Water environment health multi-index system evaluation method for winter rivers in cold regions
By constructing a multi-index evaluation system for winter rivers in cold areas, combining hierarchical analysis method, LightGBM model and game theory method, the problem that the existing evaluation methods fail to accurately reflect the water environment health status of rivers in cold areas is solved, and a higher evaluation is achieved Accuracy and scientificity provide a reliable evaluation basis for river management.
Patent Information
- Application Number
- CN202411251527.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-09-08
- Filing Date
- 2024-09-08
- Publication Date
- 2024-11-22
AI Technical Summary
The existing river water environment health evaluation methods fail to effectively consider the ecological fragility characteristics of rivers in winter in cold areas, resulting in low accuracy of evaluation results, which cannot accurately reflect the local water environment health status, and affect scientific decision-making management.
Construct a multi-index system evaluation method for winter rivers in cold areas, including the target layer, the criterion layer and the index layer. The hierarchical analysis method, the LightGBM model and the game theory method are used to calculate the combined weights based on subjective and objective factors, and the improved material elements are improved. The proliferation model calculates the correlation degree of river health and determines the level of water environment health.
It improves the accuracy of the evaluation algorithm, can more accurately reflect the health status of the river water environment in the cold area, provides scientific and comprehensive evaluation basis, and provides more reliable support for river management decisions.
Smart Images

Figure CN119003974A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of water environment health management, and in particular relates to a multi-index system evaluation method for water environment health of rivers in cold regions in winter. Background Art
[0002] The river system is a complex system composed of water, sand, riverbed and ecosystems that are interconnected in the river. It covers many aspects such as natural ecology, cultural society, and engineering regulation, and plays a vital role in social and economic development. At present, the problems of water environment deterioration and riverbank erosion are becoming increasingly prominent. How to scientifically manage rivers has become a focus of widespread concern at home and abroad. As an important basis for scientific river management, water environment health assessment is the key to the entire management system in terms of how to accurately assess the health of the river water environment.
[0003] The cold regions in the north are affected by seasonal climate, and there is a common phenomenon of uneven temporal and spatial distribution of rainfall and runoff. In addition to the flood season, dry season and normal water season, the rivers in the region also experience a 4-6 month ice-covered period each year. In addition, due to the low temperature and low flow, the species richness of river organisms in the cold regions of the north is much lower than that in other regions, the ecological structure is not complete, and the self-purification capacity is weak. Especially in winter, the water quality, water resources, water ecology and other aspects are significantly different from other regions, with strong temporal and spatial heterogeneity. Evaluating the health of the water environment of rivers in extremely fragile and sensitive ecosystems will not only help the scientific management decision-making of rivers in the region, but also have certain guiding significance for the scientific management and precise decision-making of general rivers.
[0004] In recent years, with the development of applied mathematics and new generation information technology, many comprehensive methods have been applied to the evaluation process in various fields, such as fuzzy comprehensive evaluation method, matter-element extension method, principal component analysis method, grey correlation method, ELECTRE, etc. These methods have their own characteristics and are suitable for different scenarios. At the same time, the weight of each indicator is also a key factor in determining whether the evaluation model is accurate. Nowadays, subjective weighting method or objective weighting method is usually used to determine the weight. Subjective weighting method has strong subjective initiative and relies on the personal knowledge, experience and preference of experts. It usually ignores the information contained in the data itself. Commonly used methods include: hierarchical analysis method, Delphi method, etc. Objective weighting method weakens subjective consciousness and pays more attention to mining the information contained in the data itself, but sometimes unreasonable weight coefficients may occur. Commonly used methods include entropy weight method, CRITIC method, coefficient of variation method, etc. In order to obtain more scientific and reasonable weight values, it is an effective way to solve the problem by coupling subjective weighting method with objective weighting method and carrying out river water environment health status evaluation based on combined weight values. However, at present, most scholars still use the subjective method of determining preference coefficients in the process of combination weighting, which weakens the objectivity and integration effect of combination weighting. The weight allocation of subjective and objective assignment methods still needs scientific and reasonable algorithm support.
[0005] In summary, the research on river water environment health evaluation at home and abroad has developed rapidly in recent years, and the relevant evaluation systems and methods have gradually matured. However, the existing evaluation index system does not take into account the characteristics of low temperature and low flow of rivers in cold regions in winter, and the fragile ecology, which does not match the fragile ecological environment in the northern cold regions. The evaluation results obtained based on this cannot accurately reflect the real situation of local water environment health, and the evaluation algorithm has low accuracy, which is not conducive to the implementation of scientific decision-making and management. Summary of the invention
[0006] The purpose of the present invention is to provide a multi-index system evaluation method for water environment health of rivers in cold regions in winter. By constructing a multi-index evaluation system that conforms to the characteristics of rivers in cold regions in winter, the problem of low accuracy of evaluation results caused by the traditional evaluation index system not taking into account the actual situation of ecological fragility in cold northern regions is overcome. Based on this multi-index evaluation system, combined with subjective and objective factors and based on a variety of methods, the accuracy of the evaluation algorithm is improved, providing a scientific and comprehensive evaluation basis for river management.
[0007] To achieve the above object, the present invention provides a multi-index system evaluation method for water environment health of rivers in cold regions in winter, comprising the following steps:
[0008] Step 1: Construct a cold region winter river water environment health evaluation index system including target layer, criterion layer and indicator layer;
[0009] Step 2: Collect data on various indicators of the river, divide the health status of the river water environment into five levels, namely very healthy, healthy, sub-healthy, unhealthy, and pathological, and divide each evaluation indicator into the same level level accordingly;
[0010] Step 3: Use the analytic hierarchy process and LightGBM model to calculate the weight of each indicator data from both subjective and objective perspectives, and based on the game theory method, use Nash equilibrium as the coordination goal to find the consistency and coordination between subjective and objective weights to obtain the combined weight;
[0011] Step 4: Use the improved matter-element extension model combined with the combined weights obtained in step 3 to calculate the correlation of river health, and determine the river water environment health evaluation level based on the maximum value of the correlation.
[0012] Preferably, in step 1, the target layer is the health status of river water environment in cold regions in winter; the criterion layer includes water resources, water environment, water ecology, social services, and habitat conditions; water resources indicators include water resource development and utilization rate, ecological flow satisfaction degree, and river longitudinal connectivity index; water environment indicators include water quality and water function area compliance rate; water ecology indicators include Chao index, Shannon index, large benthic invertebrate biological integrity index, and fish retention index; social service indicators include public satisfaction; and habitat conditions indicators include riparian vegetation coverage and riparian artificial disturbance degree.
[0013] Preferably, Chao Index and Shannon Index are used as the specific indicators of cold region winter rivers in the indicator layer; due to the ice-covered period in cold region winter rivers, the lives of many aquatic organisms are restricted, and water ecology only considers the growth indicators of invertebrates and fish, which will lead to a large deviation between the evaluation results and the actual situation; therefore, the selection of microbial indicators such as Chao Index and Shannon Index that occupy a wider ecological niche is more in line with the actual situation of water environment health evaluation of cold region winter rivers.
[0014] Preferably, when calculating the water resource development and utilization rate and the ecological flow satisfaction degree, since there is a freezing period in the cold winter rivers, the characteristics of the rivers in the cold winter are taken into consideration, and the calculation expression is as follows:
[0015] The calculation expression of water resources development and utilization rate is as follows:
[0016]
[0017] Where WURI is the water resources development and utilization rate, in %; WS is the water supply of the river basin, in 10,000 m 3 WR is the water resources in the river basin, including the ice storage during the ice-covered period, in units of 10,000 m 3 ;
[0018] The calculation expression of ecological flow satisfaction is as follows:
[0019]
[0020] In the formula, q d is the minimum daily average flow, in m 3 / s; Q is the multi-year average flow rate in the corresponding period, in m 3 / s; EF is the percentage of the minimum average daily flow from October to March of the following year (the period of river ice cover in cold regions in winter) to the multi-year average flow in the corresponding period.
[0021] Preferably, the LightGBM model is specifically: the gradient boosting decision tree is optimized by constructing a decision tree based on histogram, a leaf growing decision tree algorithm, histogram difference calculation, unilateral sampling of gradients, mutually exclusive feature bundling and sparse feature optimization to obtain the LightGBM model. The specific contents are as follows:
[0022] The gradient boosting decision tree is to gradually build a series of decision trees so that each tree reduces the residual of all previous trees. The calculation expression is as follows:
[0023]
[0024] In the formula, F m (x) is the model of the mth round; F m-1 (x) is the model of the m-1th round; h m (x) is the weak learner trained in the mth round; γ m is the learning rate, used to reduce the contribution of the new model;
[0025] The histogram-based decision tree construction optimizes the gradient boosting decision tree. Specifically, the eigenvalues are grouped to reduce the computational complexity. The calculation expression is as follows:
[0026]
[0027] In the formula, x i is the value of a sample on the i-th feature; n bin is the number of groups, min(x) represents the minimum value of feature x, and max(x) represents the maximum value of feature x;
[0028] The decision tree algorithm based on leaf growth optimizes the gradient boosting decision tree so that the LightGBM model adopts a leaf-growing strategy. In each iteration, the leaf with the largest gain among all current leaf nodes is selected for splitting. The calculation expression of the gain is as follows:
[0029]
[0030] In the formula, G L and G R is the sum of the gradients of the left child node and the right child node; H L and H R is the sum of the second-order derivatives of the left child node and the right child node; λ is the regularization parameter; γ is the penalty term for the number of split leaf nodes;
[0031] The gradient boosting decision tree is optimized based on histogram difference calculation to reduce the calculation amount of the LightGBM model. The calculation expression is as follows:
[0032] H left =H parent -H right (6)
[0033] In the formula, H right is the histogram of the right child node, H parent is the histogram of the parent node, H left is the histogram of the left child node;
[0034] The gradient boosting decision tree is optimized based on gradient one-side sampling, specifically the samples of the LightGBM model are optimized.
[0035] Preferably, the process of obtaining the combined weight in step 3 is as follows:
[0036] S31. Standardized processing index data, specifically divided into positive indicators and negative indicators, the standardized value r ij The calculation formula is as follows:
[0037] Positive indicators:
[0038]
[0039] Negative indicators:
[0040]
[0041] In the formula, x ij is the value of the jth evaluation index in the i-th evaluation unit; min(x j ) represents the minimum value of the jth index in the original data, max(x j ) represents the maximum value of the jth index in the original data;
[0042] S32. Obtain the subjective weights of various indicators for evaluating river water environment health in cold regions in winter based on the analytic hierarchy process;
[0043] S33, based on the LightGBM model, the objective weights of various indicators for evaluating the health of river water environment in cold regions in winter were obtained;
[0044] S34. Use game theory methods and Nash equilibrium as the coordination goal to find the consistency and coordination between subjective and objective weights and obtain the combined weights.
[0045] Preferably, the process of obtaining the objective weights of various indicators for evaluating the health of river water environment in cold regions in winter based on the LightGBM model in S33 is as follows:
[0046] S331. First, sort the training instances in descending order according to the absolute value of the gradient. Second, retain the head a×100% data instances with larger gradients to obtain a subset of instance a. For the remaining set A consisting of (1-a)×100% data instances with smaller gradients, c , randomly select a size of b×|A c |Data subset B; Finally, according to the estimated variance gain The subset A∪B is divided and the calculation expression is as follows:
[0047]
[0048] In the formula, A l ={x i ∈A:x ij ≤d},A r ={x i ∈A:x ij >d}, B l ={x i ∈B:x ij ≤d},B r ={x i ∈B:x ij >d}; coefficient Used to normalize the gradient on B to A c Size; g i Represents the gradient of the data.
[0049] S332. Combine the LightGBM model with the selected indicators to rank the importance of regional water quality characteristics. The more times an indicator feature is split in the decision tree, the higher its relative importance. The feature importance score is calculated based on the number of times each indicator feature splits the training data in all trees. The importance of the indicator feature obtained by the LightGBM model is recorded as F[i]. The indicator feature importance F[i] is sorted by numerical size. The expression of scale ranking is RANK(X). The order of each indicator after scale ranking is RANK(F[i]). The ranking result reflects the importance of each indicator.
[0050] S333. Determine the objective weight of the LightGBM model based on the order centroid. The ROC method maintains the sorting order, minimizes the maximum error of the weight, and reduces the risk of overlap due to inappropriate weights. The calculation formula of the order centroid is as follows:
[0051]
[0052] In the formula, N is the number of indicators, RANK(F[i]) is the ranking result of feature importance, and W o is the objective weight.
[0053] Preferably, in S34, a game theory method is used to find the consistency and coordination between the subjective and objective weights with Nash equilibrium as the coordination goal, and the process of obtaining the combined weights is as follows:
[0054] S341. Construct any linear combination of vectors ω:
[0055]
[0056] Where L is the number of methods for determining weight coefficients; α k is the linear combination weight coefficient α k ={α1, α2, ..., α L};ω k is the basic weight vector set; Table ω k The transpose of
[0057] S342, with ω and ω k The goal is to minimize the deviation of, construct the objective function, and obtain the combined weight coefficient α in ω k , the calculation formula is:
[0058]
[0059] S343, the combined weight coefficient α obtained by the above calculation k Normalize to get α k * , combined with formula (11), we get the final optimized combined weight value ω * :
[0060]
[0061] Preferably, in step 4, the process of calculating the correlation degree of river health by using the improved matter-element extension model combined with the combined weight obtained in step 3 is as follows:
[0062] S41: Construct a classical domain and normalize it to obtain a new classical domain R k :
[0063]
[0064] Where N j is the jth evaluation level; C i is the i-th evaluation index of level j; (a ji ,b ji ) is the value range of the i-th evaluation index under level j; q i is the maximum value of each indicator parameter range; (a ji ',b ji ') is the value range of the evaluation index after normalization;
[0065] S42: Constructing the section domain R p R x , and normalize the object-element to be evaluated to obtain the new object-element R L :
[0066]
[0067] Where N p represents the total number of evaluation categories, (a pi ,b pi ) means C i The range of the section domain;
[0068]
[0069] In the formula, v i is the value of the i-th evaluation index in the object-element to be evaluated;
[0070] S43: Calculate the new matter-element R L About New Classical Domain R K The distance D of the measurement range ji , the calculation formula is:
[0071] D ji =|v i ′-(a ji ′+b ji ′) / 2|-(b ji ′-a ji ′) / 2 (17)
[0072] S44: Calculate the comprehensive correlation K j (N), if K j =max{K j (N)}, the river water environment health evaluation level is j, and the river water environment health status in cold regions in winter is obtained, K j The calculation formula for (N) is:
[0073]
[0074] Therefore, the present invention adopts the above-mentioned multi-index system evaluation method for water environment health of rivers in cold regions in winter, which has the following beneficial effects:
[0075] (1) Comprehensively consider the characteristics of rivers in cold regions in winter, and select characteristic indicators of rivers in cold regions in winter, such as Chao index and Shannon index, based on the original traditional river health evaluation index system, to better adapt to the unique local environment and obtain more accurate evaluation results;
[0076] (2) The LightGBM model is introduced into the river water environment health assessment, which can process a large amount of data and high-dimensional features, thereby effectively and accurately selecting key indicators from a given data set. Moreover, LightGBM is optimized based on the traditional GBDT (gradient boosting decision tree), which effectively reduces the training set size for training the next ensemble tree, making the training of new trees faster;
[0077] (3) Combining subjective and objective weights through game theory can reduce the arbitrariness of weight combination compared with traditional methods;
[0078] (4) The evaluation is carried out based on the improved physical-element extension model, which makes up for the problem that the correlation function cannot obtain the calculation result when the evaluation index parameters exceed the node domain range, and has stronger practicality and logic.
[0079] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0080] Figure 1 It is a flow chart of a multi-index system evaluation method for water environment health of rivers in cold regions in winter according to the present invention;
[0081] Figure 2 The present invention is an evaluation index system diagram of a multi-index system evaluation method for water environment health of rivers in cold regions in winter. DETAILED DESCRIPTION
[0082] The following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0083] See also Figure 1-2 , a multi-index system evaluation method for water environment health of rivers in cold regions in winter, including the following steps:
[0084] S1. In order to quantitatively, systematically and operationally evaluate the health of river water environment in cold regions in winter, a cold region winter river water environment health evaluation index system is constructed, which includes target layer A, criterion layer B and indicator layer C.
[0085] Target layer A represents the goal to be achieved in constructing the cold region winter river index evaluation system, specifically, obtaining the health status of the water environment of cold region winter rivers.
[0086] Criterion layer B represents the dimensional factors that need to be considered in order to calculate the water environment health rating of cold regions in winter, specifically water resources B1, which reflects the storage and utilization of river water resources; water environment B2, which reflects the compliance of physical and chemical indicators of river water quality; water ecology B3, which reflects the survival and reproduction of aquatic organisms in rivers; social services B4, which reflects the public satisfaction with the social service functions of rivers; and habitat conditions B5, which reflects the basic environmental conditions of river banks.
[0087] The indicator layer C represents the secondary evaluation indicators involved in each dimension of the criterion layer B, including water resources development and utilization rate C1 under water resources B1, ecological flow satisfaction degree C2, river longitudinal connectivity index C3, water quality quality C4 under water environment B2, water function area compliance rate C5, Chao index C6 under water ecology B3, Shannon index C7, large benthic invertebrate biological integrity index C8, fish retention index C9, public satisfaction C1 under social services B4. 10 , Riparian vegetation coverage rate C under habitat condition B5 11 , Degree of artificial disturbance of riparian zone C 12 .
[0088] The construction of the evaluation index system is based on the reference to the "Technical Guidelines for River and Lake Health Assessment" and other domestic technical specifications, taking into full consideration the water environment characteristics of rivers in cold regions in winter, with the key point being to meet the needs of water environment management and protection in cold regions in winter. The selection of criteria in the criterion layer is based on the ability to comprehensively and objectively reflect the health status of the water environment of rivers in cold regions in winter, and the selection of evaluation indicators in the indicator layer follows the basic principles of systematicity, scientificity, representativeness of rivers in cold regions in winter, and easy access to specific implementation data.
[0089] S2. In order to plan the river water environment health classification standard, give priority to the control of rivers with poor health, and divide the river water environment health into five levels (very healthy, healthy, sub-healthy, unhealthy, and pathological). Each evaluation indicator in the indicator layer C is divided into the same level according to the corresponding specifications (I: very healthy; II: healthy; III: sub-healthy; IV: unhealthy; V: pathological). The specific classification is shown in Table 1:
[0090] Table 1 Classification of river water environment health levels in cold regions in winter
[0091]
[0092]
[0093]
[0094] Due to the influence of special circumstances such as natural freezing, the water resource utilization of rivers in cold regions in winter is relatively low. Therefore, when classifying the health index levels, in order to accurately reflect the health status of rivers in cold regions in winter, targeted standards should be set, specifically: I: 20-40%; II: 40-50%; III: 50-67%; IV: 67-75%; V: 75-100%.
[0095] The water resources development and utilization rate C1 is calculated as follows:
[0096]
[0097] Where: WURI is the water resources development and utilization rate, %; WS is the water supply of the river basin, 10,000 m 3 WR is the water resources in the river basin (including ice storage during the ice-covered period), 10,000 m 3 .
[0098] The annual distribution of rainfall and runoff in rivers in the cold regions of the north is uneven, especially in winter when the runoff drops significantly, with strong temporal heterogeneity. Therefore, when calculating the ecological flow satisfaction degree of rivers in cold regions in winter, the time period from October to March of the following year (the ice-covered period of rivers in cold regions in winter) should be focused on, and the health rating standards should be divided specifically, as follows: I: 30-40%; II: 20-30%; III: 10-20%; IV: 5-10%; V: 0-5%. The ecological flow satisfaction degree C2, represented by EF, is calculated as follows:
[0099]
[0100] Where: q d is the minimum daily average flow m 3 / s; Q is the multi-year average flow rate of the corresponding period m 3 / s; EF is the percentage of the minimum average daily flow from October to March of the following year (the period of river ice cover in cold regions in winter) to the multi-year average flow in the corresponding period.
[0101] The river longitudinal connectivity index C3 is determined based on the number of buildings or structures that affect river connectivity within a unit river length.
[0102] The water quality level C4 is based on various conventional water quality indicators and is divided into water quality grades according to the water quality classification standards in the "Surface Water Environmental Quality Standards" (GB3838-2002). The assessment is conducted based on the proportion of Class I to Class III water bodies according to the length of the river.
[0103] The compliance rate of water functional areas is C5, and is assessed by the percentage of the number of water functional areas that meet the standards to the number of assessed water functional areas.
[0104] Chao index C6 is used to estimate the number of OTUs in a community. The calculation formula is:
[0105]
[0106] Where: S obs is the number of OTUs actually observed; n1 is the number of OTUs with only one sequence; n2 is the number of OTUs with only two sequences.
[0107] The Chao Index is divided into grades based on expert scoring results, using a percentage system.
[0108] Shannon index C7 is used to estimate the index of microbial diversity in samples. The calculation formula is:
[0109]
[0110] Where: n i is the number of OTUs containing i sequences; N is the total number of sequences.
[0111] The Shannon Index is divided into grades based on expert scoring results, using a percentage system.
[0112] The biological integrity index of macrobenthic invertebrates, C8, is expressed as the BIBI index monitoring value, and the calculation formula is as follows:
[0113]
[0114] Wherein: BIBIS is the score for evaluating the biological integrity index of large benthic invertebrates in rivers; BIBIS is the monitoring value for evaluating the biological integrity index of large benthic invertebrates in rivers; BIBIE is the optimal expected value of the biological integrity index of large benthic invertebrates in the water ecological zone where the river is located.
[0115] The fish abundance index C9, expressed as FOEI, is used to assess the difference between the current fish species and the historical reference point fish species. The calculation formula is as follows:
[0116]
[0117] Where: FOEI is the fish endowment index, %; FO is the number of fish species obtained from the river assessment survey (excluding alien species), species; FE is the number of fish species assessed in the river before or in the 1980s, species.
[0118] Public Satisfaction C 10 , assess the public's satisfaction with the river environment, water quality and quantity, water-related landscapes, etc., and take the average score assigned by the public participating in the survey, on a percentage basis.
[0119] Riparian vegetation coverage C 11 , and the evaluation is carried out based on the total coverage of natural and artificial vegetation in the river banks obtained from the survey.
[0120] The degree of artificial disturbance of riparian zone C 12 , investigate the "four chaos" conditions of the river bank, and evaluate them based on the scoring standards specified in the "Technical Guidelines for River and Lake Health Assessment" on a 100-point scale.
[0121] S3. The analytic hierarchy process-machine learning (LightGBM) is used to calculate the weight value of each indicator data from both subjective and objective perspectives, and the combined weight is obtained by comprehensively optimizing the two methods based on game theory.
[0122] The LightGBM model is specifically: the gradient boosting decision tree is optimized through histogram-based decision tree construction, leaf growth decision tree algorithm, histogram difference calculation, gradient unilateral sampling, mutually exclusive feature bundling and sparse feature optimization to obtain the LightGBM model. The specific contents are as follows:
[0123] The gradient boosting decision tree is to gradually build a series of decision trees so that each tree reduces the residual of all previous trees. The calculation expression is as follows:
[0124] F m (x) = F m-1 (x)+γ m h m (x) (3)
[0125] In the formula, F m (x) is the model of the mth round; F m-1 (x) is the model of the m-1th round; h m (x) is the weak learner trained in the mth round; γx is the learning rate used to reduce the contribution of the new model;
[0126] The histogram-based decision tree construction optimizes the gradient boosting decision tree. Specifically, the eigenvalues are grouped to reduce the computational complexity. The calculation expression is as follows:
[0127]
[0128] In the formula, xi is the value of a sample on the i-th feature; n bin is the number of groups, min(x) represents the minimum value of feature x, and max(x) represents the maximum value of feature x;
[0129] The decision tree algorithm based on leaf growth optimizes the gradient boosting decision tree so that the LightGBM model adopts a leaf-growing strategy. In each iteration, the leaf with the largest gain among all current leaf nodes is selected for splitting. The calculation expression of the gain is as follows:
[0130]
[0131] In the formula, G L and G R is the sum of the gradients of the left child node and the right child node; H L and H R is the sum of the second-order derivatives of the left child node and the right child node; λ is the regularization parameter; γ is the penalty term for the number of split leaf nodes;
[0132] The gradient boosting decision tree is optimized based on histogram difference calculation to reduce the calculation amount of the LightGBM model. The calculation expression is as follows:
[0133] H left =H parent -H right (6)
[0134] In the formula, H right is the histogram of the right child node, H parent is the histogram of the parent node, H left is the histogram of the left child node;
[0135] The gradient boosting decision tree is optimized based on gradient one-side sampling, specifically the samples of the LightGBM model are optimized.
[0136] S31. Collect the graded values corresponding to the above evaluation indicators of the rivers to be evaluated, and standardize the data results to eliminate the differences in the dimensions of the indicators. Specifically, they are divided into positive indicators (the larger the better) and negative indicators (the smaller the better). The standardized values r ij The calculation formulas are as shown in formula (7) and formula (8).
[0137] Positive indicators:
[0138]
[0139] Negative indicators:
[0140]
[0141] In the formula, x ij is the value of the jth evaluation index in the i-th evaluation unit; min(x j ) represents the minimum value of the jth index in the original data, max(x j ) represents the maximum value of the jth index in the original data;
[0142] S32. Based on the analytic hierarchy process, the subjective weights of various indicators for evaluating the health of river water environment in cold regions in winter are obtained. When calculating the weights according to the analytic hierarchy process, the relative importance data are obtained by comparing the relative importance of experts in pairs, and the judgment matrix is constructed;
[0143] Collect the relative importance tables of each factor filled out by the experts, and obtain the relative importance a between each two factors for each expert ij The arithmetic mean of is filled into the corresponding judgment matrix. Taking the criterion layer B-target layer A and the indicator layer C-criterion layer B1 as examples, the relative importance scale table is shown in Table 2 and Table 3.
[0144] Table 2 Relative importance scale table of criterion layer B-target layer A
[0145] <![CDATA[Water Resource B1]]> <![CDATA[Water environment B2]]> <![CDATA[Water Ecology B3]]> <![CDATA[Social Service B4]]> <![CDATA[Habitat condition B5]]> <![CDATA[Water Resource B1]]> 1 0.2 0.17 0.2 0.25 <![CDATA[Water environment B2]]> 5 1 0.2 0.5 2 <![CDATA[Water Ecology B3]]> 6 5 1 4 5 <![CDATA[Social Service B4]]> 5 2 0.25 1 2 <![CDATA[Habitat condition B5]]> 4 0.5 0.2 0.5 1
[0146] The judgment matrix A1 of the criterion layer B-target layer A can be obtained:
[0147]
[0148] Table 3 Relative importance scale table of indicator layer C-criterion layer B1
[0149]
[0150] The judgment matrix A2 of the indicator layer C-criterion layer B1 can be obtained:
[0151]
[0152] Similarly, the judgment matrix A3 of the indicator layer C-criterion layer B2, the judgment matrix A4 of the indicator layer C-criterion layer B3, the judgment matrix A5 of the indicator layer C-criterion layer B4, and the judgment matrix A6 of the indicator layer C-criterion layer B5 are shown in the following formulas:
[0153]
[0154] A5=[1] (27)
[0155]
[0156] Calculate the maximum eigenvalue λ of the above judgment matrix respectively max , and perform consistency check. The consistency check table is shown in Table 4.
[0157] Table 4 Consistency test table
[0158]
[0159] The CR values of each judgment matrix are less than 0.1, which means that they pass the consistency test.
[0160] The judgment matrix is normalized so that the sum of the weights of each element in the final output weight vector is equal to 1. Taking the judgment matrix A1 as an example, the corresponding matrix A'1 after normalization is as follows:
[0161]
[0162] Solving the average value of each row of matrix A'1, we can get the single-order weight vector See below:
[0163]
[0164] Weight vector The values in the figure represent that the weight of water resources B1 in target layer A is 0.044, the weight of water environment B2 is 0.148, the weight of water ecology B3 is 0.509, the weight of social services B4 is 0.192, and the weight of habitat conditions B5 is 0.107.
[0165] Similarly, the single ranking weight vector corresponding to each judgment matrix can be obtained Summarized in Table 5:
[0166] Table 5 Single ranking weight vectors of each factor
[0167]
[0168]
[0169] After the total sorting consistency check is passed, the weight value of each indicator layer indicator for the target layer A can be calculated layer by layer.
[0170] The weight values of each indicator of water environment health evaluation in cold regions in winter based on the hierarchical analysis method are as follows: water resources development and utilization rate C1 (0.009), ecological flow satisfaction degree C2 (0.032), river longitudinal connectivity index C3 (0.004), water quality quality C4 (0.111), water function area compliance rate C5 (0.037), Chao index C6 (0.216), Shannon index C7 (0.176), large benthic invertebrate biological integrity index C8 (0.072), fish retention index C9 (0.045), public satisfaction C 10(0.192), riparian vegetation coverage rate C 11 (0.021), degree of artificial disturbance of riparian zone C 12 (0.086).
[0171] It can be seen that the weights of the Chao index C6 (0.216) and the Shannon index C7 (0.176), the characteristic indicators of winter rivers in cold regions, are relatively high, which have a great impact on the final evaluation results, thus verifying the necessity of incorporating them into the water environment health evaluation system of winter rivers in cold regions.
[0172] S33. Obtain the objective weights of each indicator of river water environment health evaluation in cold regions in winter based on machine learning (LightGBM). When machine learning calculates the weights, it determines the weights of each indicator based on the amount of information provided by the specific data collected on the river water environment health evaluation indicators in cold regions in winter. The specific process is as follows:
[0173] S331: First, the training instances are sorted in descending order according to the absolute value of the gradient. Second, the algorithm retains the head a×100% data instances with larger gradients to obtain a subset of instance a. For the remaining set A consisting of (1-a)×100% data instances with smaller gradients c , randomly select a size of b×|A c |Data subset B. Finally, according to the estimated variance gain Partition the subset A∪B:
[0174]
[0175] Where: A l ={x i ∈A:x ij ≤d},A r ={x i ∈A:x ij >d}, B r ={x i ∈B:x ij ≤d},B r ={x i ∈B:x ij >d}. Coefficient Used to normalize the gradient on B to A c size.
[0176] S332: Combine the LightGBM tree model established above with the indicators selected in step S1 to rank the importance of regional water quality characteristics. The more times an indicator feature makes key decisions (splitting) in the decision tree, the higher its relative importance. The feature importance score is calculated based on the number of times each indicator feature splits the training data in all trees. The importance of the indicator feature obtained by the LightGBM operation is denoted as F[i]. The indicator feature importance F[i] is numerically sorted. The expression for scale ranking is RANK(X), then the order of each indicator after scale ranking is RANK(F[i]), and the ranking result reflects the importance of each indicator.
[0177] S333: Determine the objective weight of LightGBM based on the order centroid (ROC). The ROC method maintains the sorting order and minimizes the maximum error of the weight, which can reduce the risk of overlap due to inappropriate weights. The ROC calculation formula is as follows:
[0178]
[0179] Where N is the number of indicators, RANK(F[i]) is the ranking result of feature importance, and W o is the objective weight.
[0180] S34, using the game theory method, based on the constructed objective function, seeks the coordination and balance between subjective and objective weights. The specific process is as follows:
[0181] S341: Construct any linear combination of vectors ω:
[0182]
[0183] Where: L is the number of methods for determining weight coefficients; α k is the linear combination weight coefficient α k ={α1, α2, ..., α L};ω k is the basic weight vector set; Table ω k The transpose of
[0184] S342: With ω and ω k The goal is to minimize the deviation of, construct the objective function, and obtain the combined weight coefficient α in ω k , the calculation formula is:
[0185]
[0186] The linear equation of the first-order derivative condition optimized in formula (12) is:
[0187]
[0188] S343: The obtained combined weight coefficient α k Normalized processing to obtain α k * , combined with formula (11), we get the final optimized combined weight value ω * , the calculation formula is:
[0189]
[0190] Where: α k * is the normalized combined weight coefficient.
[0191] Step S4, using the improved matter-element extension model combined with the combined weights to evaluate the health of the river water environment.
[0192] According to the construction rules, a multi-index system evaluation algorithm for water environment health of rivers in cold regions in winter is developed. j for:
[0193]
[0194] Improved matter-element extension model for classical domain R j Normalize the index by dividing both ends of the range by the region R. p The maximum value q of each indicator parameter range i , and obtain the new classical domain R k for:
[0195]
[0196] The collected values of various evaluation indicators of river water environment health in cold regions in winter are used to construct the object element R x After that, the new object element R is obtained by normalization according to formula (16): L . And according to formula (17) and formula (18), we can get R L With R K The distance D between the evaluation index value range ji , and the comprehensive correlation K j (N).
[0197] If K j =max{K j (N)}, the comprehensive rating of the water environment health of the river to be evaluated is j.
[0198] When the comprehensive rating of the water environment health of the river to be evaluated is I (very healthy) or II (healthy), it can be considered that the water environment health of the river is relatively good. There is no need to invest additional time and energy in the management of the river's health status. A health assessment can be conducted once a quarterly random inspection.
[0199] When the comprehensive rating of the water environment health of the river to be evaluated is III (sub-healthy), it can be considered that the water environment health of the river is unstable. It is recommended to conduct a water environment health assessment once a month to timely understand the changing trend of the river water environment health and avoid further development to a severe condition.
[0200] When the comprehensive rating of the water environment health of the river to be evaluated is IV (unhealthy) or V (sick), it can be considered that there are problems with the health of the river, and the water environment health status should be monitored weekly. And according to the problem indicators pointed out in the process of water environment health evaluation, corresponding control measures should be taken in a targeted manner.
[0201] Therefore, the present invention adopts the above-mentioned multi-index system evaluation method for the water environment health of rivers in cold regions in winter, which has strong operability and high credibility, can accurately reflect the true health status of the water environment of rivers in cold regions in winter, and provides support for managers to fully grasp the current status of river water environment and maintain the stability of regional ecological environment. It has important practical significance for scientific management decisions of rivers.
[0202] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.
Claims
1. A multi-index system evaluation method for water environment health of rivers in cold regions in winter, characterized by: The following steps are involved: Step 1: Construct a cold region winter river water environment health evaluation index system including target layer, criterion layer and indicator layer; Step 2: Collect data on various indicators of the river, divide the health status of the river water environment into five levels, namely very healthy, healthy, sub-healthy, unhealthy, and pathological, and divide each evaluation indicator into the same level level accordingly; Step 3: Use the analytic hierarchy process and LightGBM model to calculate the weight of each indicator data from both subjective and objective perspectives, and based on the game theory method, use Nash equilibrium as the coordination goal to find the consistency and coordination between subjective and objective weights to obtain the combined weight; Step 4: Use the improved matter-element extension model combined with the combined weights obtained in step 3 to calculate the correlation of river health, and determine the river water environment health evaluation level based on the maximum value of the correlation.
2. The method for evaluating the water environment health of rivers in cold regions in winter according to claim 1 is characterized by: In step 1, the target layer is the health status of river water environment in cold regions in winter; the criterion layer includes water resources, water environment, water ecology, social services, and habitat conditions; water resources indicators include water resources development and utilization rate, ecological flow satisfaction degree, and river longitudinal connectivity index; water environment indicators include water quality and water function area compliance rate; water ecology indicators include Chao index, Shannon index, large benthic invertebrate biological integrity index, and fish retention index; social service indicators include public satisfaction; habitat conditions indicators include riparian vegetation coverage and riparian artificial disturbance degree.
3. The method for evaluating the water environment health of rivers in cold regions in winter according to claim 2 is characterized by: The Chao index and Shannon index are used as the specific indicators of rivers in cold regions in winter in the index layer.
4. The method for evaluating the water environment health of rivers in cold regions in winter according to claim 3 is characterized by: When calculating the water resources development and utilization rate and the ecological flow satisfaction degree, the characteristics of rivers in cold regions in winter are taken into consideration because there is a freezing period in rivers in cold regions in winter. The calculation expression is as follows: The calculation expression of water resources development and utilization rate is as follows: Where WURI is the water resources development and utilization rate, in %; WS is the water supply of the river basin, in 10,000 m 3 WR is the water resources in the river basin, including the ice storage during the ice-covered period, in units of 10,000 m 3 ; The calculation expression of ecological flow satisfaction is as follows: In the formula, q d is the minimum daily average flow, in m 3 / s; Q is the multi-year average flow rate in the corresponding period, in m 3 / s; EF is the percentage of the minimum average daily flow during the winter ice-covered period of cold regions to the multi-year average flow during the corresponding period.
5. The method for evaluating the water environment health of rivers in cold regions in winter according to claim 4 is characterized in that: The LightGBM model is specifically: the gradient boosting decision tree is optimized through histogram-based decision tree construction, leaf growth decision tree algorithm, histogram difference calculation, gradient unilateral sampling, mutually exclusive feature bundling and sparse feature optimization to obtain the LightGBM model. The specific contents are as follows: The gradient boosting decision tree is to gradually build a series of decision trees so that each tree reduces the residual of all previous trees. The calculation expression is as follows: F m (x)=F m-1 (x)+γ m h m (x) (3) In the formula, F m (x) is the model of the mth round; F m-1 (x) is the model of the m-1th round; h m (x) is the weak learner trained in the mth round; γm is the learning rate used to reduce the contribution of the new model; The histogram-based decision tree construction optimizes the gradient boosting decision tree. Specifically, the eigenvalues are grouped to reduce the computational complexity. The calculation expression is as follows: In the formula, x i is the value of a sample on the i-th feature; n bin is the number of groups, min(x) represents the minimum value of feature x, and max(x) represents the maximum value of feature x; The decision tree algorithm based on leaf growth optimizes the gradient boosting decision tree so that the LightGBM model adopts a leaf-growing strategy. In each iteration, the leaf with the largest gain among all current leaf nodes is selected for splitting. The calculation expression of the gain is as follows: In the formula, G L and G R is the sum of the gradients of the left child node and the right child node; H L and H R is the sum of the second-order derivatives of the left child node and the right child node; λ is the regularization parameter; γ is the penalty term for the number of split leaf nodes; The gradient boosting decision tree is optimized based on histogram difference calculation to reduce the calculation amount of the LightGBM model. The calculation expression is as follows: H left =H parent -H right (6) In the formula, H right is the histogram of the right child node, H parent is the histogram of the parent node, H left is the histogram of the left child node; The gradient boosting decision tree is optimized based on gradient one-side sampling, specifically the samples of the LightGBM model are optimized.
6. The method for evaluating the water environment health of rivers in cold regions in winter according to claim 5 is characterized in that: Step 3 The process of obtaining the combined weight is as follows: S31. Standardized processing index data, specifically divided into positive indicators and negative indicators, the standardized value r ij The calculation formula is as follows: Positive indicators: Negative indicators: In the formula, x ij is the value of the jth evaluation index in the i-th evaluation unit; min(x j ) represents the minimum value of the jth index in the original data, max(x j ) represents the maximum value of the jth index in the original data; S32. Obtain the subjective weights of various indicators for evaluating river water environment health in cold regions in winter based on the analytic hierarchy process; S33, based on the LightGBM model, the objective weights of various indicators for evaluating the health of river water environment in cold regions in winter were obtained; S34. Use game theory methods and Nash equilibrium as the coordination goal to find the consistency and coordination between subjective and objective weights and obtain the combined weights.
7. The method for evaluating the water environment health of rivers in cold regions in winter according to claim 6 is characterized in that: The process of obtaining the objective weights of various indicators for evaluating the health of river water environment in cold regions in winter based on the LightGBM model in S33 is as follows: S331. First, sort the training instances in descending order according to the absolute value of the gradient. Second, retain the head a×100% data instances with larger gradients to obtain a subset of instance a. For the remaining set Ac consisting of (1-a)×100% data instances with smaller gradients, randomly select a data subset B of size b×|Ac|. Finally, according to the estimated variance gain The subset A∪B is divided and the calculation expression is as follows: In the formula, A l ={x i ∈A:x ij ≤d},A r ={x i ∈A:x ij >d},B l ={x i ∈B:x ij ≤d}, B r ={x i ∈B:x ij >d}; coefficient Used to normalize the gradient on B to A c Size; g i Represents the gradient of the data; S332, combining the LightGBM model with the selected indicators, ranking the importance of regional water quality characteristics, recording the importance of the indicator characteristics obtained by the LightGBM model as F[i], and ranking the indicator characteristics F[i] by numerical value, the expression of scale ranking is RANK(X), and the order of each indicator after scale ranking is RANK(F[i]); S333. Determine the objective weight of the LightGBM model based on the order centroid. The calculation formula of the order centroid is as follows: In the formula, N is the number of indicators, RANK(F[i]) is the ranking result of feature importance, and W o is the objective weight.
8. The method for evaluating the water environment health of rivers in cold regions in winter according to claim 7 is characterized in that: S34 adopts the game theory method to find the consistency and coordination between subjective and objective weights with Nash equilibrium as the coordination goal. The process of obtaining the combined weight is as follows: S341. Construct any linear combination of vectors ω: Where L is the number of methods for determining weight coefficients; α k is the linear combination weight coefficient α k ={α1, α2, ..., α L };ω k is the basic weight vector set; Table ω k The transpose of S342, with ω and ω k The goal is to minimize the deviation of, construct the objective function, and obtain the combined weight coefficient α in ω k , the calculation formula is: S343, the combined weight coefficient α obtained by the above calculation k Normalize to get α k * , combined with formula (11), the final optimized combined weight value ω* is obtained:
9. The method for evaluating the water environment health of rivers in cold regions in winter according to claim 8, characterized in that: In step 4, the improved matter-element extension model is combined with the combined weights obtained in step 3 to calculate the correlation degree of river health as follows: S41: Construct a classical domain and normalize it to obtain a new classical domain R k : Where N j is the jth evaluation level; C i is the i-th evaluation index of level j; (a ji , b ji ) is the value range of the i-th evaluation index under level j; q i is the maximum value of each indicator parameter range; (a ji ′,b ji ′) is the value range of the evaluation index after normalization; S42: Constructing the section domain R p R x , and normalize the object-element to be evaluated to obtain the new object-element R L : Where N p represents the total number of evaluation categories, (a pi , b pi ) means C i The range of the section domain; In the formula, v i is the value of the i-th evaluation index in the object-element to be evaluated; S43: Calculate the new matter-element R L About New Classical Domain R K The distance D of the measurement range ji , the calculation formula is: D ji =|v i ′-(a ji ′+b ji ′) / 2|-(b ji ′-a ji ′) / 2 (17) S44: Calculate the comprehensive correlation K j (N), if K j =max{K j (N)}, the river water environment health evaluation level is j, and the river water environment health status in cold regions in winter is obtained, K j The calculation formula for (N) is:
Citation Information
Cited By
Artificial intelligence-driven water ecology sensitive area coupling analysis and evaluation method
CN121787984A
An artificial intelligence driven water ecological sensitive area coupling analysis and evaluation method
CN121787984B
