3032: Interpret the critical boundary analysis of the performance evaluation results. The critical boundary analysis is used to calculate the distance between the current indicator value and the threshold of the adjacent level in order to identify the indicator and gap closest to the threshold of the previous level.

[0097] Step S3033: Interpret the importance ranking of the indicators in the effectiveness evaluation results, wherein the importance ranking of the indicators is determined based on the position of the first appearance of each attribute in the decision tree and the information gain value.

[0098] Step S304: The performance evaluation results are scored and interpreted based on the decision tree model. See details... Figure 1 Step S104 in the embodiment will not be described again here.

[0099] Specifically, the scoring and interpretation of performance evaluation results based on decision tree models aims to transform the "black box" performance evaluation results output by neural networks into understandable, rule-based explanations, thereby improving the model's credibility and practicality. In practical applications, various decision tree algorithms can be used to construct this model. For example, the CART (Classification and Regression Tree) algorithm can be used, which splits nodes using Gini impurity or mean squared error, handling classification and regression tasks and generating an intuitive binary tree structure. Alternatively, Conditional Inference Trees can be used, which selects splitting variables and splitting points through statistical tests, helping to avoid the bias problems of traditional decision trees and providing more robust interpretations.

[0100] When interpreting the judgment path of performance evaluation results, its core role is to reveal how a specific evaluation result is ultimately derived through a series of indicator judgments and threshold comparisons, thereby concretizing the abstract scoring process into a traceable logical chain. One implementation method is that, after the decision tree model is constructed, for each performance evaluation result to be interpreted (i.e., a sample), starting from the root node of the decision tree, based on the evaluation indicator values ​​of the sample, the tree is traversed downwards along the branches until a leaf node is reached. During this process, the evaluation indicators used by all internal nodes traversed, the splitting thresholds, and the comparison results between the sample and the thresholds (e.g., "vegetation coverage greater than 0.8" or "agricultural output growth rate per mu less than or equal to 0.05") are recorded, thus forming a complete judgment path. Another approach is to use visualization tools to present the structure of the decision tree and highlight the decision path followed by a specific sample. At the same time, each decision point on the path can be output in natural language, for example: "Because the 'vegetation coverage' is greater than 0.8 and the 'new area of ​​high-standard farmland' is greater than 100 mu, the rating is 'excellent'."

[0101] The interpretative role of boundary analysis for performance evaluation results is to quantify which key indicators the current project is closest to a grade improvement, thus providing clear and actionable directions for improvement. One approach is to first determine the current grade category of a given performance evaluation result. Then, find the leaf node leading to that grade category and its upstream path in the decision tree. Next, identify all adjacent grade thresholds that could lead to a grade improvement (e.g., the indicator threshold required to improve from "Good" to "Excellent"). For each relevant indicator, calculate the difference between the current indicator value and the threshold required for that grade improvement, i.e., the "distance" or "gap." Finally, identify the indicator with the smallest gap and report its specific value and the amount of improvement required. Another approach is through simulation analysis. For a project, keeping other indicators constant, fine-tune each indicator value one by one and observe when it can leap to a higher grade. Record the minimum adjustment required for each indicator and use it as the result of the boundary analysis. For example, if improving "vegetation coverage" from 0.75 to 0.8 would change the rating from "good" to "excellent", then the report shows that "vegetation coverage" is the closest indicator to the threshold of the next higher level, with a difference of 0.05.

[0102] Furthermore, interpreting the importance ranking of indicators in the effectiveness evaluation results serves to objectively quantify the contribution of each evaluation indicator to the final evaluation result, helping users identify key factors affecting the effectiveness of the rectification efforts. One implementation method is to calculate the information gain at each node split during the decision tree construction process. The larger the information gain, the greater the contribution of the indicator to the classification at the current node split. Therefore, the total information gain (or weighted information gain, with weights determined based on the number of samples contained in the node) of each evaluation indicator appearing as a splitting attribute in the decision tree can be calculated, and the indicators can be sorted in descending order based on this value to obtain the importance ranking. Simultaneously, the position of the indicator's first appearance in the tree (the closer to the root node, the more important) can also serve as an auxiliary judgment criterion. Another implementation method is to calculate the sum of the information gain brought by each indicator in all node splits and normalize it as the importance score.

[0103] The first appearance position according to the indicator In a decision tree, the depth of the node that splits for the first time is converted into a positional score, reflecting its overall splitting ability. , The maximum depth of the decision tree is given by the information gain, which is the information gain ratio of each split node for that metric. The weighted cumulative normalized value of the sample reflects the strength of its purity contribution. The comprehensive score is obtained by linearly weighting the two components. The weights are calibrated with reference to the importance of the validation set permutation, and the output ranking is after consistency and sensitivity tests to identify the indicators that need to be improved first.

[0104] It is understandable that, through the above technical solution, this application combines the fitting advantages of neural network performance evaluation models in handling nonlinear complex relationships with the interpretability advantages of decision tree models, effectively solving the problem that neural network models, being "black boxes," lead to difficult-to-understand evaluation results. By explaining the judgment path, users can clearly trace the logical source of the evaluation results, transforming abstract scores into traceable business logic. Through boundary analysis, it is possible to quantify which indicators the current project is closest to achieving an improvement level, thus providing users with clear and actionable improvement directions. By ranking the importance of indicators, it is possible to objectively identify key factors affecting the effectiveness of land consolidation, making resource allocation more targeted. Overall, this solution not only retains high-precision evaluation capabilities but also endows the model with transparent interpretability, significantly enhancing the practical value and decision support capabilities of the comprehensive land consolidation performance evaluation method, making the evaluation results easier for business personnel to understand and adopt.

[0105] In some embodiments, this application further proposes a method for evaluating the effectiveness of comprehensive land consolidation, the method further comprising: Step e1 involves deriving actionable optimization suggestions based on the rule structure of the decision tree. These suggestions include path gap quantification suggestions, multi-indicator synergistic improvement suggestions, and improvement priority ranking suggestions. Path gap quantification suggestions characterize the difference between the current indicator value and the corresponding threshold for the second rating, calculated node by node along the judgment path for projects in the first rating, where the second rating is superior to the first rating. Multi-indicator synergistic improvement suggestions characterize the constraints that require simultaneous improvement of multiple evaluation indicators to achieve a rating leap. Improvement priority ranking suggestions consider the magnitude of the gaps and the feasibility of improvement for each evaluation indicator, providing a priority ranking recommendation.

[0106] Specifically, "deriving actionable optimization suggestions based on the rule structure of decision trees" refers to using the "if-then" rules formed by the discretization and classification of performance evaluation results in a decision tree model to analyze in reverse or forward directions to generate specific, actionable improvement measures. For example, by traversing the leaf nodes of the decision tree and backtracking to the root node, all path rules leading to a specific rating (e.g., "unsatisfactory") can be extracted. Then, suggestions for reverse operations can be proposed for the evaluation indicators and thresholds in these rules, such as "If evaluation indicator A is below threshold X, resulting in unsatisfactory, then it is recommended to raise evaluation indicator A to above X." Alternatively, a rule engine can be built to transform each branch path of the decision tree into an independent business rule. When current evaluation indicator data for a project is input, the rule engine not only provides a rating but also identifies which evaluation indicators are the main factors leading to the current rating based on the decision path it is on, and generates improvement suggestions based on these factors.

[0107] "Optimization suggestions include path gap quantification suggestions, multi-indicator collaborative improvement suggestions, and improvement priority ranking suggestions"—this is a specific classification and content limitation of optimization suggestions, aiming to ensure the comprehensiveness and practicality of the suggestions. For example, the system can be designed with a suggestion generation module containing three sub-modules, each responsible for generating these three types of suggestions. Upon receiving the performance evaluation results and the decision tree model, this module will call these three sub-modules in parallel or sequentially to generate complete optimization suggestions. Another implementation approach is to pre-set or dynamically calculate information related to these three suggestion types at each node or leaf node of the decision tree. For example, the corresponding rating can be stored at the leaf node, and evaluation indicators and threshold information can be stored on the path for subsequent gap calculation and identification of collaborative relationships.

[0108] The "Path Gap Quantification Suggestion" aims to provide specific, numerical improvement targets for projects rated "First Grade." It calculates the gap between the current indicator value and the corresponding threshold for the Second Grade, which is superior to the First Grade, node by node along the decision path. Instead of simply stating "a certain evaluation indicator is bad," it quantifies "how much improvement is needed to reach a higher level." For example, for a project rated "First Grade," the system first identifies its decision path in the decision tree. Then, it finds adjacent paths or parent nodes leading to the "Second Grade" (e.g., from "Medium" to "Good") and compares the evaluation indicator values ​​involved on that path with the threshold required for the "Second Grade," thus calculating the specific gap. Another example is through reverse traversal of the decision tree. Starting from the leaf node of the current project, it backtracks to the root node, recording all traversed nodes and splitting conditions. Then, it selects a leaf node for the target "Second Grade" and compares the evaluation indicator thresholds on the two paths, calculating the minimum gap between the current evaluation indicator value and the threshold required for the target grade.

[0109] The "Multi-Indicator Collaborative Improvement Suggestions" are designed to characterize the constraints requiring simultaneous improvement of multiple evaluation indicators to achieve a higher rating. This aims to identify complex scenarios where improvement of a single evaluation indicator is insufficient to raise the rating, necessitating the combined effect of multiple indicators to reach a higher rating. This reflects the non-linear coupling relationships between evaluation indicators. For example, the system can analyze complex branches in a decision tree with "AND" or "OR" relationships. If a project needs to improve from "Qualified" to "Excellent," it might require both "Economic Benefits" and "Ecological Benefits" to reach a certain threshold. The system will identify these "AND" relationships and suggest simultaneous improvements to these evaluation indicators. Another approach is to analyze the pruning paths or alternative split points of the decision tree. If the splitting condition of a node involves a combination of multiple evaluation indicators, or if reaching a threshold for one evaluation indicator still requires another indicator to reach a threshold to enter a branch with a higher rating, collaborative improvement suggestions can be generated.

[0110] The "Improvement Priority Ranking Suggestions," which consider the size of the gaps and the feasibility of improvement for each evaluation indicator, aim to provide practical guidance, avoid resource waste, and ensure the effectiveness and efficiency of improvement measures. For example, the system can preset or dynamically assess an "improvement difficulty coefficient" or "improvement cost" for each evaluation indicator. After calculating the path gap, this difficulty coefficient is combined with a weighted function or decision rule to rank all evaluation indicators to be improved; for example, indicators with large gaps but low improvement difficulty are given priority. As another example, priority ranking can be based on an expert experience rule base. Experts can define that "ecological benefits" evaluation indicators have the highest priority in certain situations, or that "economic benefits" evaluation indicators have priority when the return on investment is high. The system combines the quantified gaps with these rules to generate a ranking.

[0111] It is understood that, through the aforementioned technical solutions, this application can transform the performance evaluation results from passive interpretation into proactive and actionable optimization guidance. By utilizing the judgment paths and threshold logic already constructed in the decision tree model, the system can accurately calculate the specific gap between the current evaluation index value and the higher-level threshold, giving the improvement target a clear numerical direction and avoiding vague improvement directions. Simultaneously, by identifying the constraints requiring simultaneous improvement of multiple evaluation indicators to achieve a level leap, this application can guide users to focus on the linkage relationships between evaluation indicators, ensuring that the improvement plan logically meets the leap conditions of the decision tree, thereby effectively addressing the nonlinear coupling problem between evaluation indicators. Furthermore, by comprehensively considering the magnitude of the gaps in evaluation indicators and the feasibility of improvement, this application provides a scientific execution sequence for remediation projects, avoiding blind resource investment and ensuring the targeting and efficiency of remediation work. These technical features work together to enable the evaluation system not only to identify the problems in the project but also to provide actionable path planning, thereby effectively solving the problem of the disconnect between evaluation results and actual remediation actions, achieving closed-loop optimization of the comprehensive land remediation performance evaluation across the entire region, and significantly improving remediation efficiency.

[0112] In some embodiments, this application further proposes a dynamic update and calibration mechanism, the method of which includes: Step f1: Collect data on remediation projects that have been completed and accepted within a preset time period. After data verification and preprocessing, add the data to the training set to trigger retraining of the neural network performance evaluation model.

[0113] Step f2: When the cumulative number of corrected samples reviewed by experts reaches a preset threshold, or when the consistency rate between the prediction results of the neural network performance evaluation model and the expert evaluation results is lower than a preset level within multiple preset time periods, the neural network performance evaluation model is retrained.

[0114] Specifically, the "collection of data from remediation projects that have been completed and accepted within a preset timeframe" aims to ensure the freshness and timeliness of the model training data, enabling it to reflect the latest remediation practices and environmental changes. For example, it can be set to collect data on remediation projects that have been completed and put into use within a specific time period, quarterly, semi-annually, or annually. This data typically originates from project management systems, geographic information systems, or field survey reports. Furthermore, the system can also interface with a project acceptance management platform, so that once a project completes acceptance, its relevant data is automatically or semi-automatically added to the collection queue.

[0115] The purpose of "adding newly collected data to the training set after data validation and preprocessing" is to ensure the quality and consistency of the data, making it meet the requirements of model training and avoiding the negative impact of dirty data on model performance. Data validation can include integrity checks (e.g., missing values), consistency checks (e.g., consistency of data types and units), and outlier detection (e.g., identifying unreasonable data through statistical methods or expert experience). Preprocessing may involve data cleaning (e.g., missing value imputation, outlier removal), data transformation (e.g., normalization, standardization), and feature engineering (e.g., extracting or constructing new features from the original data). Another approach is to use automated scripts combined with manual review. The automated scripts are responsible for initial format checks and anomaly marking. For the marked problematic data, data engineers or domain experts conduct manual review and correction to ensure data quality.

[0116] "Triggering retraining of the neural network performance evaluation model" refers to initiating the model update process, enabling it to learn new data patterns and evaluation logic, thereby maintaining its predictive accuracy and adaptability. Retraining can be incremental learning based on existing model parameters, that is, fine-tuning the model using new data to adapt to new data distributions. Alternatively, it can be retraining the model from scratch using all historical data (including newly added data) to ensure that the model can fully learn all available information.

[0117] The "when the cumulative number of corrected samples reviewed by experts reaches a preset threshold" mechanism provides a calibration mechanism based on expert experience. When there is a significant difference between the model prediction and the expert's judgment, and the number of samples corrected by the expert accumulates to a certain extent, model retraining is triggered to correct the model's bias in specific evaluation logic. For example, the system records the number of samples in which experts manually review and correct the model's prediction results. When the number of these corrected samples (e.g., 100 samples corrected or a correction rate of 5%) reaches a preset threshold, retraining is triggered. Alternatively, it can be set to trigger retraining when the number of times experts correct the model's prediction results reaches a certain frequency within a specific time window.

[0118] The provision "when the consistency rate between the prediction results of the neural network performance evaluation model and the expert evaluation results is lower than a preset level within multiple preset time periods" provides a calibration mechanism based on model performance monitoring. When the overall prediction performance of the model deviates significantly from the expert evaluation over a period of time, model retraining is triggered to maintain the overall accuracy and reliability of the model. For example, the system periodically (e.g., monthly or quarterly) calculates the consistency rate between the model's prediction results for new projects in practical applications and the final expert evaluation results for these projects. When this consistency rate is lower than a preset minimum acceptable level (e.g., lower than 85%) for multiple consecutive periods (e.g., two consecutive quarters), retraining is triggered. Alternatively, retraining can be triggered when the model's prediction results for a batch of new samples within a specific time period, and metrics such as F1 score, precision, or recall, are lower than preset levels compared to the expert evaluation results.

[0119] It is understood that, through the aforementioned technical solution, this application introduces a dynamic data update and model retraining mechanism, effectively addressing the issues of timeliness and accuracy degradation in long-term application of evaluation models. By periodically collecting and verifying preprocessed new data, it ensures that the training set reflects the latest governance practices and environmental changes in real time, enabling the neural network performance evaluation model to continuously learn new data distribution characteristics and maintain its high-precision evaluation capabilities. Simultaneously, by setting a cumulative threshold for expert review and correction samples, deviations in specific evaluation logic can be promptly captured and corrected, ensuring that the model output always aligns with industry expert experience. Furthermore, by monitoring the consistency rate between model predictions and expert evaluations, retraining is automatically triggered when the consistency level falls below a preset standard, achieving closed-loop monitoring and self-correction of model performance. This mechanism not only ensures that the evaluation model can continuously optimize over time but also enhances the model's adaptability to complex governance environments through a data-driven approach, ensuring the reliability and authority of evaluation results throughout their entire lifecycle. This allows the neural network-based performance evaluation method to provide accurate and reliable evaluation results stably and consistently over the long term.

[0120] In one example, such as Figure 4 , Figure 5 and Figure 6 As shown, the above technical solution will be explained in more detail through a more specific example: Suppose a comprehensive land consolidation project has been completed in a certain "demonstration area," and its effectiveness needs to be evaluated.

[0121] like Figure 4As shown, firstly, based on the specific circumstances of the project, a comprehensive evaluation index system is constructed. This system includes multiple primary indicators such as ecological benefits, economic benefits, social benefits, and spatial structure optimization. For example, the primary indicator of ecological benefits includes secondary indicators such as vegetation coverage rate, biodiversity index, and the proportion of ecological restoration area; the primary indicator of economic benefits includes secondary indicators such as the newly added area of ​​high-standard farmland, the growth rate of agricultural output value per mu, and industrial investment intensity; the primary indicator of social benefits includes secondary indicators such as the per capita net income of rural residents, the number of newly created jobs, and resident satisfaction; and the primary indicator of spatial structure optimization includes secondary indicators such as the contiguousness of cultivated land, the intensity of construction land use, and the coverage rate of public service facilities.

[0122] When determining the weight of each evaluation indicator, the "demonstration area" project is first identified as a composite remediation model of "farmland improvement" and "ecological restoration" based on the remediation implementation content recorded in the project approval documents. For these two remediation models, the system determines the corresponding weight vector. Specifically, a judgment matrix is ​​constructed using expert opinions, and the analytic hierarchy process (AHP) weights are calculated. Simultaneously, based on the data recorded in the remediation project, the information entropy and difference coefficient of each evaluation indicator are calculated, thereby determining the entropy weight. Information entropy reflects the dispersion of the indicator, and the difference coefficient characterizes the indicator's sample discrimination ability; the stronger the sample discrimination ability, the greater the assigned weight. Finally, the AHP weights and entropy weights are weighted and fused according to a preset fusion coefficient to obtain the final weight vector. This method of integrating expert experience and objective data effectively avoids the problem of poor weight stability caused by the excessive reliance on subjective judgment in traditional AHP.

[0123] Next, as Figure 5 As shown, the evaluation index data collected from the "demonstration area" project are input into a pre-trained neural network performance evaluation model for performance assessment. This model is based on a three-layer backpropagation neural network, where the number of input nodes is equal to the number of evaluation indicators, and the number of hidden layer nodes is determined by a specific expression. As the activation function of the hidden layer, in order to overcome the problem of traditional backpropagation neural networks easily getting trapped in local optima and unstable training results during the model training phase, this method uses a genetic algorithm to globally optimize the initial weights and thresholds of the neural network. Specifically, the weights and thresholds of the neural network are encoded into a one-dimensional vector as a chromosome, and the inverse of the mean squared error of the training set is used as the fitness function. Convergence training is achieved by setting the population size, maximum number of generations, crossover probability, mutation probability, and elite retention ratio. After convergence, the weights and thresholds corresponding to the optimal individuals are used as the initial parameters of the neural network, and the parameters are then adjusted based on the backpropagation algorithm to obtain the trained neural network performance evaluation model. This training method combined with a genetic algorithm significantly improves the model's global optimization ability and stability, avoiding fluctuations caused by random initialization.

[0124] The performance evaluation result output by the neural network model is a continuous predicted score (e.g., 0.75). For ease of understanding and interpretation, this predicted score is discretized into multiple level categories according to preset level thresholds, such as "Excellent," "Good," "Average," and "Poor." Subsequently, a decision tree model is constructed using the various evaluation indicators in the evaluation index system as splitting attributes. As shown in Table 1, information gain is used as the node splitting index during the construction process; information gain measures the reduction in information entropy before and after level division. Starting from the root node, all possible thresholds of the evaluation indicators are traversed for each node, and the attribute with the largest information gain and the corresponding threshold combination are selected for splitting, recursively constructing subtrees. To prevent overfitting, cost complexity pruning is used, and the optimal complexity parameter is selected through cross-validation.

[0125] Table 1 Information Gain of Each Evaluation Indicator

[0126] Finally, as Figure 6As shown, the system uses a constructed decision tree model to score and interpret the effectiveness evaluation results of the "demonstration area" project. For example, if the project is rated as "good," the decision tree clearly shows its judgment path, i.e., the indicators, thresholds, and comparison results used in the complete path from the root node to the leaf node (e.g., "vegetation coverage is higher than 0.8 and the growth rate of agricultural output per mu is higher than 0.1"). This solves the "black box" problem of traditional neural network models, making the evaluation results traceable and understandable. Simultaneously, the system performs boundary analysis, calculating the distance between the current indicator value and the threshold of the adjacent level, identifying the indicator closest to the previous level threshold and its gap (e.g., "rural residents' per capita net income" is only 500 yuan away from the "excellent" level threshold). Furthermore, based on the first appearance position and information gain value of each attribute in the decision tree, the system ranks the importance of the evaluation indicators, helping managers understand which indicators have the greatest impact on the final rating.

[0127] Furthermore, based on the rule structure of the decision tree, the system can deduce actionable optimization suggestions. For example, for a project rated "Good," the system calculates the gap between the current indicator value and the threshold corresponding to the "Excellent" level node by node along its judgment path, and provides a quantitative suggestion for the path gap (e.g., increasing "rural residents' per capita net income" by 500 yuan can achieve the "Excellent" level). Simultaneously, the system can also identify constraints that require simultaneous improvement of multiple evaluation indicators to achieve a level jump, providing suggestions for coordinated improvement of multiple indicators (e.g., simultaneously increasing "industrial investment intensity" and "number of new jobs" can make the project jump from "Good" to "Excellent"). Finally, combining the magnitude of the gaps in each evaluation indicator and the feasibility of improvement, the system provides suggestions for prioritizing improvements to guide subsequent rectification and optimization work.

[0128] To ensure the long-term effectiveness and accuracy of the model, this method also includes a dynamic retraining mechanism. For example, when data from remediation projects that have been completed and accepted within a preset time period are collected and added to the training set after data verification and preprocessing, retraining of the neural network performance evaluation model will be triggered. In addition, when the cumulative number of corrected samples reviewed by experts reaches a preset threshold, or when the consistency rate between the prediction results of the neural network performance evaluation model and the expert evaluation results is lower than a preset level within multiple preset time periods, retraining of the model will also be triggered, thereby ensuring that the model can adapt to new data patterns and evaluation standards, and maintain its high accuracy and robustness.

[0129] This embodiment also provides an efficiency evaluation device for comprehensive land consolidation, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0130] This embodiment provides a device for evaluating the effectiveness of comprehensive land consolidation, such as... Figure 7 As shown, it includes: Module 701 is used to construct an evaluation index system and determine the weight of each evaluation index in the evaluation index system. The evaluation index system includes multiple primary indicators, and each primary indicator includes multiple secondary indicators. Evaluation module 702 is used to evaluate the performance of the input evaluation index based on the trained neural network performance evaluation model and obtain the performance evaluation result. The neural network performance evaluation model includes a three-layer backpropagation neural network. The partitioning module 703 is used to discretize the performance evaluation results into multiple level categories, and to construct a decision tree model using the evaluation indicators in the evaluation indicator system as the splitting attributes. The interpretation module 704 is used to score and interpret the performance evaluation results based on the decision tree model.

[0131] In some alternative implementations, the construction module 701 includes: The first unit is used to classify the remediation project into various remediation models based on the remediation implementation content. The remediation models include farmland improvement, village remediation, ecological restoration, and industry introduction. Each remediation project must include at least one remediation model. The remediation implementation content is the recorded content in the project approval document. The second unit is used to determine the corresponding weight vector in the evaluation index system according to different remediation models; The third unit is used to mark the corresponding evaluation indicator as inapplicable in the evaluation indicator system when there is no corresponding evaluation indicator in the rectification model, and to allocate the weight of the evaluation indicator to other evaluation indicators of the same level.

[0132] The comprehensive land consolidation efficiency evaluation device provided in this application embodiment can execute the comprehensive land consolidation efficiency evaluation method provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0133] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0134] The following is a detailed reference. Figure 8 The diagram illustrates a structural schematic suitable for implementing the electronic device described in the embodiments of this application. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 801, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 802 or a program loaded from memory 808 into random access memory (RAM) 803. The RAM 803 also stores various programs and data required for the operation of the electronic device. The processor 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0135] Typically, the following devices can be connected to I / O interface 805: input devices 806 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 807 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 808 including, for example, magnetic tapes, hard disks, etc.; and communication devices 809. Communication device 809 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 8 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0136] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 809, or installed from a memory 808, or installed from a ROM 802. When the computer program is executed by the processor 801, it performs the functions defined in the method for evaluating the effectiveness of comprehensive land consolidation according to embodiments of this application.

[0137] Figure 8 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0138] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the effectiveness evaluation method for comprehensive land consolidation shown in the above embodiments is implemented.

[0139] A portion of this application can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0140] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and all such modifications and variations fall within the scope defined by the appended claims.

Claims

1. A method for evaluating the effectiveness of comprehensive land consolidation, characterized in that, The method includes: An evaluation index system is constructed, and the weight of each evaluation index in the evaluation index system is determined. The evaluation index system includes multiple primary indicators, and each primary indicator includes multiple secondary indicators. Based on the trained neural network performance evaluation model, the performance of the input evaluation index is evaluated to obtain the performance evaluation result. The neural network performance evaluation model includes a three-layer backpropagation neural network. The performance evaluation results are discretized into multiple level categories, and a decision tree model is constructed using the evaluation indicators in the evaluation indicator system as the splitting attributes. The performance evaluation results are scored and interpreted based on the decision tree model.

2. The method according to claim 1, characterized in that, Determining the weight of each evaluation indicator in the evaluation indicator system includes: Based on the content of the remediation implementation, the remediation projects are divided into multiple remediation models, including farmland improvement, village remediation, ecological restoration, and industry introduction. Each remediation project shall include at least one remediation model. According to the different remediation models, the corresponding weight vectors are determined in the evaluation index system; When the corresponding evaluation indicator does not exist in the rectification mode, the corresponding evaluation indicator is marked as inapplicable in the evaluation indicator system, and the weight of the evaluation indicator is allocated to other evaluation indicators of the same level.

3. The method according to claim 2, characterized in that, The step of determining the corresponding weight vector in the evaluation index system according to different remediation models includes: Based on the evaluation indicators in the evaluation indicator system, construct a judgment matrix; The weight vector of the constructed judgment matrix is ​​calculated and a consistency check is performed based on expert opinions to obtain the hierarchical analysis weights. Based on the data information recorded in the remediation project, the information entropy and difference coefficient of each evaluation indicator in the evaluation indicator system are calculated, and the entropy weight is determined based on the information entropy and difference coefficient. The difference coefficient is used to characterize the sample discrimination ability of the evaluation indicator, and the sample discrimination ability is positively correlated with the corresponding assigned weight. The weight vector is obtained by weighting and fusing the hierarchical analysis weights and the entropy weights according to a preset fusion coefficient.

4. The method according to claim 1, characterized in that, The neural network performance evaluation model is constructed based on a three-layer backpropagation neural network. The number of input nodes in the model is equal to the number of evaluation indicators. The number of hidden layer nodes is determined by the following expression: ,in The number of nodes in the input layer. This represents the number of nodes in the output layer. To adjust the parameters, the hidden layer activation function of the neural network performance evaluation model is: ; The training of the neural network performance evaluation model includes: The initial weights and thresholds of the three-layer backpropagation neural network are globally optimized using a genetic algorithm to obtain the trained neural network performance evaluation model.

5. The method according to claim 4, characterized in that, The step of globally optimizing the initial weights and thresholds of the three-layer backpropagation neural network using a genetic algorithm to obtain the trained neural network performance evaluation model includes: Using real-number encoding, the weights and thresholds of the three-layer backpropagation neural network are concatenated into a one-dimensional vector as a chromosome, where the chromosome length is n. i ×n h +n h ×n0+n h +n0, where n i n is the number of nodes in the input layer, n0 is the number of nodes in the output layer, and n h This represents the number of hidden layer nodes. The reciprocal of the mean square error of the training set is taken as the fitness function; After configuring the population size, maximum number of generations, crossover probability, mutation probability, and elite retention ratio, convergence training is performed. After convergence, the weights and thresholds corresponding to the optimal individuals representing the evaluation indicators are used as the initial parameters of the three-layer backpropagation neural network. The parameters are adjusted based on the backpropagation algorithm to obtain the trained neural network performance evaluation model.

6. The method according to claim 1, characterized in that, The performance evaluation results include a predictive score. The step of discretizing the performance evaluation results into multiple level categories and constructing a decision tree model using the evaluation indicators in the evaluation indicator system as splitting attributes includes: The predicted scores are divided into multiple categories according to a grade threshold. Information gain is used as the node splitting metric, where the information gain characterizes the reduction in information entropy before and after the level split, and the information gain includes Gain(D,A) = H(D) - H(D|A), where... For the training sample set, For information entropy, express For feature attributes of the training set The information gain, where A is the candidate feature attribute; Starting from the root node, iterate through all possible thresholds of evaluation indicators for each node, and select the attribute with the largest information gain and the corresponding threshold combination for splitting, recursively constructing subtrees. The stopping conditions for recursion include the number of samples in a node being lower than a preset minimum value, the tree depth reaching a preset upper limit, and a preset number of samples in a node belonging to the same category.

7. The method according to claim 1, characterized in that, The scoring and interpretation of the performance evaluation results based on the decision tree model includes: The judgment path of the performance evaluation results is explained, wherein the judgment path represents the indicators, thresholds and comparison results used for the complete path from the root node to the leaf node; The performance evaluation results are interpreted using a critical boundary analysis, wherein the critical boundary analysis is used to calculate the distance between the current indicator value and the adjacent level threshold, in order to identify the indicator closest to the previous level threshold and the gap. The importance ranking of the indicators in the performance evaluation results is interpreted, wherein the importance ranking is determined based on the position of the first occurrence of each attribute in the decision tree and the information gain value.

8. The method according to claim 1, characterized in that, The method further includes: Based on the rule structure of the decision tree, actionable optimization suggestions are derived; wherein, the optimization suggestions include path gap quantification suggestions, multi-indicator synergistic improvement suggestions, and improvement priority ranking suggestions; The path gap quantification suggestion is used to characterize the project with the first rating, and calculate the gap between the current indicator value and the corresponding threshold of the second rating node by node along its judgment path, where the second rating is better than the first rating. The proposed multi-indicator collaborative improvement suggestions are used to characterize the constraints that require simultaneous improvement of multiple evaluation indicators to achieve a leap in level. The improvement priority ranking suggestion is given by taking into account the size of the gap between various evaluation indicators and the feasibility of improvement.

9. The method according to claim 1, characterized in that, The method further includes: Data on remediation projects that have been completed and accepted within a preset time period are collected, and after data verification and preprocessing, they are added to the training set to trigger retraining of the neural network performance evaluation model. When the cumulative number of corrected samples reviewed by experts reaches a preset threshold, or when the consistency rate between the prediction results of the neural network performance evaluation model and the expert evaluation results is lower than a preset level within multiple preset time periods, the neural network performance evaluation model is retrained.

10. A performance evaluation device for comprehensive land consolidation, characterized in that, The device includes: A construction module is used to construct an evaluation index system and determine the weight of each evaluation index in the evaluation index system, wherein the evaluation index system includes multiple primary indicators, and each primary indicator includes multiple secondary indicators. An evaluation module is used to evaluate the efficiency of the input evaluation index based on the trained neural network efficiency evaluation model, and obtain the efficiency evaluation result. The neural network efficiency evaluation model includes a three-layer backpropagation neural network. The partitioning module is used to discretize the performance evaluation results into multiple level categories, and to construct a decision tree model using the evaluation indicators in the evaluation indicator system as the splitting attributes. An interpretation module is used to score and interpret the performance evaluation results based on the decision tree model.

Images