Seasonal river pollution control engineering effect evaluation and environmental protection technology popularization service system

Abstract

The application discloses a seasonal river pollution control engineering effect evaluation and environmental protection technology popularization service system and relates to the technical field of environmental monitoring; the system comprises the following modules: a matter-element extension evaluation module, which determines the comprehensive evaluation grade of the pollution control engineering according to a matter-element correlation matrix; a projection pursuit weak link identification module, which generates an improvement priority sequence by sorting the projection values of each weak link index along the optimal projection direction vector from large to small; and a multi-objective optimization popularization module, which is used for setting candidate environmental protection technologies for each weak link index according to the improvement priority sequence and screening a final environmental protection technology popularization service scheme from a Pareto optimal solution set. The application realizes a complete decision link from quantitative evaluation to accurate positioning of weak links and then to multi-objective balanced optimization of a popularized technology scheme, and overcomes the limitation of traditional methods that can only output a comprehensive evaluation grade and cannot identify specific shortcomings and improvement priorities.

Description

Technical Field

[0001] This invention relates to the field of big data analysis technology, and more particularly to the field of environmental monitoring and processing. Specifically, it relates to a system for evaluating the effectiveness of seasonal river pollution control projects and promoting environmental protection technologies, which optimizes strategies for evaluating the effectiveness of seasonal river pollution control projects and promoting environmental protection technologies through big data analysis. Background Technology

[0002] Seasonal rivers, influenced by climate and precipitation cycles, exhibit significant seasonal fluctuations in runoff between high and low water seasons, with some sections even experiencing complete flow interruption during the dry season. These drastic changes in hydrological conditions result in marked differences in the river's self-purification capacity, pollutant migration and diffusion characteristics, and ecosystem carrying capacity across different seasons, presenting unique technical challenges to the design, operation, and effectiveness evaluation of pollution control projects along rivers, distinct from those for perennial rivers. In recent years, with the continuous increase in national investment in comprehensive watershed water environment management, numerous projects such as sewage interception and collection, constructed wetlands, ecological floating beds, sediment dredging, and aeration have been implemented along seasonal rivers. How to scientifically and quantitatively evaluate the comprehensive effects of these projects and efficiently promote mature management experiences to other watersheds has become a crucial issue urgently needing to be addressed in the field of water environment management.

[0003] Currently, methods for evaluating the effectiveness of river pollution control projects mainly focus on traditional approaches such as the single-factor index method, the comprehensive pollution index method, and the fuzzy comprehensive evaluation method. The single-factor index method determines the water quality level based on the worst single indicator. While simple to operate, it ignores the synergistic relationship between indicators, making it prone to distortion of the overall evaluation result due to individual outliers. The comprehensive pollution index method obtains a comprehensive score by weighted averaging of multiple water quality indicators, reflecting the overall situation of multiple indicators to some extent. However, its evaluation result is a continuous value, lacking clear level classification ability and failing to intuitively answer the core question of "what level is the current project effectiveness at?" The fuzzy comprehensive evaluation method introduces a membership function to fuzzily distinguish water quality levels. However, when the measured value of an indicator falls precisely in the boundary region between two levels, the choice of membership function has a significant impact on the evaluation result, and it cannot effectively handle extreme cases where indicator values ​​exceed the preset range of the evaluation system. Furthermore, the above methods generally rely on the pre-setting of indicator weights. The introduction of subjective weighting methods such as the analytic hierarchy process in determining weights can easily lead to a lack of objectivity and repeatability in the evaluation results due to differences in expert experience. Summary of the Invention

[0004] The purpose of this invention is to provide a system for evaluating the effectiveness of seasonal river pollution control projects and promoting environmental protection technologies. It realizes a complete decision-making chain from quantitative assessment to precise identification of weak links and then to multi-objective balanced optimization of promotion technology solutions. It overcomes the limitations of traditional methods that only output comprehensive evaluation levels but cannot identify specific shortcomings and improvement priorities. At the same time, it avoids the shortcomings of environmental protection technology promotion that rely on qualitative experience recommendations and lack systematic quantitative optimization support.

[0005] To address the aforementioned technical problems, this invention provides a system for evaluating the effectiveness of seasonal river pollution control projects and promoting environmental protection technologies, comprising: The matter-element extension evaluation module is used to collect current monitoring values ​​of multiple evaluation indicators for seasonal river pollution control projects. It constructs matter-element objects to be evaluated using each evaluation indicator as a feature and the current monitoring value as a feature value. Based on the classical domain and section domain corresponding to multiple preset evaluation levels, it calculates the correlation degree of each evaluation indicator with respect to each evaluation level according to the matter-element extension correlation function, forming a matter-element correlation degree matrix. Based on the matter-element correlation degree matrix, it determines the comprehensive evaluation level of the pollution control project. The projection tracking weak link identification module is used to select the evaluation index with the maximum correlation value of the level to be improved from the matter-element correlation matrix as the weak link index. The module constructs an input data matrix with the correlation values ​​of all weak link indices to each evaluation level and performs standardization processing. The module performs projection tracking dimensionality reduction on the standardized input data matrix and searches for the optimal projection direction vector that maximizes the function value of the projection index through an optimization algorithm. The module generates an improvement priority sequence based on the projection values ​​of each weak link index along the optimal projection direction vector from largest to smallest. The multi-objective optimization and promotion module is used to set candidate environmental protection technologies for each weak link indicator according to the improvement priority sequence. The technology selection corresponding to each weak link indicator is used as the decision variable, and the three optimization objectives are insufficient technology coverage, promotion and implementation cost and insufficient technology adaptation. Under the preset constraints, the NSGA-II multi-objective genetic algorithm is used to iteratively optimize through non-dominated sorting and crowding distance selection, outputting the Pareto optimal solution set, and selecting the final environmental protection technology promotion service plan from the Pareto optimal solution set.

[0006] Furthermore, the evaluation indicators include nine benefit-oriented indicators: chemical oxygen demand reduction rate, ammonia nitrogen reduction rate, total phosphorus reduction rate, aquatic biodiversity recovery index, riparian vegetation coverage rate, sediment pollutant reduction rate, facility integrity rate, treatment capacity compliance rate, and operation and maintenance response timeliness rate.

[0007] Furthermore, the Matter-Element Extensibility Evaluation Module divides the effectiveness of pollution control engineering into four evaluation levels: excellent, good, qualified, and needing improvement. For each evaluation level, the characteristic value range of each evaluation indicator is pre-set as the classical domain, and the complete value range after merging the classical domains of the four evaluation levels is taken as the section domain.

[0008] Furthermore, when calculating the correlation between each evaluation indicator and each evaluation level, the Matter-Element Extension Evaluation Module first calculates the midpoint value and half-width of the classical domain interval, and then determines the correlation according to the following three cases: If the current monitoring value is within the classical domain interval, the correlation is equal to 1 minus the absolute value of the difference between the current monitoring value and the midpoint value of the classical domain interval divided by the half-width of the classical domain interval; If the current monitoring value is outside the classical domain interval but within the section domain, the distance from the current monitoring value to the boundary of the classical domain interval is first calculated, and then the distance is divided by the difference between the section domain interval width and the classical domain interval width, and the negative value is taken as the correlation. If the difference between the section domain interval width and the classical domain interval width is less than the preset minimum difference threshold, the minimum difference threshold is used instead; If the current monitoring value exceeds the section domain range, the correlation is set to negative 1.

[0009] Furthermore, the matter-element extension evaluation module arranges the correlation degree of each evaluation indicator to each evaluation level in the order of indicator and level to form a matter-element correlation degree matrix. The matter-element correlation degree matrix is ​​summed column by column to obtain the comprehensive correlation degree value of each evaluation level. The evaluation level with the largest comprehensive correlation degree value is selected as the comprehensive evaluation level.

[0010] Furthermore, the projection tracking weak link identification module compares the correlation values ​​of each evaluation index in the matter-element correlation matrix with each evaluation level. If the correlation value of a certain evaluation index with the level to be improved is the maximum value among its correlation values ​​with each evaluation level, then the evaluation index is marked as a weak link index. An input data matrix is ​​constructed using the correlation values ​​of all weak link indices with each evaluation level. The input data matrix is ​​then subjected to range standardization processing by column. Specifically, for each column of data, each element is subtracted from the minimum value of that column and then divided by the difference between the maximum value and the minimum value of that column.

[0011] Furthermore, when the projection tracing weak link identification module performs projection tracing dimensionality reduction, it projects each row of the standardized input data matrix along the candidate projection direction vector. The projection value of each weak link index is equal to the sum of the standardized correlation values ​​of the corresponding row multiplied by each component of the candidate projection direction vector. The projection index function value is equal to the product of the average local density and the dispersion of all weak link index projection values. The average local density is calculated by using 0.1 times the standard deviation of all projection values ​​as the window radius, counting the number of other projection values ​​within the window radius for each projection value, summing all the numbers, and then dividing by the total number of weak link indices. The dispersion is the square root of the sum of the squares of the differences between each projection value and the mean of the projection values.

[0012] Furthermore, the projection tracking weak link identification module uses a particle swarm optimization algorithm to search for the optimal projection direction vector. In each iteration, the position vector of each particle is normalized to make the modulus equal to 1, and then the corresponding projection index function value is calculated. The particle velocity and position are updated based on the individual optimal position and the global optimal position. After the iteration is completed, the global optimal position is used as the optimal projection direction vector. The projection values ​​of each weak link index are recalculated along the optimal projection direction vector and then sorted in descending order of projection values ​​to generate an improvement priority sequence.

[0013] Furthermore, the multi-objective optimization and promotion module sets no more than 5 candidate environmental protection technologies and their corresponding promotion cost and technology suitability values ​​for each weak link indicator. Each weak link indicator corresponds to one decision variable, with values ​​ranging from 0 to the number of corresponding candidate environmental protection technologies. 0 indicates that no technology promotion measures will be taken for the corresponding weak link. The three optimization objectives are unified in the direction of minimization: the technology coverage insufficiency is equal to 1 minus the ratio of the number of decision variables with values ​​greater than 0 to the total number of all weak link indicators; the promotion implementation cost is equal to the sum of the promotion cost values ​​of the candidate environmental protection technologies corresponding to all decision variables with values ​​greater than 0; and the technology suitability insufficiency is equal to 1 minus the arithmetic mean of the technology suitability values ​​corresponding to all decision variables with values ​​greater than 0 when there are decision variables with values ​​greater than 0, and is set to 1 when all decision variables have values ​​of 0.

[0014] Furthermore, the constraints set by the multi-objective optimization promotion module include that the promotion implementation cost does not exceed the preset budget limit and that the decision variables corresponding to the top-ranked weak link indicators in the improvement priority sequence are greater than 0, with the number of top-ranked indicators being the smaller of the total number of weak link indicators and 3. When performing NSGA-II iteration, the parent and offspring populations are merged and non-dominated sorting is performed. The next generation population is filled in order of increasing non-dominated level, and individuals within the same non-dominated level are selected in order of decreasing crowding distance. After the iteration is completed, for each individual in the Pareto optimal solution set, the three optimization objective values ​​are standardized by range and then added together to obtain a comprehensive inferiority score. The decision variable corresponding to the individual with the smallest comprehensive inferiority score is decoded and output as the final environmental protection technology promotion service plan.

[0015] The seasonal river pollution control engineering effectiveness evaluation and environmental protection technology promotion service system of the present invention has the following beneficial effects: This invention uses matter-element extension correlation functions to calculate the correlation degree of each evaluation index with respect to each evaluation level. By setting classical domain and node domain and using distance measurement mechanism, it can not only determine which evaluation level each index belongs to, but also quantify the matching quality of index values ​​within the level range. At the same time, it provides clear correlation degree calculation rules for extreme cases where index values ​​fall into the level boundary area and exceed the evaluation range, thus overcoming the problem of insufficient processing ability in the boundary area of ​​traditional fuzzy comprehensive evaluation method.

[0016] This invention integrates assessment indicators from three dimensions—water quality reduction, ecological restoration, and engineering operation and maintenance—into a matter-element correlation matrix for comprehensive evaluation, avoiding the shortcomings of focusing solely on water quality and resulting in biased assessment conclusions. By comparing the correlation values ​​of each assessment indicator at each assessment level, this invention automatically identifies the weakest link indicators with the highest degree of matching to the required improvement level. It then uses a projection pursuit method to reduce the correlation information of the weak link indicators across multiple levels to a one-dimensional projection value. Based on the magnitude of the projection value, an improvement priority sequence is directly generated, providing decision-makers with a quantitative basis for ranking weak links. This overcomes the shortcomings of traditional methods that only output a comprehensive evaluation level without further identifying specific weaknesses and determining improvement priorities.

[0017] This invention considers three optimization objectives simultaneously: insufficient technology coverage, promotion and implementation costs, and insufficient technology adaptation. Under budget constraints and mandatory coverage constraints for key weak links, it uses the NSGA-II multi-objective genetic algorithm for global search, outputs a Pareto optimal solution set, and selects the final solution through a comprehensive inferiority score. This achieves balanced optimization decision-making in multi-objective conflict situations and overcomes the limitations of existing environmental protection technology promotion that relies on qualitative experience recommendations and lacks systematic quantitative optimization support. Attached Figure Description

[0018] Figure 1 A schematic diagram illustrating the variation of the matter-element extension correlation function across the entire range of evaluation index values, as provided in this embodiment of the invention. Figure 2 A schematic diagram of the correlation matrix and weak link identification process based on the matter-element extension theory provided in this embodiment of the invention; Figure 3 This is a schematic diagram of the hierarchical results of the non-dominated sorting mechanism in the NSGA-II multi-objective genetic algorithm provided in this embodiment of the invention; Figure 4 This is a schematic diagram showing the distribution of the Pareto optimal solution set obtained after convergence of the NSGA-II multi-objective optimization algorithm in the objective space, as provided in an embodiment of the present invention. Detailed Implementation

[0019] A system for evaluating the effectiveness of seasonal river pollution control projects and promoting environmental protection technologies, including: The matter-element extension evaluation module is used to collect current monitoring values ​​of multiple evaluation indicators for seasonal river pollution control projects. It constructs matter-element objects to be evaluated using each evaluation indicator as a feature and the current monitoring value as a feature value. Based on the classical domain and section domain corresponding to multiple preset evaluation levels, it calculates the correlation degree of each evaluation indicator with respect to each evaluation level according to the matter-element extension correlation function, forming a matter-element correlation degree matrix. Based on the matter-element correlation degree matrix, it determines the comprehensive evaluation level of the pollution control project. The projection tracking weak link identification module is used to select the evaluation index with the maximum correlation value of the level to be improved from the matter-element correlation matrix as the weak link index. The module constructs an input data matrix with the correlation values ​​of all weak link indices to each evaluation level and performs standardization processing. The module performs projection tracking dimensionality reduction on the standardized input data matrix and searches for the optimal projection direction vector that maximizes the function value of the projection index through an optimization algorithm. The module generates an improvement priority sequence based on the projection values ​​of each weak link index along the optimal projection direction vector from largest to smallest. The multi-objective optimization and promotion module is used to set candidate environmental protection technologies for each weak link indicator according to the improvement priority sequence. The technology selection corresponding to each weak link indicator is used as the decision variable, and the three optimization objectives are insufficient technology coverage, promotion and implementation cost and insufficient technology adaptation. Under the preset constraints, the NSGA-II multi-objective genetic algorithm is used to iteratively optimize through non-dominated sorting and crowding distance selection, outputting the Pareto optimal solution set, and selecting the final environmental protection technology promotion service plan from the Pareto optimal solution set.

[0020] In the implementation of the seasonal river pollution control project effectiveness evaluation, it is first necessary to obtain the current monitoring values ​​reflecting the project's treatment effect. Considering the characteristics of seasonal river pollution control projects, this invention selects nine evaluation indicators covering three dimensions: water quality improvement, ecological restoration, and project operation and maintenance. These indicators are: chemical oxygen demand (COD) reduction rate, ammonia nitrogen reduction rate, total phosphorus reduction rate, aquatic biodiversity recovery index, riparian vegetation coverage rate, sediment pollutant reduction rate, facility integrity rate, treatment capacity compliance rate, and timely operation and maintenance response rate. All nine evaluation indicators are benefit-oriented, meaning that higher values ​​indicate better treatment effects. During each evaluation period, current monitoring values ​​for each evaluation indicator are collected through water quality sensor arrays, ecological monitoring stations, and the project operation and maintenance management platform deployed along the river. All current monitoring values ​​are recorded as percentages, ranging from 0 to 100. In an optional implementation, the evaluation period can be set to monthly or quarterly data collection, with the specific period length determined based on the river's seasonal characteristics and the project's operational rhythm.

[0021] After obtaining the current monitoring data, the seasonal river pollution control project is taken as the object to be evaluated. Nine evaluation indicators are used as the characteristics of the matter element, and the current monitoring data corresponding to each evaluation indicator is used as the characteristic value of the matter element to construct the matter element to be evaluated. The matter element is the basic information unit used to describe things in extension theory. It consists of three elements: name, characteristics, and characteristic value. It can express the multidimensional information of the evaluation object in a structured way. Assuming that the current monitoring data collected in a certain evaluation period are as follows: chemical oxygen demand reduction rate 65, ammonia nitrogen reduction rate 58, total phosphorus reduction rate 52, aquatic biodiversity recovery index 45, riparian vegetation coverage rate 70, sediment pollutant reduction rate 48, facility integrity rate 88, treatment capacity compliance rate 76, and operation and maintenance response timeliness rate 82, then the above nine values ​​are filled into the corresponding characteristic value positions of the matter element to be evaluated.

[0022] Next, a grading system is established for the treatment effectiveness of pollution control projects. This invention divides the effectiveness evaluation results into four evaluation levels, from highest to lowest: Excellent, Good, Qualified, and Needs Improvement. Four levels, rather than more or fewer, are chosen because this four-level system effectively distinguishes differences in treatment effectiveness and aligns with the grading conventions commonly used in the domestic environmental protection industry, facilitating horizontal comparisons of evaluation results across different watersheds.

[0023] For each assessment level, a pre-defined range of characteristic values ​​for each assessment indicator within that level is required, known as the classical domain. The classical domain defines the numerical range that an assessment indicator must meet to fall within a specific assessment level. For example, the classical domain for the chemical oxygen demand (COD) reduction rate can be set to 80-100 for the excellent level, 60-80 for the good level, 40-60 for the qualified level, and 0-40 for the level requiring improvement. The classical domains for the remaining eight assessment indicators are determined based on recommended values ​​in national surface water environmental quality standards and industry technical specifications. In an optional implementation, the boundary values ​​of the classical domain can also be determined through consultation among multiple experts in the field based on practical engineering experience.

[0024] refer to Figure 1 In the graph, the horizontal axis represents the quantitative value of the evaluation index, and the vertical axis represents the correlation degree. The value of . In this embodiment, the section domain is set to . Furthermore, the section is divided into four classical domains, each corresponding to a classical domain of the level to be improved. Classical domains with qualified levels Classical domains of good quality And classic domains of excellent quality The central value of each classical domain and half width respectively by and Confirmed, among which and The left and right endpoints represent the classical domain. The figure shows four correlation curves, each corresponding to an evaluation level. When the evaluation index value... Falling into a classical domain Internally, the corresponding correlation degree is based on The calculation shows that the correlation is positive at this point, reaching a maximum of 1 at the center of the classical domain and decaying to 0 at the boundaries of the classical domain. When the index value deviates from the classical domain but remains within the domain... Within the range, the degree of correlation is based on Calculation, where The shortest distance from the index value to the classical domain. The correlation coefficient is the difference between the length of the section domain interval and the length of the classic domain interval. A negative correlation indicates a deviation between the indicator value and that level. The four curves exhibit an alternating distribution, with convex positive values ​​within their respective classic domain intervals and decreasing to negative values ​​outside the classic domains. A vertical dashed line marks an example of a current monitoring value. This line intersects with each of the four correlation curves, and the ordinate of each intersection point represents the correlation coefficient between the monitoring value and the corresponding assessment level. By comparing the correlation coefficients at the four intersection points, the assessment level that best matches the monitoring value is determined as the level with the highest correlation, and the degree to which the monitoring value deviates from other levels can be quantitatively reflected. The bottom labels indicate the interval range of each classic domain and the span of the section domain, visually demonstrating the boundary structure of the level division.

[0025] After determining the classical domains, it is also necessary to define the sub-domains. The sub-domain is the complete value range of each assessment indicator across all four assessment levels, representing the overall possible range of variation for that indicator within the assessment system. Taking the chemical oxygen demand (COD) reduction rate as an example, since the classical domains of the four levels sequentially cover the range from 0 to 100, its sub-domain is 0 to 100. The purpose of the sub-domain is to provide a reference benchmark for subsequent correlation calculations: when the current monitoring value of an assessment indicator deviates from all classical domain ranges, the sub-domain range is used to measure the degree of deviation, thereby providing a reasonable negative correlation value.

[0026] After determining the classical domain and the section domain, the matter-element extension correlation function is used to calculate the correlation degree of the current monitoring value of each evaluation indicator relative to each evaluation level. The correlation degree is a real value used to quantitatively describe the degree of closeness between the current monitoring value and a certain evaluation level. A positive correlation degree, and the larger the value, the more the current monitoring value meets the requirements of that evaluation level; a negative correlation degree, and the larger the absolute value of the negative value, the greater the degree of deviation.

[0027] The specific calculation process is as follows: For the... The evaluation indicator is set to a current monitoring value of . Let it be in the first The left boundary value of the classical domain under each evaluation level is The right boundary value is Let the left boundary value of its domain be... The right boundary value is .in, The value range is 1 to 9, corresponding to 9 evaluation indicators; The value ranges from 1 to 4, corresponding to four evaluation levels. First, the midpoint value of the classical domain interval is calculated. and half width Midpoint value Left boundary value of classical domain With right boundary value Arithmetic mean: ; half width Right boundary value of classical domain With left boundary value Half of the difference: Midpoint value This represents the ideal center position of the evaluation indicator at this evaluation level, half-width. This reflects the permissible range of fluctuation for that level.

[0028] Then, the current monitoring values ​​were calculated. Distance to the classical domain .distance The meaning is to measure the degree to which the current monitoring value deviates from the classical range. Its calculation is based on the relative positional relationship between the current monitoring value and the classical range, and is divided into three cases. When the current monitoring value... Located inside the classical domain interval, i.e. At time, distance The calculation method is as follows: ;at this time The value is non-positive because the current monitoring value lies within the interval, and logically, its distance from the interval should not be positive. Less than the left boundary value of the classical domain At time, distance Left boundary value Subtract the current monitoring value Current monitoring values Greater than the right boundary value of the classical domain At time, distance Current monitoring data Subtract the right boundary value In the latter two cases All values ​​are positive, indicating that the current monitoring value has fallen outside the classic domain of this level.

[0029] In obtaining distance Then, the correlation degree is calculated based on its positive or negative value. .when When the current monitoring value falls within the classical domain interval, the correlation is... The calculation method is as follows: At this point, the correlation coefficient ranges from 0 to 1. This occurs when the current monitored value is exactly equal to the midpoint value of the classic domain interval. When the correlation reaches its maximum value of 1, it indicates a perfect match with the evaluation level; when the current monitoring value is located at the boundary of the classic domain interval, the correlation is 0. This calculation method allows the correlation not only to determine whether the current monitoring value falls into a certain level interval, but also to further distinguish the degree of matching within the interval, thus providing a more refined evaluation result than simple threshold judgment.

[0030] when When the current monitoring value falls outside the classical domain interval, it is necessary to further calculate the negative correlation using segment domain information. In this case, the difference between the segment domain interval width and the classical domain interval width is first calculated. : ;in The width of the section interval, This represents the width of the classical domain interval. Difference This indicates that even if the current monitoring value deviates from the classic domain of this level, it may still belong to the activity space of other levels. If If the value is too small, it will lead to an abnormal amplification in the correlation calculation results, with the divisor approaching zero. Therefore, a minimum difference threshold is set for protective truncation. When the difference is less than the minimum difference threshold, Replace with the minimum difference threshold. In one optional implementation, the minimum difference threshold can be set to 0.001. After the above protective truncation is completed, the correlation... The calculation method is as follows: At this point, the correlation is negative. The larger the absolute value, the greater the deviation of the current monitoring value from the assessment level.

[0031] Another extreme scenario exists: when the current monitoring data... Beyond the section range, i.e. or When this happens, it means that the value has completely fallen outside the measurable range of the evaluation system, and at this point, the correlation degree is directly... Set to -1 as a uniform minimum correlation marker.

[0032] The correlation calculation process described above was repeated for each of the nine evaluation indicators, resulting in four correlation values ​​for each of the four evaluation levels. Taking the current monitoring value of 65 for the chemical oxygen demand reduction rate in the aforementioned example data as an example, this value falls within the classic range of 60 to 80 for the good level, therefore its correlation with the good level is positive; at the same time, this value deviates from the classic range of 80 to 100 for the excellent level, so its correlation with the excellent level is negative; this value also deviates from the classic range of 40 to 60 for the qualified level, so its correlation with the qualified level is also negative, but the absolute value is smaller because 65 is closer to the upper boundary of the qualified level of 60.

[0033] The correlation values ​​of all nine evaluation indicators with respect to the four evaluation levels are organized into a matrix according to the order of the evaluation indicators and the order of the evaluation levels, forming a 9-row, 4-column matter-element correlation matrix. The matrix's... Line number The column element is the first one. Evaluation indicator relative to the first Correlation between assessment levels The object-element correlation matrix, with its compact two-dimensional structure, comprehensively records the proximity relationships between the evaluated object and each level across all dimensions, providing a unified data foundation for subsequent comprehensive level determination and weak link identification.

[0034] When determining the comprehensive evaluation level, the matter-element correlation matrix is ​​summed column-by-column. Specifically, all nine correlation values ​​in the first column of the matrix are summed to obtain the comprehensive correlation value for the excellent level; all nine values ​​in the second column are summed to obtain the comprehensive correlation value for the good level; and so on, to obtain the comprehensive correlation values ​​for the qualified level and the level requiring improvement. Summing column-by-column means that each evaluation indicator contributes equally to the comprehensive evaluation level, eliminating the need for additional indicator weights and avoiding potential subjective biases in weight determination. Among the four comprehensive correlation values, the evaluation level corresponding to the largest comprehensive correlation value is selected as the comprehensive evaluation level for the seasonal river pollution control project in the current period. The largest comprehensive correlation value indicates the highest degree of matching between the current project's overall treatment effect and that level.

[0035] refer to Figure 2 The figure presents the correlation calculation results of nine assessment indicators across four assessment levels in matrix form. The rows of the matrix represent the following nine indicators: Chemical Oxygen Demand (COD) reduction rate, ammonia nitrogen reduction rate, and total phosphorus reduction rate under the water quality reduction dimension; aquatic biodiversity recovery index, riparian vegetation coverage rate, and sediment pollutant reduction rate under the ecological restoration dimension; and facility integrity rate, treatment capacity compliance rate, and timely operation and maintenance response rate under the engineering operation and maintenance dimension. The columns of the matrix correspond to the four assessment levels: Excellent, Good, Satisfactory, and Needs Improvement. Each matrix cell contains the correlation value between the corresponding indicator and the corresponding level. ,in For the first The actual monitoring value of the indicator This is used to evaluate the level sequence number. A positive correlation value indicates that the indicator value falls within the classical domain of that level, while a negative correlation value indicates that the indicator value deviates from the classical domain of that level. For each row of indicators, the matrix marks the cell containing the maximum correlation value. When an indicator achieves the maximum correlation in the "Improvement Level" column, it indicates that the indicator's current performance is closest to the "Improvement Level," meaning that the indicator is identified as a weak link indicator and is marked with an arrow on the right side of the matrix. In this embodiment, the correlation of ammonia nitrogen reduction rate is... The correlation between total phosphorus reduction rate and total phosphorus reduction rate is The correlation between the aquatic biodiversity recovery index and the aquatic biodiversity recovery index is: The correlation between sediment pollutant reduction rate and sediment pollutant reduction rate is The indicators that achieve the maximum value within their respective rows in the "Needs Improvement" column are therefore identified as weak links. The remaining indicators achieve the highest correlation in the "Excellent" or "Good" columns. The left side of the matrix groups and labels the nine indicators according to three dimensions: water quality reduction, ecological restoration, and engineering operation and maintenance. The bottom of the matrix also summarizes the overall correlation of each level column, used to determine the comprehensive evaluation level of the overall project effectiveness.

[0036] In one optional implementation, when two evaluation levels have the same overall correlation value and both are the maximum value, the evaluation level with the lower level number (i.e., better performance) is selected as the overall evaluation level to encourage the project to continuously strive for higher standards. In another optional implementation, a weighted summation can be used instead of direct summation by column to calculate the overall correlation value. In this case, the weight of each evaluation indicator can be determined by the current monitoring data based on the coefficient of variation method or the entropy method.

[0037] After completing the construction of the matter-element correlation matrix and determining the comprehensive evaluation level, it is necessary to further identify the specific links in the current pollution control project that have obvious deficiencies from the matter-element correlation matrix, and prioritize these weak links according to the urgency of improvement, so as to formulate targeted environmental protection technology promotion plans in the future.

[0038] The selection criteria for weak link indicators are based on the hierarchical information inherent in the matter-element correlation matrix itself. The matter-element correlation matrix has a 9-row, 4-column structure, with each row corresponding to one evaluation indicator and each column corresponding to one evaluation level. For a given evaluation indicator, the evaluation level corresponding to the highest correlation value among the four evaluation levels reflects the level to which the indicator is most likely to be assigned. If the correlation value of an evaluation indicator for the level to be improved is exactly the highest among its correlation values ​​for the four evaluation levels, it indicates that the current performance of this indicator matches the level to be improved the most, meaning that the engineering governance effect reflected by this indicator is in the worst hierarchical state and needs to be focused on for improvement.

[0039] The specific screening process is as follows: scan the matter-element correlation matrix row by row, for the first row... row (i.e., the first row) (Evaluation indicators), compare their correlation values ​​across 4 columns. , , , ,in to The correlation between these levels corresponds to the grades of Excellent, Good, Satisfactory, and Need for Improvement, respectively. If... If the value is the largest among these four values, then the first one will be... Each of the nine evaluation indicators is marked as a weak link. After iterating through all nine evaluation indicators, the number of indicators marked as weak links is counted and denoted as follows: .like A value of 0 indicates that none of the evaluation indicators fall into the "needs improvement" category. In this case, the previously determined comprehensive evaluation level is directly output as the final evaluation result, and the subsequent process ends. If If the value is greater than 0, then it is necessary to continue prioritizing these weak links.

[0040] Let's continue with the example data from before. Assume that among the nine assessment indicators, the current monitoring value for the aquatic biodiversity recovery index is 45, the current monitoring value for the sediment pollutant reduction rate is 48, and the current monitoring value for the total phosphorus reduction rate is 52. After calculating the matter-element extension correlation, the aquatic biodiversity recovery index has the highest correlation with the improvement level among its four correlation values, the sediment pollutant reduction rate also has the highest correlation with the improvement level, while the total phosphorus reduction rate has the highest correlation with the qualified level. Therefore, the aquatic biodiversity recovery index and the sediment pollutant reduction rate are marked as weak link indicators, while the total phosphorus reduction rate is not marked. It equals 2.

[0041] When the number of weak links indicators When the value is only 1, no sorting is needed; this single weak point indicator is the highest priority for improvement. When the correlation coefficient is greater than or equal to 2, the urgency of improving each weak link indicator may differ. Simply comparing a single correlation coefficient value is insufficient to fully reflect the comprehensive disadvantage of each indicator across multiple dimensions. Therefore, the projection pursuit method is introduced to reduce the dimensionality of the weak link indicators and rank them. The core idea of ​​the projection pursuit method is to find an optimal projection direction in the high-dimensional space, projecting the high-dimensional data into a one-dimensional space so that the projected values ​​can retain the structural difference information in the original data to the greatest extent, thus effectively distinguishing the differences between samples even after dimensionality reduction.

[0042] First, construct the input data matrix for projection pursuit analysis. Extract all rows containing the weak link indicators from the matter-element correlation matrix. Each row contains the correlation values ​​of that weak link indicator with the four evaluation levels, thus forming one [database / database]. The input data matrix has 4 rows and 4 columns. The matrix's first... Line number Column elements are denoted as ,in The value range is 1 to The corresponding sequence number of the indicators for each weak link The value ranges from 1 to 4, corresponding to 4 evaluation levels.

[0043] Because the dimensions and numerical ranges of the correlation values ​​in different columns may differ—for example, the correlation values ​​in one evaluation level column may be generally high while those in another column may be generally low—directly using the original correlation values ​​for projection can lead to columns with larger numerical ranges having an excessive influence on the projection results, masking the structural differences contained in columns with smaller numerical ranges. To eliminate this dimensional effect, range standardization needs to be performed on the input data matrix column by column. For the [specific data point]... First, find the maximum value in the column of data. and minimum value Then, each element in the column is standardized as follows: ;in The standardized value. The original correlation coefficient value before standardization. For the first The maximum value of the column. For the first The minimum value of each column. After range standardization, the value range of each column is normalized to between 0 and 1, eliminating the numerical differences between columns and ensuring that the contribution of each evaluation level dimension to the projection result is on an equal scale during the subsequent projection process. In an optional implementation, when the maximum value and minimum value of a column are equal, all standardized values ​​in that column are uniformly set to 0 to avoid the anomaly of a denominator of zero.

[0044] After standardization, the projection pursuit dimensionality reduction calculation begins. Let the projection direction vector be... ,in , , , The projection components correspond to the four evaluation level dimensions, and the projection direction vectors must satisfy the normalization constraint, i.e. The purpose of the normalization constraint is to ensure that the projected direction vector only represents directional information without introducing additional scale changes. The normalized input data matrix will then be used to... The weak link index is projected along the projection direction vector to obtain the projected value of the weak link index. : ;in , , , The first The correlation values ​​of the weak link indicators after standardization across four assessment levels. Projected values. It is a scalar that compresses the original four-dimensional correlation information into a single value, allowing the indicators of each weak link to be directly compared on the same scale.

[0045] Different projection direction vectors lead to different projection value distributions, and the key to projection tracing methods lies in finding an optimal projection direction vector that results in the best projection value distribution. Each projected value exhibits both local clustering and sufficient global dispersion. Local clustering ensures that weak links with similar levels of similarity are projected to nearby locations, while global dispersion ensures that weak links with significant differences are projected to more distant locations. This balance allows the projected values ​​to reveal the intrinsic differences between the various weak links to the greatest extent possible.

[0046] Therefore, a projection index function is constructed. The search target is the one with the optimal projection direction. Projection index function. Defined as the average local density of the projected values ​​of all weak link indicators. With the degree of dispersion The product of: ;in For average local density, The degree of dispersion. Average local density. The calculation process is as follows. First, determine the window radius. Window radius Set to All Projected values The window radius is 0.1 times the standard deviation. The size of the window directly affects the granularity of density statistics: if the window is too large, all projected values ​​will be included in each other's neighborhoods, and the density values ​​will tend to be uniform and lose their distinguishing ability; if the window is too small, most projected values ​​will not contain other projected values ​​in their neighborhoods, and the density values ​​will tend to be zero and also cannot be effectively distinguished. Choosing 0.1 times the standard deviation as the window radius is because the standard deviation reflects the overall dispersion level of the current projected value distribution, and taking 0.1 times it can obtain a stable density estimate while maintaining sufficient sensitivity. In an optional implementation, the window radius can also be set between 0.05 and 0.2 times the standard deviation, and the specific value can be adjusted according to the number of weak link indicators. When the magnification is large, the magnification can be appropriately reduced to improve the resolution. When the value is small, the factor can be increased appropriately to obtain a smoother density estimate.

[0047] Determine the window radius Then, for each projection value Statistics in all Among the projected values, with The absolute value of the difference does not exceed The number of other projected values ​​is denoted as Note that projected values ​​should be excluded during the statistical analysis. It only takes into account other projected values. All The number of other projected values ​​in the neighborhood of each projected value Summing and dividing by That is, to obtain the average local density : ;in For the first The projection value is within the window radius. The number of other projected values ​​within the range, This represents the total number of indicators indicating weak links. Average local density. The larger the value, the higher the degree of clustering between the projected values.

[0048] Dispersion This measures the dispersion of all projected values ​​in a one-dimensional projected space. First, calculate all... The mean of the projected values Then, square the difference between each projected value and the mean, sum them up, and finally take the square root of the sum: ;in For the first The projected value of each weak link indicator is the arithmetic mean of all projected values. Dispersion degree. The larger the value, the stronger the dispersion between the projected values.

[0049] Projection index function Average local density and degree of dispersion Multiplication allows for simultaneous consideration of local clustering and global dispersion. Focusing solely on maximizing local density might lead to all projected values ​​clustering in a narrow region, losing discriminative power; focusing solely on maximizing dispersion might result in projected values ​​being uniformly scattered without forming meaningful clusters. Through multiplication, the projection index function only becomes truly valid when both factors simultaneously reach large values. Only then can a larger value be obtained, thus ensuring that the found projection direction can simultaneously meet the dual requirements of clustering and discreteness.

[0050] The search enables the projection index function. Finding the optimal projection direction vector that maximizes the value is a global optimization problem on a 4-dimensional unit hypersphere, due to the projection index function. The dependence on the projection direction vector is highly nonlinear, and traditional gradient descent methods are prone to getting trapped in local optima. This invention employs a particle swarm optimization algorithm to perform this search process. By simulating swarm intelligence behavior, the particle swarm optimization algorithm can effectively approximate the global optimum in a multi-dimensional search space.

[0051] The execution process of the particle swarm optimization algorithm is as follows. Initially, 30 particles are selected. Each particle's position is represented by a 4-dimensional vector, indicating a candidate projection direction vector; each particle's velocity is represented by a 4-dimensional vector, indicating its movement trend in the search space. Initially, the position component of each particle randomly takes values ​​between -1 and 1, and the velocity component randomly takes values ​​between -0.5 and 0.5. In an optional implementation, the number of particles can be adjusted to an integer between 20 and 50 depending on the problem complexity.

[0052] At the start of each iteration, the position vector of each particle is normalized. Specifically, the square root of the sum of the squares of the four components of the particle's position vector is calculated, and then each of the four components is divided by this square root value, ensuring that the magnitude of the processed position vector equals 1, satisfying the normalization constraint of the projected direction vector. This normalization ensures that the particle always moves within the 4-dimensional unit hypersphere and does not deviate into the meaningless region outside the hypersphere.

[0053] After normalization, the position vector of each particle is used as the current candidate projection direction vector. Following the calculation process of the aforementioned projection value, average local density, and dispersion degree, the projection index function value corresponding to each particle is calculated respectively. Subsequently, the following update operation is performed on each particle: if the current projection index function value of the particle is greater than the maximum projection index function value it has ever reached in history, then the current position is recorded as the individual optimal position of the particle; among the individual optimal positions of all 30 particles, the position with the largest projection index function value is selected as the global optimal position.

[0054] The velocity and position of each particle are adjusted using the individual optimal position and the global optimal position. For each velocity component of each particle, the following steps are taken: multiply the current velocity component by the inertia weight coefficient, add the individual learning factor multiplied by a random number between 0 and 1, multiply by the difference between the individual optimal position component and the current position component, add the social learning factor multiplied by another random number between 0 and 1, and multiply by the difference between the global optimal position component and the current position component to obtain the updated velocity component. The inertia weight coefficient controls the degree to which the particle maintains its previous direction of motion, the individual learning factor controls the tendency of the particle to move closer to its historical optimal position, and the social learning factor controls the tendency of the particle to move closer to the global optimal position. In one specific implementation, the inertia weight coefficient is set to 0.7, and both the individual learning factor and the social learning factor are set to 1.5. After updating the velocity, the updated velocity component is added to the current position component of each particle to obtain the new position component.

[0055] The iteration process is repeated 100 times before termination, and the final global optimal position is taken as the optimal projection direction vector. In an optional implementation, the number of iterations can also be set between 50 and 200, or a convergence criterion can be used to terminate early: when the change in the global optimal projection index function value in 10 consecutive iterations is less than a preset convergence threshold (e.g., 0.0001), the iteration ends early.

[0056] After obtaining the optimal projection direction vector, the projection values ​​of each weak link indicator are recalculated using this vector. Each weak link indicator in the standardized input data matrix is ​​then projected along the optimal projection direction vector using the aforementioned projection value calculation method, yielding its projection value along that direction. The larger the projection value of a weak link indicator, the more prominent it is in the overall disadvantage structure revealed by the optimal projection direction, indicating a higher urgency for improvement. All weak link indicators are then sorted from largest to smallest projection value to generate an improvement priority sequence. Weak link indicators ranked high in this priority sequence should be addressed first in the subsequent development of environmental technology promotion plans.

[0057] Continuing with the previous example, in When the value equals 2, after standardization and projection, the aquatic biodiversity recovery index and sediment pollutant reduction rate are respectively assumed to have a projected value of 0.83 for the aquatic biodiversity recovery index and a projected value of 0.61 for the sediment pollutant reduction rate. The improvement priority sequence is then that the aquatic biodiversity recovery index is ranked first and the sediment pollutant reduction rate is ranked second.

[0058] It is worth noting that when When the number of weak links is large (e.g., more than 5 weak link indicators), the advantages of the projection pursuit method are particularly obvious. This is because it is difficult to obtain an intuitive and consistent ranking conclusion by directly comparing the multi-level correlation values ​​of each indicator in a high-dimensional space. However, the projection pursuit method effectively compresses multi-dimensional information into a 1-dimensional projection value, which greatly simplifies the ranking process while preserving key difference features.

[0059] After identifying the weak links and generating an improvement priority sequence, specific environmental technology promotion and service plans need to be developed for each weak link. In actual engineering projects, multiple competing optimization objectives exist—the desire is to cover as many weak links as possible, control the total cost of promotion and implementation, and maximize the matching degree between the selected technologies and the weak links. These objectives often cannot be simultaneously optimized; rather, they are contradictory, with one gaining at the expense of the other. For example, increasing technology coverage usually means investing more funds, while cost reduction may result in some weak links not being covered by technology or having to use alternative technologies with lower suitability. Faced with this multi-objective conflict, the NSGA-II multi-objective genetic algorithm can be used for optimization. Without pre-setting the priority relationships between objectives, it can simultaneously search for a set of Pareto optimal solutions that achieve different compromises among the objectives, providing decision-makers with diverse alternatives.

[0060] refer to Figure 3 The horizontal axis in the graph represents the degree of insufficient technology coverage. The vertical axis represents the cost of promotion and implementation. The unit is 10,000 yuan. In the bi-objective optimization space, candidate solutions in the population are divided into multiple non-dominated levels according to Pareto dominance. The first non-dominated level, namely the Pareto optimal frontier, is represented by a line connecting circular markers and contains 8 non-dominated candidate solutions. These solutions do not have any other solutions in the current population that can simultaneously achieve the same level. and The two target dimensions are superior to or equal to them. The second non-dominated level is represented by lines connecting square markers and contains 7 candidate solutions, which are dominated only by some solutions in the first level. The third non-dominated level is represented by lines connecting triangular markers and contains 7 candidate solutions, which are dominated by solutions in the first two levels. The solutions within each level show a clear trade-off trend along the frontier direction, namely, the degree of technology coverage inadequacy. The lower the cost of implementation and promotion of the solution. The higher the level, the lower the priority, reflecting the inherent conflict between the two optimization objectives. The diagram uses dashed arrows to indicate the dominance relationship between adjacent levels, with the arrows pointing from the dominated scheme to the dominant scheme, visually illustrating the hierarchical logic of non-dominated ranking. In the selection operation of the NSGA-II algorithm, schemes with smaller non-dominated level numbers have higher selection priority; that is, schemes in level 1 are prioritized over level 2, and schemes in level 2 are prioritized over level 3. When further differentiation of individual quality is needed within the same level, a secondary ranking is performed using the crowding distance index. Individuals with larger crowding distances are prioritized for retention to maintain a uniform distribution of the population on the Pareto front. The lower left corner of the diagram also marks the ideal point location with a star. The coordinates of the ideal point represent the optimal combination of the objective functions and serve as a reference benchmark in subsequent comprehensive inferiority score calculations.

[0061] First, a coding scheme for the decision variables is established. Based on the improvement priority sequence output by the aforementioned projection pursuit stage, there are a total of weak link indicators. For each weak link indicator, no more than five candidate environmental protection technologies are pre-selected by domain experts or environmental technology databases. Each candidate technology is accompanied by two attribute values: promotion cost value and technology adaptability value. The promotion cost value is in units of ten thousand yuan, reflecting the expected cost required to promote and deploy the candidate environmental protection technology to the target river, covering equipment procurement, installation and commissioning, personnel training, and initial operation and maintenance. The technology adaptability value is a value between 0 and 1, which is comprehensively evaluated by experts based on the degree of matching between the candidate environmental protection technology and the target river's climate conditions, hydrological characteristics, existing infrastructure, etc. The closer the value is to 1, the higher the degree of matching. In an optional implementation, the technology adaptability value can also be automatically calculated by constructing a technology adaptability scoring model. Input factors include the target river's runoff level, the proportion of seasonal flow interruption, the type of existing treatment facilities, and the level of regional economic development.

[0062] Each weak link indicator corresponds to one decision variable, denoted as . ,in The value range is 1 to This corresponds to the sequence number of each weak link indicator in the improvement priority sequence. Decision variables. The value can be an integer ranging from 0 to the number of candidate environmental protection technologies corresponding to the weak link indicator. When the value is 0, it means that the first... No technical promotion measures will be taken for any weak link indicator; when When the value is a positive integer, it indicates that the candidate environmental protection technology with the corresponding index is selected for promotion to address the weak link. For example, if the first weak link indicator (aquatic biodiversity restoration index) has 3 candidate environmental protection technologies, then... The value range is 0, 1, 2, and 3, representing no action, selection of candidate technology 1, selection of candidate technology 2, and selection of candidate technology 3, respectively. (This will be repeated for emphasis.) The decision variables are concatenated according to the priority order of the weak link indicators, forming a length of [length missing]. decision variable vector Each decision variable vector constitutes a complete candidate environmental protection technology promotion plan.

[0063] Next, we define three optimization objectives and unify them into a minimization direction so that a unified comparison rule can be used to determine the dominance relationship in the subsequent non-dominated sorting process.

[0064] The first optimization objective is to reduce the inadequacy of technology coverage. The number of decision variables with values ​​greater than 0 in the statistical decision variable vector, i.e., the number of weak link indicators for which actual technology promotion measures were implemented, is calculated by dividing this number by the total number of weak link indicators. The degree of inadequacy in technology coverage is obtained by subtracting the technology coverage rate from 1. ;in This represents the number of decision variables in the decision variable vector that take values ​​greater than 0. This represents the total number of indicators indicating weak links. The value range is between 0 and 1. A smaller value indicates a wider range of technology coverage. This applies when technology promotion measures are implemented for all weak link indicators. The value equals 0 when no weak links are identified and measures are implemented. It equals 1. Using insufficiency rather than coverage as the objective is to unify all objectives towards minimization, ensuring consistency in the dominance relationship judgment logic in non-dominated sorting.

[0065] The second optimization objective is to reduce implementation costs. The promotion cost of each candidate environmental protection technology corresponding to a decision variable with a value greater than 0 in the decision variable vector is accumulated one by one: ;in exist Time equals the first The first weak link indicator The promotion cost of the candidate environmental protection technologies is in When it equals 0. The unit is ten thousand yuan, and the smaller the value, the lower the total cost of the promotion plan.

[0066] The third optimization objective is to address the issue of insufficient technology compatibility. When there are decision variables in the decision variable vector that have values ​​greater than 0, first calculate the arithmetic mean of the technology fit values ​​of all decision variables with values ​​greater than 0 for the candidate environmental protection technologies, and then subtract this arithmetic mean from 1 to obtain the technology inadequacy degree. ; in Let be the set of indices of the decision variables in the decision variable vector that take values ​​greater than 0. For set The number of elements in For the first The first weak link indicator The technology fit value of each candidate environmental protection technology. When all decision variables are 0, it means that no technology has been selected, and the average technology fit value cannot be calculated in this case. Setting it directly to 1 indicates that the inadequacy of adaptation has reached its maximum. The smaller the value, the higher the degree of matching between the selected technology and the corresponding weak link.

[0067] During the optimization process, constraints also need to be set to exclude solutions that do not meet the limitations of actual engineering projects. The first constraint is the cost of promotion and implementation. The budget limit shall not exceed the preset upper limit. The upper limit is determined by the project management team based on the available funds for the current period, for example, set at 5 million yuan. Anything that makes... Candidate solutions exceeding the budget ceiling are considered infeasible. The second constraint requires that the decision variables corresponding to the top-priority weak links in the improvement priority sequence must have values ​​greater than 0; that is, technical promotion measures must be arranged for the most pressing weak link indicators. The number of top-priority indicators is the total number of weak link indicators. The smaller value in 3. This constraint is set because the order of the improvement priority sequence has been determined by the projection pursuit method based on the degree of comprehensive disadvantage. The weak link indicators at the forefront are the links with the highest urgency of improvement. If these links are allowed to be left unaddressed during the optimization process, the optimization results may neglect the most critical weak links due to an excessive focus on cost minimization, causing the promotion plan to lose its practical significance.

[0068] After the constraints and objective function are determined, the iterative solution process of the NSGA-II multi-objective genetic algorithm begins. The population size is set to 100, meaning each generation contains 100 individuals, and each individual corresponds to a sequence number of... The decision variable vector. In one alternative implementation, the population size can be adjusted according to the number of weak link indicators and the number of candidate technologies, generally taking an integer between 50 and 200.

[0069] During population initialization, each individual's decision variables are assigned values ​​using random integers, with each variable randomly selected from 0 to the corresponding number of candidate environmental technologies. After random initialization, each individual undergoes constraint checks and remediation to ensure all individuals in the initial population meet the constraints. The remediation process is as follows: First, check the second constraint. If the decision variable corresponding to a top-ranking weakness indicator in the improvement priority sequence has a value of 0, then reassign that decision variable a random integer value between 1 and the corresponding number of candidate environmental technologies. After remediating the second constraint, check the first constraint. If the implementation cost... If the budget limit is exceeded, one decision variable will be randomly selected from the non-mandatory retention decision variables (i.e., decision variables that are not among the top priority items in the improvement sequence) and set to 0, and the implementation cost will be recalculated. If the budget limit is still exceeded, the next non-mandatory retention decision variable is selected and set to 0. This process is repeated until the implementation cost meets the budget constraint or all non-mandatory retention decision variables have been set to 0.

[0070] After population initialization and repair are completed, the iterative process begins. Each generation iteration includes steps such as non-dominated sorting, crowding distance calculation, selection operation, crossover operation, mutation operation, and environment selection.

[0071] The purpose of non-dominated sorting is to divide individuals in a population into several levels according to their dominance relationships. Since all three optimization objectives are minimization directions, the dominance relationship between individuals is defined as follows: For any two individuals, if individual A's three optimization objective values ​​are all less than or equal to individual B's corresponding objective value in every objective, and are strictly less than individual B in at least one objective, then individual A is said to dominate individual B. The specific sorting process is as follows: Iterate through each individual in the population, comparing its three objective values ​​with those of all other individuals in the population, recording the number of times each individual is dominated by other individuals and the list of other individuals dominated by that individual. All individuals with a dominance count of 0 are assigned to the first non-dominated level. These individuals are not surpassed by any other individual simultaneously in all objectives in the current population, representing the optimal compromise set in the current population. Then, logically remove all individuals from the first non-dominated level from the population, recount the dominance counts for the remaining individuals, and assign individuals with a dominance count of 0 to the second non-dominated level. This process continues until all individuals in the population are assigned to a certain non-dominated level. The smaller the level number, the better the overall optimization performance of the corresponding individual.

[0072] For individuals within the same non-dominated hierarchy, crowding distance needs to be calculated to measure the sparsity of their distribution in the target space. Individuals with larger crowding distances have fewer neighboring individuals in the target space, meaning they are located in the marginal or sparse regions of the solution set. Prioritizing the retention of individuals with larger crowding distances in environment selection maintains the diversity of the solution set in the target space, preventing a large number of individuals from clustering in a localized area of ​​the Pareto front while ignoring effective compromise solutions in other areas.

[0073] The crowding distance is calculated as follows: For each of the three optimization objectives, all individuals within the current non-dominated level are sorted in ascending order of the objective value. After sorting, the crowding distance between the first and last individuals on that objective is set to infinity, ensuring that individuals at the extremes of the objective's value range are always prioritized. For an individual in the middle position after sorting, its crowding distance on that objective is equal to the difference between the objective values ​​of its two adjacent individuals divided by the range of that objective in the entire population. The range is the difference between the maximum and minimum values ​​of that objective among all individuals in the population. Dividing by the range aims to normalize the crowding distances of different objectives to the same dimension, ensuring that the subsequent summation of the crowding distances for each objective does not introduce a dominant bias due to the large value range of a particular objective. The total crowding distance for each individual is obtained by summing the crowding distances calculated for each of the three optimization objectives.

[0074] The selection operation uses a binary tournament approach to generate the parent pairing pool. Each time, two individuals are randomly selected from the current population for comparison, prioritizing the individual with the smaller non-dominated level number for the pairing pool; if two individuals belong to the same non-dominated level, the individual with the larger total crowding distance is selected. This selection and comparison process is repeated until the number of individuals in the pairing pool reaches the population size of 100.

[0075] The crossover operation involves sequentially selecting two parent individuals from the pairing pool and performing a single-point crossover with a crossover probability of 0.9: (The last part, "1 to...", appears to be an unrelated fragment and is omitted from the translation.) A random crossover point is selected between the two parent individuals. The decision variable segments of the two parent individuals after the crossover point are swapped, generating two offspring individuals. If the randomly generated probability value is greater than 0.9, the two parent individuals are directly copied as offspring without crossover. The crossover probability is set to 0.9 because a higher crossover probability promotes information exchange between different individuals, accelerating the exploration efficiency of the search space. In an optional implementation, the crossover probability can be adjusted to a value between 0.8 and 0.95.

[0076] The mutation operation re-randomizes each decision variable of each offspring individual produced by crossover with a mutation probability of 0.1. Specifically, for each decision variable, a random number between 0 and 1 is generated. If this random number is less than 0.1, the decision variable is reassigned a random integer between 0 and the corresponding number of candidate environmental technologies. The mutation operation introduces random perturbation to prevent the population from prematurely converging to a local optimum and losing its ability to explore other regions of the search space. The mutation probability is set to 0.1, which is relatively low compared to the crossover probability, ensuring that crossover is the dominant mechanism in population evolution, with mutation serving only as an auxiliary diversity supplement. In an optional implementation, the mutation probability can be adjusted to a value between 0.05 and 0.2.

[0077] After crossover and mutation are completed, the aforementioned constraint checks and repairs are performed on all offspring individuals to ensure that all individuals in the offspring population meet the constraints.

[0078] During the environmental selection phase, the parent and offspring populations are merged to form a mixed population of 200. The aforementioned non-dominated sorting and crowding distance calculations are then performed on the mixed population. Subsequently, starting from the first non-dominated level, all individuals from each level are sequentially selected for the next generation population. If the next generation population size still does not reach 100 after all individuals from a certain non-dominated level have been added, all individuals from the next level are added. When the number of individuals in a certain non-dominated level exceeds the remaining capacity of the next generation population, the individuals within that level are sorted from largest to smallest based on their total crowding distance, and individuals with larger total crowding distances are sequentially selected to fill the next generation population until the population size reaches 100.

[0079] The algorithm checks if the current iteration count has reached the preset maximum iteration count of 200. If not, it returns to the non-dominated sorting step with the current next generation as the new parent population to continue the next iteration. If the maximum iteration count has been reached, the iteration terminates. In an optional implementation, the maximum iteration count can be set to an integer between 100 and 500. Alternatively, a convergence criterion can be added as an auxiliary termination condition, such as terminating early when the composition of individuals in the first non-dominated level has not changed in 20 consecutive iterations.

[0080] After the iteration terminates, all individuals in the first non-dominated level of the final population constitute the Pareto optimal solution set. Each individual in the Pareto optimal solution set represents a non-dominated governance scheme that achieves different compromises among the three optimization objectives. There is no dominance relationship between these schemes, that is, there is no situation where one scheme is superior to another scheme in all objectives.

[0081] The process of selecting the final solution from the Pareto optimal solution set is as follows: For each individual in the Pareto optimal solution set, range standardization is performed on the three optimization objective values. Taking the first optimization objective, technology coverage deficiency, as an example, the maximum value of technology coverage deficiency in the Pareto optimal solution set is found. and minimum value The degree of technological undercapacity for each individual will be standardized in the following way: ;in The lack of coverage of standardized technologies Due to insufficient technology coverage before standardization, and These represent the maximum and minimum values ​​of the technical coverage inadequacy in the Pareto optimal solution set. When the maximum and minimum values ​​are equal, the standardized value for all individuals on that objective is uniformly set to 0. (This relates to the cost of implementation and promotion.) Inadequate technology compatibility The same range standardization operation described above is performed. The purpose of standardization is to eliminate differences in units and numerical ranges between different objectives, so that the three objectives are on the same comparison benchmark in the subsequent comprehensive scoring.

[0082] refer to Figure 4 The horizontal axis in the graph represents the degree of insufficient technology coverage. The vertical axis represents the cost of promotion and implementation. The unit is 10,000 yuan, and it contains 12 Pareto optimal solutions, labeled as solution 1 to solution 12. Each solution is plotted as a scatter plot in a two-dimensional objective space, and the fill depth of the scatter plots maps to the third optimization objective, namely, the degree of technology mismatch. The value of , The value ranges from 0 to 0.40, with deeper fill indicating greater underfitting. The 12 solutions exhibit typical Pareto front distribution characteristics, extending from the upper left to the lower right along the front trend line. Solutions located in the upper left region of the front have lower underfitting. However, this corresponds to higher promotion and implementation costs. For example, solution 1 , Ten thousand yuan; solutions located in the lower right region of the frontier have lower costs but significant limitations in technical coverage, such as solution 12. , The distribution pattern reflects the inevitable trade-off between the adequacy of technology coverage and the cost of promotion. The figure shows the overall trend of the Pareto front with dashed lines, connecting the scattered solutions to demonstrate the convex structure of the front. Under the joint consideration of the three objectives, although some solutions... and Performed well on both objectives but The values ​​are too large, while other solutions are in... It performs better in the above objectives but compromises on other objectives. No single solution in the Pareto optimal set can simultaneously achieve the desired result. , and The solution is superior to all other solutions in all three objectives, therefore a comprehensive inferiority score method is needed to ultimately filter the solution set. This is done by calculating the comprehensive inferiority score of each solution in the normalized objective space. To determine the final recommended solution.

[0083] The standardized values ​​for each individual across the three objectives are summed to obtain the overall poorness score: ;in To assess the overall poorness, , , These are the standardized values ​​for the three optimization objectives. The overall poorness score is the weighted sum of the standardized values ​​of the three objectives; a smaller score indicates a more balanced and better overall performance across the three objectives for that individual. The decision variable vector corresponding to the individual with the smallest overall poorness score is selected as the final environmental technology promotion service plan.

[0084] After obtaining the final environmental technology promotion service plan, the decision variable vector is decoded and output. Each decision variable in the vector is iterated over; if its value is greater than 0, the name of the corresponding weak link indicator, the name of the selected candidate environmental technology, the promotion cost, and the technology suitability value are output; if the value is 0, it is indicated that no technology promotion measures will be arranged for that weak link indicator. The decoded output forms a complete list of environmental technology promotion service plans, specifying which environmental technologies are recommended for which weak links, the expected investment, and the degree of technology suitability with the target watershed. This list can be directly used to guide subsequent technology promotion implementation.

[0085] In one alternative implementation, when multiple individuals with equal overall inferiority scores exist in the Pareto optimal solution set, the one with the highest implementation cost is prioritized. Lower-performing individuals are considered as the final solution to further control economic input while maintaining comparable overall performance. Alternatively, all non-dominated solutions in the Pareto optimal solution set can be output together for decision-makers to consider, allowing them to choose the final implementation plan based on current budget priorities, policy orientation, or watershed management objectives.

[0086] The present invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make various improvements and modifications to the present invention without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A system for evaluating the effectiveness of seasonal river pollution control projects and promoting environmental protection technologies, characterized in that: include: The matter-element extension evaluation module is used to collect current monitoring values ​​of multiple evaluation indicators for seasonal river pollution control projects. It constructs matter-element objects to be evaluated using each evaluation indicator as a feature and the current monitoring value as a feature value. Based on the classical domain and section domain corresponding to multiple preset evaluation levels, it calculates the correlation degree of each evaluation indicator with respect to each evaluation level according to the matter-element extension correlation function, forming a matter-element correlation degree matrix. Based on the matter-element correlation degree matrix, it determines the comprehensive evaluation level of the pollution control project. The projection tracking weak link identification module is used to select the evaluation index with the maximum correlation value of the level to be improved from the matter-element correlation matrix as the weak link index. The module constructs an input data matrix with the correlation values ​​of all weak link indices to each evaluation level and performs standardization processing. The module performs projection tracking dimensionality reduction on the standardized input data matrix and searches for the optimal projection direction vector that maximizes the function value of the projection index through an optimization algorithm. The module generates an improvement priority sequence based on the projection values ​​of each weak link index along the optimal projection direction vector from largest to smallest. The multi-objective optimization and promotion module is used to set candidate environmental protection technologies for each weak link indicator according to the improvement priority sequence. The technology selection corresponding to each weak link indicator is used as the decision variable, and the three optimization objectives are insufficient technology coverage, promotion and implementation cost and insufficient technology adaptation. Under the preset constraints, the NSGA-II multi-objective genetic algorithm is used to iteratively optimize through non-dominated sorting and crowding distance selection, outputting the Pareto optimal solution set, and selecting the final environmental protection technology promotion service plan from the Pareto optimal solution set.

2. The system according to claim 1, characterized in that, The evaluation indicators include nine benefit-oriented indicators: chemical oxygen demand reduction rate, ammonia nitrogen reduction rate, total phosphorus reduction rate, aquatic biodiversity recovery index, riparian vegetation coverage rate, sediment pollutant reduction rate, facility integrity rate, treatment capacity compliance rate, and operation and maintenance response timeliness rate.

3. The system according to claim 1, characterized in that, The Matter-Element Expansion Evaluation Module divides the effectiveness of pollution control engineering into four evaluation levels: excellent, good, qualified, and needing improvement. For each evaluation level, the characteristic value range of each evaluation indicator is pre-set as the classical domain, and the complete value range after merging the classical domains of the four evaluation levels is used as the section domain.

4. The system according to claim 3, characterized in that, When calculating the correlation between each evaluation indicator and each evaluation level, the Matter-Element Extension Evaluation Module first calculates the midpoint value and half-width of the classical domain interval, and then determines the correlation according to the following three cases: If the current monitoring value is within the classical domain interval, the correlation is equal to 1 minus the absolute value of the difference between the current monitoring value and the midpoint value of the classical domain interval divided by the half-width of the classical domain interval; If the current monitoring value is outside the classical domain interval but within the section domain, the distance from the current monitoring value to the boundary of the classical domain interval is first calculated, and then the distance is divided by the difference between the section domain interval width and the classical domain interval width, and the negative value is taken as the correlation. If the difference between the section domain interval width and the classical domain interval width is less than the preset minimum difference threshold, the minimum difference threshold is used instead; If the current monitoring value exceeds the section domain range, the correlation is set to -1.

5. The system according to claim 1, characterized in that, The Matter-Element Extension Evaluation Module arranges the correlation degree of each evaluation indicator to each evaluation level in the order of indicator and level to form a Matter-Element Correlation Degree Matrix. The Matrix is ​​summed column by column to obtain the comprehensive correlation degree value of each evaluation level. The evaluation level with the largest comprehensive correlation degree value is selected as the comprehensive evaluation level.

6. The system according to claim 1, characterized in that, The projection tracking weak link identification module compares the correlation values ​​of each evaluation index in the object-element correlation matrix with each evaluation level. If the correlation value of a certain evaluation index with the level to be improved is the maximum value among its correlation values ​​with each evaluation level, then the evaluation index is marked as a weak link index. An input data matrix is ​​constructed using the correlation values ​​of all weak link indices with each evaluation level. The input data matrix is ​​then subjected to range standardization processing by column. Specifically, for each column of data, each element is subtracted from the minimum value of that column and then divided by the difference between the maximum value and the minimum value of that column.

7. The system according to claim 6, characterized in that, When the projection tracing weak link identification module performs projection tracing dimensionality reduction, it projects each row of the standardized input data matrix along the candidate projection direction vector. The projection value of each weak link index is equal to the sum of the standardized correlation values ​​of the corresponding row multiplied by each component of the candidate projection direction vector. The projection index function value is equal to the product of the average local density and the dispersion of all weak link index projection values. The average local density is calculated by using 0.1 times the standard deviation of all projection values ​​as the window radius, counting the number of other projection values ​​within the window radius for each projection value, summing all the numbers, and then dividing by the total number of weak link indices. The dispersion is the square root of the sum of the squares of the differences between each projection value and the mean of the projection values.

8. The system according to claim 7, characterized in that, The projection tracking weak link identification module uses a particle swarm optimization algorithm to search for the optimal projection direction vector. In each iteration, the position vector of each particle is normalized to make the modulus equal to 1, and then the corresponding projection index function value is calculated. The particle velocity and position are updated based on the individual optimal position and the global optimal position. After the iteration is completed, the global optimal position is used as the optimal projection direction vector. The projection values ​​of each weak link index are recalculated along the optimal projection direction vector and then sorted in descending order of projection values ​​to generate an improvement priority sequence.

9. The system according to claim 1, characterized in that, The multi-objective optimization and promotion module sets no more than 5 candidate environmental protection technologies and their corresponding promotion cost and technology suitability values ​​for each weak link indicator. Each weak link indicator corresponds to one decision variable, with a value ranging from 0 to the number of corresponding candidate environmental protection technologies. 0 indicates that no technology promotion measures will be taken for the corresponding weak link. The three optimization objectives are unified in minimizing the following direction: the technology coverage insufficiency is equal to 1 minus the ratio of the number of decision variables with values ​​greater than 0 to the total number of all weak link indicators; the promotion implementation cost is equal to the sum of the promotion costs of the candidate environmental protection technologies corresponding to all decision variables with values ​​greater than 0; and the technology suitability insufficiency is equal to 1 minus the arithmetic mean of the technology suitability values ​​corresponding to all decision variables with values ​​greater than 0 when there are decision variables with values ​​greater than 0, and is set to 1 when all decision variables have values ​​of 0.

10. The system according to claim 9, characterized in that, The constraints set by the multi-objective optimization promotion module include that the promotion implementation cost does not exceed the preset budget limit and that the decision variables corresponding to the top-ranked weak link indicators in the improvement priority sequence are greater than 0, and the number of top-ranked indicators is the smaller value between the total number of weak link indicators and 3. During the NSGA-II iteration, the parent and offspring populations are merged and then non-dominated sorting is performed. The next generation population is filled in order of increasing non-dominated level. Within the same non-dominated level, individuals are selected in order of decreasing crowding distance. After the iteration is completed, for each individual in the Pareto optimal solution set, the three optimization objective values ​​are standardized by range and then added together to obtain a comprehensive inferiority score. The decision variable corresponding to the individual with the smallest comprehensive inferiority score is decoded and output as the final environmental protection technology promotion service plan.

Images