Bayesian network risk model construction method for reservoir dam risk evaluation

Abstract

The invention discloses a Bayesian network risk model construction method for reservoir dam risk evaluation, and the method comprises the steps: constructing a historical dam break database, analyzing risk causes, and building a behavior influence factor index system comprising internal factors and external factors; performing factor correlation analysis by using an approximate value matrix and a Pearson's correlation coefficient, determining subjective weights of indexes at all levels by using an analytic hierarchy process of expert scoring judgment, determining objective weights of the indexes by using a BP neural network, and performing weighted combination on the weights; establishing a reservoir dam Bayesian network risk model considering human factors according to an analysis result, establishing an evaluation standard of each secondary index in the Bayesian network in combination with engineering practice, and formulating a judgment basis of each grade in detail according to the evaluation indexes so as to quantify a risk error probability and improve the risk accuracy. And calculating a reliability result through the Bayesian network and the combined weight of each level of indexes, and taking the reliability result as an index for judging the risk level of the reservoir.

Description

Technical Field

[0001] This invention relates to a Bayesian network risk model construction method for reservoir dam risk assessment based on clustering and subjective / objective weight analysis, belonging to the field of dam safety prediction technology in water conservancy engineering. Background Technology

[0002] Dams play a vital role in flood control, power generation, water supply, irrigation, navigation, and ecological environment improvement. However, they also pose risks to downstream life, property, infrastructure, and the environment. The losses from dam failures can be enormous. Frequent extreme weather events and aging reservoirs, among other disaster-causing factors, increase the probability of dam failures. Rapid economic and social development exacerbates the losses from dam failures. Dam safety is no longer just an engineering safety concern for water administrative departments and reservoir management units at all levels; it is also a public safety issue of concern to the government and the public.

[0003] Dam risk assessment and control has become an indispensable and crucial component of all dam engineering projects worldwide. It involves the integration of multiple indicators, and accurately assessing and mitigating the risk level of a reservoir remains a challenge. Dam risk refers to the measure of the likelihood and severity of a dam's negative impacts on life, health, property, and the environment. Risk factors include both engineering factors and human activity factors.

[0004] Humans are the primary actors in reservoir dam management, operating the human-machine system. With the continuous development of science and technology, policies such as reservoir standardization, digital twins, and operation and management matrices have been implemented, monitoring facilities are gradually expanding coverage, and operation and management systems are being upgraded iteratively. However, technological advancements cannot fully guarantee the safety of reservoir dams; the ability to manage risks at the grassroots level and effectively utilize systems is also a crucial factor in risk management. Existing research primarily focuses on engineering risks within the risk factors of reservoir dams, with limited research on risks arising from human activities. A risk network considering the clustering relationships between various influencing factors has not been established, and a specific evaluation standard system has not been generated, resulting in incomplete risk assessment factors and inaccurate dam risk assessments. Summary of the Invention

[0005] Purpose of the invention: To address the problems and shortcomings of existing technologies, this invention provides a method for constructing a Bayesian network risk model for reservoir dam risk assessment based on clustering and subjective / objective weight analysis.

[0006] Technical Solution: A method for constructing a Bayesian network risk model for reservoir dam risk assessment. This method involves building a historical dam failure database, analyzing risk causes, and establishing a behavioral influencing factor index system including internal and external factors. Factor correlation analysis is performed using an approximation matrix and Pearson correlation coefficient. The analytic hierarchy process (AHP) with expert scoring is used to determine the subjective weights of each level of indicator, and a backpropagation (BP) neural network is used to determine the objective weights of the indicators. These weights are then weighted and combined. Based on the analysis results, a Bayesian network risk model for reservoir dams considering human factors is established. Evaluation criteria for each secondary indicator in the Bayesian network are established in conjunction with engineering practice, and detailed judgment criteria for each level are formulated based on the evaluation indicators to quantify the probability of risk failure. The reliability result is calculated by combining the weights of the Bayesian network and each level of indicators, serving as an indicator for judging the reservoir risk level. The construction method includes the following steps: S1, based on historical dam failure data, constructs a historical dam failure database.

[0007] S2. Classify the causes of errors to obtain the root causes of risks in dam failure accidents and conduct importance analysis.

[0008] S3 analyzes the factors influencing dam risks and establishes an indicator system for error-related factors. The indicator system is divided into internal and external factors. Internal factors include two primary indicators: management and individuals. External factors include two primary indicators: system and environment. Management is further divided into four secondary indicators: supervision intensity, risk awareness, situation assessment, and daily management. Individuals are divided into three secondary indicators: sense of responsibility, experience, and professional skills. System is divided into three secondary indicators: organizational structure, institutional development, and emergency response plans. Environment is divided into two secondary indicators: external environment and internal environment.

[0009] S4. Factor correlation analysis was performed using the approximation matrix and Pearson correlation coefficient to obtain the cluster analysis phylogenetic diagram and the relationship between the influencing factors, which served as the basis for the connection of the directed acyclic graph of the Bayesian network.

[0010] S5, based on expert scoring, uses the analytic hierarchy process (AHP) to determine the subjective weights of each level of indicators.

[0011] S6 uses a BP neural network to calculate the objective weights between indicators.

[0012] S7 uses empirical factors to weight the subjective and objective weights.

[0013] S8. A directed acyclic graph of the Bayesian network is obtained from the clustering and correlation analysis results. The conditional probabilities between nodes are obtained based on expert evaluation scores and BP neural network results. A risk model based on the Bayesian network is then established.

[0014] S9 combines the error impact factor index system with the actual situation of reservoir dam engineering, establishes evaluation standards for each index, and realizes the qualitative and quantitative conversion of the reliability of each index.

[0015] S10: Perform risk assessment calculations on the obtained Bayesian network risk model, and use the results as an indicator to evaluate the risk of the reservoir dam.

[0016] The specific implementation process of S4 is as follows: S4.1, let the impact factor score of the j-th error in the i-th reservoir be... Then, establish the evaluation characteristic matrix A: S4.2 introduces Pearson correlation coefficient for cluster analysis to measure the correlation among error-influencing factors. Error-influencing factors and The Pearson correlation coefficient between them is calculated as follows: In the formula, m is the number of variables, that is, each error influencing factor has m reservoir scores; and Error influencing factors and The average value, yes and covariance; yes Standard deviation; yes Standard deviation; yes and The Pearson correlation coefficient. The Pearson correlation coefficient is between [-1, 1], and the larger its absolute value, the stronger the correlation between the variables. When When, it means and A positive correlation; when When, it means and A negative correlation; when When, it means and A perfect positive correlation; express and A perfect negative correlation; when Time indicates and Irrelevant.

[0017] The specific implementation process of S5 is as follows: S5.1, Establish a hierarchical structure model.

[0018] Section S5.2 describes pairwise comparisons between indicators and the construction of the judgment matrix, where the indicators refer to error impact factors. After establishing the hierarchical structure, judgments need to be made based on the relative importance of each indicator at each level. These judgments are expressed numerically using appropriate scaling and written in the form of a judgment matrix. Generally, the judgment matrix should be independently provided by an expert familiar with the problem.

[0019] For n error-affecting factors, the judgment matrix can be obtained through pairwise comparisons. .

[0020]

[0021] in Let represent the importance of error impact factors i and j relative to the target. The judgment matrix C has the following properties: 1) 2) 3) 4) C is a matrix with positive and negative sides. If for any i, j, k, we have .

[0022] S5.3 Calculate the weights of each judgment matrix and perform a consistency check.

[0023] 1) Calculate the geometric mean of all elements in each row of the judgment matrix. :

[0024] 2) Normalization and calculation : but The desired eigenvector is, i.e. The desired subjective weight.

[0025] 3) Calculate the largest eigenvalue of the judgment matrix. :

[0026] In the above formula, For vectors The i-th element.

[0027] 4) Calculate the consistency index (CI) and perform a consistency test. The formula is as follows:

[0028] In the above formula, n is the order of the judgment matrix, and the random consistency index RI is determined by the order. Calculate the ratio CI / RI. When CI / RI < 0.1, the consistency of the judgment matrix meets the requirements. Otherwise, re-judgment is performed, and a new judgment matrix is ​​written.

[0029] The specific implementation process of S6 is as follows: S6.1, use a BP neural network to calculate the objective weights between indicators. By inputting dam failure data from the historical dam failure database into the BP neural network, the appropriate indicator weights are determined through multiple fittings, which are the desired objective weight values.

[0030] S6.2, the basic idea of ​​a BP neural network is: weighting, summation, and transition. Input layer input... x 1 , x 2 , x 3 , ..., x n The output layer provides index scores for each dam failure case and control group in the historical dam failure database; y 1 , y 2 , y 3 , ..., y n , where is the risk value for each case and control group in the historical dam failure database; if a dam fails, it is set to 1.

[0031] In this invention, the BP neural network is configured with three layers. ω g1 , ω g2 , ω g3 , ..., ω gn For the first g The connection weight coefficients between each neuron and the neurons in the previous layer. g =1,2,3; b g For threshold; f(·) Let be a transfer function. Then the _____ g The net input value of a neuron can be represented as:

[0032] in,X =[ x 1 ,x 2 , … ,x i … ,x n ] T , W =[ ω g1 ,ω g2 ,…,ω gi ,…,ω gn ].

[0033] If let x 0 =1, ω g0 = b g ,but X and W g It can be changed to:

[0034]

[0035] g The net input of a node can be expressed by the following formula:

[0036] Then the first g The output of each neuron y g for:

[0037] because f(·) It is a monotonically increasing function with an upper limit, so the signal transmitted by the neural network also has an upper limit.

[0038] S6.3, run the BP neural network, input several sets of data from the historical dam failure database, continuously fit, and obtain the connection weight coefficients of neurons in each layer. The weight transmission result of the multi-layer neural network connecting each indicator factor is the weight of each indicator factor.

[0039] In the formula, g =1,2,3, where 1,2,3 represent the number of layers in the hidden neural network, and the objective weights are calculated. .

[0040] The specific implementation process of S7 is as follows: S7.1. Let the weights obtained by the analytic hierarchy process be subjective weights. The weights obtained by fitting a BP neural network are objective weights. Then, the comprehensive weighting using the combination of empirical factors is: In the formula, Subjective weight and objective weight Empirical factors at the time of combination The weighting is determined by the evaluator and reflects the evaluator's preference for subjective and objective weighting. The larger the value, the more the evaluator values ​​subjective weight, i.e., the expert evaluation in this invention; The smaller the value, the more the evaluator values ​​objective evaluation.

[0041] The specific implementation process of S8 is as follows: S8.1. A Bayesian network is a directed acyclic graph (DAG), consisting of nodes representing variables and directed edges connecting these nodes. A Bayesian network comprises two parts: a DAG with N nodes, where the nodes represent random variables and the directed edges between nodes represent the relationships between them; and a conditional probability table associated with each node, representing the relationship between the node and its parent node—the conditional probability relationship. Drawing the connections between influencing factors in a Bayesian network based on the relationships between these influencing factor indicators represents the connection relationships between the nodes in the graph.

[0042] Based on the error impact factor index system established in S3, and the cluster analysis spectrum of impact factor indices and the interrelationships between impact factors obtained in S4, combined with the analysis of human behavior mechanisms, a Bayesian network directed acyclic graph (DAG) is constructed to establish node relationships. A hierarchical modeling approach is adopted, with each layer only influenced by its immediate parent layer. Therefore, during the modeling process, related factors are placed in the same group as much as possible to reduce computational load while maintaining model reliability. After hierarchical design, the influence relationships of PSFs (Personal Behavior Influence Factors) are clarified and simplified.

[0043] S8.2. Further determine the parameters of the model, that is, determine the conditional distribution of each node in the Bayesian network. The following rules are adopted for this.

[0044] 1) Each node has a binary value (although it uses binary values, the probability of being associated with it can also be regarded as a measure of "degree", thus more accurately reflecting the actual situation). For consistency, these binary states are collectively referred to as "yes / no".

[0045] 2) If a node Y in the model has n parent nodes, represented as follows: Based on historical dam failure database records and combined with opinions from domain experts, each root node is assigned a weight ranging from 0 to 1, denoted as . , which represents the degree of influence of the root node on node Y. The larger the value, the greater the degree of influence.

[0046] when k nodes in When the value of node Y is "yes" and the values ​​of all other nodes are "no", the probability that node Y is "yes" is: .

[0047] A computer device, characterized in that: the computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the steps of the Bayesian network risk model construction method for reservoir dam risk assessment as described above.

[0048] A computer-readable storage medium storing a computer program that performs the Bayesian network risk model construction method for reservoir dam risk assessment as described above.

[0049] Beneficial Effects: This invention utilizes historical dam failure data to establish a fault impact factor index system encompassing internal and external factors. Factor correlation analysis is performed using approximation matrices and Pearson correlation coefficients to obtain cluster analysis phylogenetic diagrams and the relationships between impact factors. The analytic hierarchy process (AHP) with expert scoring is used to determine the subjective weights of each level of indicator, while a backpropagation (BP) neural network is used to determine the objective weights, and these weights are then weighted and combined. Based on this, a Bayesian network risk model for reservoir dams considering human behavioral factors is established. Evaluation criteria for each secondary indicator in the Bayesian network are established in conjunction with engineering practice, and detailed judgment criteria for each level are formulated based on the evaluation indicators to quantify the probability of risk failure. The reliability results calculated by combining the weights of the Bayesian network and each level of indicators serve as indicators for assessing reservoir risk levels. This model can be applied in engineering practice, providing a more scientific and intuitive measure of reservoir dam risk management levels, offering a scientific basis for reservoir risk management rating and improvement, and has broad application prospects. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the error impact factor index system according to an embodiment of the present invention; Figure 2 This is an iceberg diagram of the cluster analysis of the error impact factor index system in this invention embodiment; Figure 3 This is a cluster analysis phylogenetic diagram (dendrogram) of the error impact factor index system in this embodiment of the invention. Figure 4This is a flowchart illustrating the AHP method for calculating weights according to an embodiment of the present invention; Figure 5 This is a schematic diagram of a BP neural network according to an embodiment of the present invention; Figure 6 This is a Bayesian network risk model of the error impact factor index system in this invention embodiment; Figure 7 This is the Bayesian network risk model in the early warning and forecasting of this invention. Figure 8 This is a Bayesian network risk model in disaster relief intervention according to an embodiment of the present invention. Detailed Implementation

[0051] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading the present invention, any modifications of the present invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0052] A method for constructing a Bayesian network risk model for reservoir dam risk assessment includes the following steps: Step 1: Construct a historical dam failure database based on data from more than 3,500 dam failure accidents that have occurred in my country's history.

[0053] Step 2: Analyze and classify the causes of errors to identify the root causes of risks in dam failure accidents and conduct an importance analysis.

[0054] Typical dam failure cases were selected from the historical dam failure database proposed in Step 1 as specific research objects for dam failure accident risk research and mining, in order to obtain the root causes of human behavior risks in dam failure accidents and their main influencing factors, and to explore their inherent laws and connections.

[0055] Analysis of typical dam failure cases reveals that the main root causes of human behavioral risks in dam failure accidents include violations of regulations, decision-making errors, emotional stress, lack of safety awareness, insufficient experience, inadequate professional skills, a management system lacking corrective mechanisms, ineffective rescue efforts, lack of emergency plans, poor organizational management, inappropriate allocation of management institutions, weak sense of responsibility, and ineffective supervision. Examples of some reservoir analyses are shown in Table 1.

[0056] Table 1. Root causes of human behavior risks in typical dam failure cases

[0057] Taking into account the characteristics of reservoir risks, the factors influencing failures are divided into internal and external factors. Internal factors include two primary indicators: management and individuals. External factors, in addition to the systemic indicator, also consider environmental factors. Management is further divided into four secondary indicators: supervision intensity, risk awareness, situation assessment, and daily management. Individuals are divided into three secondary indicators: sense of responsibility, experience, and professional skills. The systemic indicator is divided into three secondary indicators: organizational structure, system construction, and emergency response plan. The environmental indicator is divided into two secondary indicators: external environment and internal environment. The external environment mainly refers to the flood and earthquake levels and meteorological conditions in the dam site area, while the internal environment refers to whether there are any engineering safety hazards in the dam itself.

[0058] The specific settings of the resulting indicator system are as follows: Figure 1 As shown.

[0059] Step 4: Use the approximate value matrix and Pearson correlation coefficient to perform factor correlation analysis, and obtain the cluster analysis phylogenetic diagram and the relationship between the influencing factors.

[0060] 4.1 The error impact factor system obtained in step three is applied to the historical dam failure database obtained in step one to obtain historical data scores for each indicator. Since the impact factors in the system do not have specific values ​​and the data cannot be determined, the evaluation criteria are divided into five levels, including "good, relatively good, average, not so good, and poor", as shown in Table 2.

[0061] Table 2 Qualitative-Quantitative Probability Conversion Table for Influencing Factors

[0062] The data source is historical dam failure data from the historical dam failure database, including m dam failure cases and corresponding n influencing factors. The evaluation criteria for each influencing factor are obtained from the textual descriptions of each influencing factor in the historical dam failure records, namely, qualitative descriptions of "good, relatively good, average, not so good, poor". Combined with the qualitative-quantitative probability conversion relationship in Table 2, the corresponding quantitative probabilities are obtained.

[0063] The normalization formula for each quantitative probability is as follows: set up The first in the dam failure data i The first reservoir, j Each influencing factor, each indicator The maximum value of the corresponding data is The minimum value is Then the within-group variable becomes

[0064] Therefore, the input table of influencing factors of dam failure accidents is obtained, as shown in Table 3. The source is historical dam failure cases from the aforementioned historical dam failure records, obtained through textual recording, qualitative judgment, quantitative probability transformation, and normalization. The table shows the first... i The first reservoir j Each impact factor score is the corresponding normalized value. .

[0065] Table 3. Dam Failure Accident Input Table

[0066] Since the dam failure database only contains dam failure records, corresponding to a dam failure risk of 1, a control group is introduced to avoid bias in the results obtained during data analysis and the application of the BP neural network. This control group also serves as the data analysis database. The data for the control group comes from risk assessment reports of several normally operating reservoirs. Qualitative evaluation criteria are derived from the textual descriptions of each indicator. The qualitative-quantitative probability conversion relationship in Table 2 is used to obtain the quantitative probability, which is then normalized. The results are shown in Table 4.

[0067] Table 4 Control Group

[0068] The data in Tables 3 and 4 above are the data sources for the subsequent correlation analysis of influencing factors in step four and the determination of objective weights using the BP neural network in step six.

[0069] 4.2 Let the first j The error impact factor is in the first... i The impact factor score of each reservoir is: Then, establish the evaluation characteristic matrix A: S4.2 introduces Pearson correlation coefficient for cluster analysis to measure the correlation among error-influencing factors. Error-influencing factors and The Pearson correlation coefficient between them is calculated as follows: In the formula, m is the number of variables, that is, each error influencing factor has m reservoir scores; and Error influencing factors and The average value, yes and covariance; yes Standard deviation; yes Standard deviation; yes and The Pearson correlation coefficient is calculated. The Pearson correlation coefficient ranges from -1 to 1; the larger the absolute value, the stronger the correlation between the variables. The correlation strength criteria are shown in Table 5.

[0070] Table 5. Pearson Coefficient Correlation Criteria

[0071] when When, it means and A positive correlation; when When, it means and A negative correlation; when When, it means and A perfect positive correlation; express and A perfect negative correlation; when Time indicates and Irrelevant.

[0072] 4.4 Because influencing factors are interconnected and can influence each other or have a combined effect on lower-level indicators, the relationships between variables need to be considered when constructing the directed acyclic graph of a Bayesian network. Therefore, the similarity between influencing factors can be obtained from the approximate value matrix. Based on the similarity between influencing factors, a clustering table of influencing factors can be obtained. Influencing factors can be clustered stepwise to determine the dependencies between them. Influencing factors with high similarity can be clustered to reduce the complexity of probability calculation in the Bayesian network.

[0073] The approximate value matrix obtained by calculation is shown in Table 6.

[0074] Table 6 Approximate Value Matrix of Influencing Factors

[0075]

[0076] The approximation matrix consists of the Pearson coefficients between various influencing factors, representing the similarity between the factors. The closer the result is to 1, the greater the similarity, and the closer the two are.

[0077] The clustering table of influencing factors is shown in Table 7.

[0078] Table 7 Clustering Table of Influencing Factors

[0079]

[0080] The table above illustrates the process of progressive clustering of variables. The influencing factors—supervision intensity, risk awareness, situation assessment, daily management, sense of responsibility, experience, professional skills, organizational structure, system construction, emergency plans, external environment, and internal environment—are labeled 1 through 12, undergoing a total of 11 clustering stages. The combined clustering column illustrates the process of variables being progressively aggregated: the first row contains 1 and 4, representing supervision intensity and daily management, with an approximate value of 0.779, indicating they are the two largest and most similar factors. The second row contains 1 and 3, representing supervision intensity, daily management, and situation assessment after clustering. The remaining rows follow the same principle.

[0081] From the approximation matrix in Table 6 and the clustering table in Table 7, we obtain the clustering analysis ice map and phylogenetic diagram, as follows: Figure 2 and Figure 3 As shown.

[0082] In the icefall chart, each influencing factor occupies a column (represented by light blue), while a separator column (represented by dark blue) is reserved between the influencing factors. The length of the separator column indicates the similarity relationship between adjacent quantitative elements. Figure 2 In the data, the column separating the supervision intensity and daily management is almost full, indicating that these two variables are highly similar and were aggregated very early. In contrast, the column separating the sense of responsibility and system construction is very short, indicating that they are less similar and were aggregated later.

[0083] Figure 3 This is a cluster analysis phylogenetic diagram (tree diagram), which intuitively represents the clustering process of each factor and visually reduces the dimensionality of elements according to distance. As shown in the diagram, supervision intensity, daily management, and situation assessment were clustered into one group early on; risk awareness and external environment were clustered into another; emergency plans and internal environment were clustered into yet another, and subsequently clustered with the former. Experience, professional skills, organizational structure, system construction, and sense of responsibility were clustered into yet another group. The cluster analysis phylogenetic diagram clearly expresses the similarity relationships and clustering order of each influencing factor. Bayesian networks can demonstrate the causal relationships and correlation strength between factors. The connections between nodes prove that there are logical relationships and strong connections between influencing factors, thus yielding the topological network of the Bayesian network.

[0084] Step 5: Based on the expert scores, use the analytic hierarchy process (AHP) to determine the subjective weights of each level of indicators.

[0085] The Analytic Hierarchy Process (AHP) decomposes complex problems into constituent factors, groups these factors according to their dominance relationships to form an ordered hierarchical structure, and determines the relative importance of factors in each level through pairwise comparisons, constructing a comparison judgment matrix. Finally, various calculations are performed on the judgment matrix to obtain the weights of each factor. The process of calculating weights using the AHP method is as follows: Figure 4 As shown.

[0086] 5.1 Establish a hierarchical structure model.

[0087] 5.2 Pairwise comparisons among factors influencing errors and construction of the judgment matrix.

[0088] Once the hierarchical structure is established, judgments need to be made based on the relative importance of each indicator in each level. These judgments are expressed numerically using appropriate scales and written in the form of a judgment matrix. Generally speaking, the judgment matrix should be independently provided by an expert familiar with the problem. Table 8 shows commonly used 1-9 scale methods.

[0089] Table 8. Matrix Scale and Its Meaning

[0090]

[0091] Note: ={2,4,6,8,1 / 2,1 / 4,1 / 6,1 / 8} indicates an importance level between ={1,3,5,7,9,1 / 3,1 / 5,1 / 7,1 / 9}, such as =2 indicates that element i is between "equally important" and "slightly important" compared to element j. These numbers are determined based on people's intuition and judgment in conducting qualitative analysis.

[0092] For n factors influencing errors, a judgment matrix can be obtained through pairwise comparisons. .

[0093]

[0094] in Let i and j represent the importance of error-influencing factors relative to the target. The judgment matrix C has the following properties: 1) 2) 3) 4) C is a matrix with positive and negative sides. If for any i, j, k, we have .

[0095] 4.3 Calculate the weights of each judgment matrix and perform a consistency check. 1) Calculate the geometric mean of all elements in each row of the judgment matrix. :

[0096] 2) Normalization and calculation :

[0097] but The desired eigenvector is, i.e. The desired subjective weight.

[0098] 3) Calculate the largest eigenvalue of the judgment matrix. :

[0099] In the above formula, For vectors The i-th element.

[0100] 4) Calculate the consistency index (CI) and perform a consistency test. The formula is as follows:

[0101] In the above formula, n is the order of the judgment matrix. The random consistency index RI can be found from Table 9, and the ratio CI / RI can be calculated. When CI / RI < 0.1, the consistency of the judgment matrix meets the requirements. Otherwise, the judgment is re-performed, and a new judgment matrix is ​​written.

[0102] Table 9 RI Value Table

[0103]

[0104] When determining the weights, the assignment of importance values ​​among the various indicators by experts is crucial. In this example, ten experts from different fields, including universities, research institutions, and water conservancy authorities, specializing in hydrology, hydraulic structures, geotechnical structures, and management, were selected to score the importance of each indicator. The weighted average of the indicator weights obtained after each expert's score was calculated to obtain the final weight values ​​of the primary and secondary evaluation indicators, as shown in Table 10.

[0105] Table 10 Subjective Weights of Risk Impact Factors at Each Level

[0106] Step Six: Calculate the objective weights between indicators using a BP neural network. The data source for this step is the dam failure accident input table obtained from the dam failure database in Step Four, Section 4.1, and the set control group, i.e., the data in Tables 3 and 4.

[0107] The basic idea of ​​S6.1 BP neural networks is: weighting, summation, and transfer, such as... Figure 5 As shown. Input layer input. x 1 , x 2 , x 3 , ..., x n The output layer provides index scores for each dam failure case and control group in the historical dam failure database; y 1 , y 2 , y 3 , ..., y n , where is the risk value for each case and control group in the historical dam failure database; if a dam fails, it is set to 1.

[0108] In this patent, the BP neural network is configured with three layers. ω g1 , ω g2 , ω g3 , ..., ω gn For the first g The connection weight coefficients between each neuron and the neurons in the previous layer. g =1,2,3; b g For threshold; f(·) Let be a transfer function. Then the _____ g The net input value of a neuron can be represented as:

[0109] in, X =[ x 1 ,x 2 , … ,x i … ,x n ] T , W =[ ω g1 ,ω g2 ,…,ω gi ,…,ω gn ].

[0110] If let x 0 =1,ω g0 = b g ,but X and W g It can be changed to:

[0111]

[0112] g The net input of a node can be expressed by the following formula:

[0113] Then the first g The output of each neuron y g for:

[0114] because f(·) It is a monotonically increasing function with an upper limit, so the signal transmitted by the neural network also has an upper limit.

[0115] S6.2 Use the scoring data for each influencing factor in Tables 3 and 4. As the input layer of a BP neural network X =[ x 1 , x 2 , … ,x i … ,x n ] T In actual dam failure accidents, the output layer is set to 1. The output layer of the control group is determined according to the dam risk level in its risk assessment report, i.e., by the qualitative-quantitative probability conversion in Table 2; for lower risk levels, the corresponding smaller failure probability value is used. The output fitted neural network hidden layer data, i.e., the connection weight coefficients of each neuron, is used. The weight transmission result of the multi-layer neural network connecting each indicator factor is the weight of each indicator factor.

[0116] In the formula, g =1,2,3, represents the number of hidden neural network layers.

[0117] S6.3 After multiple inputs, several different results are obtained. Obviously unreasonable results are removed and compared. The results are then normalized to obtain the objective weights of the influencing indicators. See Table 11.

[0118] Table 11 Objective Weights of Risk Impact Factors at Each Level

[0119]

[0120] Step 7: Use empirical factors to weight the subjective and objective weights.

[0121] S7.1. Let the weights obtained by the analytic hierarchy process be subjective weights. In step six, the weights obtained by fitting the BP neural network are objective weights. Then, the comprehensive weighting using the combination of empirical factors is: In the formula, Subjective weight and objective weight Empirical factors at the time of combination The weighting is determined by the evaluator and reflects the evaluator's preference for subjective and objective weighting. The larger the value, the more the evaluator values ​​subjective weight, i.e., the expert evaluation in this invention; The smaller the value, the more the evaluator values ​​objective evaluation.

[0122] S7.2 Considering the randomness of dam failure, in this embodiment, The weight of the weighted combination of influencing factor indicators was calculated by selecting 0.6, as shown in Table 12.

[0123] Table 12 Objective Weights of Risk Impact Factors at Each Level

[0124] Step 8: Establish a Bayesian network-based risk model.

[0125] S8.1 A Bayesian network is a directed acyclic graph, consisting of nodes representing variables and directed edges connecting these nodes. A Bayesian network comprises two parts: a directed acyclic graph with N nodes, where the nodes represent random variables and the directed edges between nodes represent the relationships between them; and a conditional probability table associated with each node, representing the relationship between the node and its parent node—the conditional probability relationship.

[0126] To build a causal model, the first step is to identify the variable nodes and their interpretations, and then establish the causal relationships between these nodes. Consider the following principles: 1) Human behavior is directly influenced by one's own state, including sense of responsibility, experience, and professional skills. These are called internal influencing factors. Simultaneously, it is also influenced by factors such as management systems and the on-site environment. These are called external influencing factors. Therefore, the established causal model is a hierarchical model, possessing good characteristics in terms of computational feasibility and application.

[0127] 2) Reduce network complexity during modeling. Determining the conditional probability distribution of each node variable involves a huge amount of computation. When the number of parent nodes of a certain node variable is k, at least 2^k parameters need to be determined. k Therefore, when the value of k exceeds 10, it becomes difficult to handle in practical applications, and it is necessary to make full use of conditional independence to simplify the calculation.

[0128] In the Bayesian network of this invention, the node variables originate from the influencing factors in the indicator system. Based on the cluster analysis of the influencing factors in the previous steps, the cluster points obtained from the clustering are introduced into the Bayesian network as intermediate child nodes to place related factors in the same group. A hierarchical approach is used for modeling, with each influencing factor connected to the cluster points layer by layer. Each layer is only affected by its direct parent layer, reducing the number of parent nodes in the network, simplifying calculations, and maintaining the reliability of the model. After hierarchical modeling, the influence relationships of the indicator factors are clarified and simplified. The resulting hierarchical causal model is as follows: Figure 6 As shown, such a hierarchical causal model is easy to adjust.

[0129] S8.2 Based on Table 12, the results of the weighted combination of subjective and objective weights from step seven, further determine the model parameters, that is, determine the conditional distribution of each node in the Bayesian network. The following rules are adopted for this: 1) Each node has a value of 2 (although it uses 2 values, the probability of being connected to it can be considered a measure of "degree," thus more accurately reflecting the actual situation). For example... Figure 6 The "Failure" node has two possible values: success and failure. The "Management" node has two possible values: good and bad. The "Supervision Intensity" node has two possible values: high and low, and so on. For consistency, these binary states are collectively referred to as "yes / no".

[0130] 2) If a node Y in the model has n parent nodes, represented as follows: Based on event data from the application context and combined with opinions from domain experts, each root node is assigned a weight ranging from 0 to 1, denoted as . , which represents the degree of influence of the root node on node Y. The larger the value, the greater the degree of influence.

[0131] when k nodes in When the value of node Y is "yes" and the values ​​of all other nodes are "no", the probability that node Y is "yes" is: .

[0132] The above are just general rules for node values. More specific rules can be developed based on actual application scenarios. For example, the node "cluster point C1" (represented by Y) is influenced by three parent nodes: "supervision intensity," "daily management," and "situation judgment" (represented by X1, X2, and X3, respectively). According to step seven, the weights of these three parent nodes are 0.2285, 0.3291, and 0.1907, respectively. Following the previous rules, the conditional probability distribution of the "individual" depending on the three parent nodes is shown in Table 13.

[0133] Table 13 conditional probability distribution

[0134]

[0135] Step 9: Combine the obtained Bayesian network with the actual reservoir dam project to establish evaluation standards for each indicator, so as to achieve the qualitative and quantitative conversion of the reliability of each indicator.

[0136] S9.1 Since the above rules use binary values ​​for model nodes, in order to more accurately reflect the actual situation, the binary state is extended to a multi-valued state. That is, the state probability of each parent node is qualitatively divided into four levels: very good, good, average, and poor. The quantitative values ​​corresponding to each level are shown in Table 14, denoted as... f i (i=l,…,n).

[0137] The probability that node Y will make a mistake is:

[0138] Based on dam failure case records and expert opinions in the historical dam failure database, and considering existing qualitative and quantitative conversion relationships, a qualitative description and probability correspondence table for the parent node is set up, as shown in Table 14.

[0139] Table 14 Qualitative Description and Probability Correspondence of Parent Nodes

[0140]

[0141] S9.2 To accurately convert between qualitative and quantitative indicators, evaluation criteria are particularly important. Only with criteria can they be divided into four levels and their qualitative descriptions converted into corresponding error probabilities. Table 15 gives the evaluation criteria for twelve secondary indicators (i.e., parent nodes) and formulates the judgment basis for each level in detail based on the evaluation indicators (see Tables 16 to 27).

[0142] Table 15 Evaluation Criteria for Secondary Indicator Level Classification

[0143]

[0144] The following tables in S9.3 provide evaluation criteria for twelve secondary indicators (i.e., parent nodes) in Bayesian networks, and detail the judgment basis for each level based on the evaluation indicators (see Tables 16-27). They are mainly for different objects and are based on historical data, reservoir dam safety management, and long-term operation. They can be directly applied to engineering practice to evaluate the management level of reservoirs.

[0145] Table 16 Estimation of the Probability of Human Behavioral Factor Errors Due to Supervision Intensity

[0146] Table 17 Probability Estimation Table of Risk Awareness Leading to Behavioral Factors

[0147] Table 18. Estimation of the probability of human behavioral factors leading to errors in situational judgment.

[0148] Table 19 Estimation of Error Probability Due to Human Behavioral Factors in Routine Management

[0149] Table 20 Estimation of the Probability of Errors Due to Human Behavioral Factors Caused by a Sense of Responsibility

[0150] Table 21 Estimation of the Probability of Errors Due to Experience in Human Behavioral Factors

[0151] Table 22 Estimation of the Probability of Errors Due to Human Behavioral Factors Caused by Professional Skills

[0152] Table 23 Estimation of the Probability of Human Behavioral Factor Errors Due to Organizational Configuration Qualitative evaluation According to Human behavior factor failure probability Very good The reservoir has a perfect management organization (bureau, office, station); the management organization contains the reservoir dispatching technical department and the safety management technical department; the above-mentioned department responsibilities are very clear, the division of labor is very reasonable, and experienced professional and technical personnel are provided; the number of professional and technical personnel in each department fully meets the actual management needs; very perfect management facilities are provided, including water and rainfall situation forecasting system, dam safety monitoring system. 10 -6 ~ 0.01 <!-- 21 -->]]> Good The reservoir has a management organization (bureau, office, station); the management organization contains the reservoir dispatching technical department and the safety management technical department; the above-mentioned department responsibilities are clear, the division of labor is reasonable, and experienced professional and technical personnel are provided; the number of professional and technical personnel in each department meets the actual management needs; perfect management facilities are provided, including water and rainfall situation forecasting system, dam safety monitoring system. 0.01～0.1 General The management organization (bureau, office, station) of the reservoir is not perfect; the management organization contains the reservoir dispatching group and the safety management group; the above-mentioned department responsibilities are basically clear, the division of labor is basically reasonable, and professional and technical personnel are provided; the number of professional and technical personnel in each group basically meets the management needs; the management facilities are general, the water and rainfall situation forecasting system and the dam safety monitoring system are old and aging, and the reliability is not high. 0.1～0.8 Poor The reservoir has no special management organization and full-time management personnel, only part-time administrators; the administrators are non-technical personnel, who do not understand flood control dispatching and safety monitoring technology; the number of part-time management personnel does not meet the management needs, there are no management facilities, and there is no water and rainfall situation forecasting system and dam safety monitoring system. 0.8～1.0 Table 24 Estimation of the Probability of Errors Due to Human Behavioral Factors Caused by Institutional Development

[0153] Qualitative evaluation According to Human behavior factor failure probability Very good The perfect reservoir dam safety administrative leadership responsibility system and the responsibility investigation system have been established; there are strict reservoir management rules and regulations and operation and maintenance manuals; there are strict observation, patrol systems and pre-flood and post-flood inspection systems; there are perfect reservoir flood control and flood dispatching schemes and emergency plans; there are perfect normal maintenance systems; there are normal channels for operation and maintenance funds; the perfect dam safety annual report system has been established; the dam safety identification and reinforcement can be considered according to the regulations. 10 -6 ~ 0.01 Good The relatively perfect reservoir dam safety administrative leadership responsibility system and the responsibility investigation system have been established; there are relatively strict reservoir management rules and regulations and operation and maintenance manuals; there are relatively strict observation, patrol systems and pre-flood and post-flood inspection systems; there are relatively perfect reservoir flood control and flood dispatching schemes and emergency plans; there are relatively perfect normal maintenance systems; there are sources of operation and maintenance funds; the relatively perfect dam safety annual report system has been established; the dam safety identification and reinforcement can be considered according to the regulations. 0.01～0.1 General A system of administrative leadership responsibility and accountability for reservoir dam safety has been established; there are reservoir management regulations and operation and maintenance manuals; there are observation and inspection systems, as well as pre-flood and post-flood inspection systems; there are reservoir flood control and flood season plans, flood dispatch plans, and emergency plans; there is a routine maintenance system; there are potential sources of funds for operation and maintenance; an annual dam safety reporting system has been established; and dam safety assessments and reinforcements are generally considered in accordance with regulations. 0.1～0.8 Difference The reservoir dam safety administrative leadership responsibility system and accountability system have not been established; there are no reservoir management regulations and operation and maintenance manuals; there are no observation and inspection systems, as well as pre-flood and post-flood inspection systems; there are no reservoir flood control and flood season plans, flood dispatching plans, and emergency plans; there is no routine maintenance system; there are no potential sources of funds for operation and maintenance; there is no annual dam safety reporting system; and dam safety assessments and reinforcement are not considered as required. No regulations or systems have been formulated. 0.8～1.0 Table 25 Estimation of the Probability of Human Behavioral Factors Leading to Errors in Emergency Response Plans

[0154]

[0155] Table 26 Estimation of the probability of human behavioral factors causing errors due to external environment Table 27 Estimation of the probability of human behavioral factors causing errors due to internal environment

[0156] Table 27 Estimation of the Probability of Errors Due to Internal Environmental Factors in Human Behavior

[0157] Step 10: Considering the actual engineering situation, apply the Bayesian network risk model obtained in the previous steps to the engineering project, determine the calculation method and calculation results, and use them as indicators to evaluate the reservoir management level.

[0158] Hongze Lake is located in northwestern Jiangsu Province, on the western side of the central part of the North Jiangsu Plain, within the cities of Huai'an and Suqian. It is a giant reservoir that regulates water flow from the upper and middle reaches of the Huai River, with a catchment area of ​​158,000 km². 2 Xingli Reservoir has a capacity of approximately 2.6 billion cubic meters. 3 Total reservoir capacity: 12.3 billion cubic meters 3 It serves multiple functions including flood control, irrigation, navigation, and aquaculture. The lakebed elevation is approximately 10.0–11.0 m, 4–8 m higher than the downstream area (eastern plain). Hongze Lake is designed for a 300-year flood event and has a check flood level of 2000 years. The near-term flood control standard is 100-year, corresponding to a design water level of 16.0 m and a check water level of 17.0 m.

[0159] The main hydraulic structures of Hongze Lake Reservoir are three: Hongze Lake Dam, Sanhe Sluice Gate, and Erhe Sluice Gate.

[0160] The Hongze Lake Dam stretches 67.25 km from Laobatou in Matou Town, Huaiyin County in the north to Zhangzhuang Heights in Xuyi County in the south, with 60.1 km of it featuring an old stone masonry wall. The Hongze Lake Dam is a Class I hydraulic structure. Its design water level is 16.0 m, check water level is 17.0 m, dam crest elevation is 19.0 m–19.5 m, lake bottom average elevation is 10.5 m, and the wave-breaking forest platform elevation is 14.5 m. The dam's soil is primarily silty clay, accounting for 60%–70%, followed by heavy silty loam and layers of clay and silt. The area above 10.0 m–10.5 m consists of artificial fill.

[0161] The Sanhe Sluice Gate, located in Hongze County, Jiangsu Province, at the southeast corner of Hongze Lake, is a key waterway controlling the flow of the Huai River into the Yangtze River and a major engineering project in the Huai River basin. It was a large-scale sluice gate designed and constructed independently by my country in the early days of the People's Republic. Construction began in October 1952, and it was completed and began releasing water in July 1953. The gate body is a reinforced concrete structure with 63 openings, each with a net width of 10 meters, for a total width of 697.75 meters. The bottom slab elevation is 7.5 meters, and the width is 18 meters, consisting of 21 bottom slabs. The net height of each opening is 6.2 meters. The Sanhe Sluice Gate was designed for a Hongze Lake water level of 16 meters and checked at 17 meters, with an original design flow rate of 8000 m³ / h.3 / s, the reinforced Sanhe Sluice Gate's designed flood discharge capacity has increased to 12000m. 3 / s, belonging to the large Class I sluice gate.

[0162] The Erhe Sluice Gate is located about 7 km northeast of Hongze County, Jiangsu Province. It is the main entrance for the waterway to the sea and the main outlet for the diversion of the Huai River and the Yi River, and plays a comprehensive role in flood discharge, irrigation, and the diversion of the Yi River to the Huai River. Construction of the project started in November 1957 and was completed in August 1958. The Erhe Sluice Gate is a Class I large (1) hydraulic structure with 35 openings, each with a net width of 10m and a total width of 402m. It is a reinforced concrete breast wall structure. The bottom slab elevation is 8.0m, the top elevation of the gate pier is 19.5m, and the working bridge deck elevation is 28.25m. It uses arc-shaped steel gates for water blocking and is controlled by a winch-type hoist. The design standard of the Erhe Sluice Gate is: the design flow rate for the diversion of the Huai River and the Yi River is 3000 m³ / h. 3 / s, the verification flow rate is 9000m³ / s. 3 / s.

[0163] During the flood season, the Hongze Lake dike has repeatedly breached due to the significant difference between the upstream water flow and the downstream flood discharge capacity, causing enormous disasters downstream. Since 2000, the dike has repeatedly experienced concentrated seepage in the dike foundation and large-scale tidal surges and seepage behind the dike. Although emergency measures have been taken, limited funds have prevented a comprehensive and thorough solution. In addition, many of the structures along the dike are old and dilapidated culverts (sluice gates). In the event of a major flood, the Hongze Lake dike will be unable to guarantee flood control safety.

[0164] S10.1 further refines the Bayesian network risk model based on Bayesian networks for engineering applications.

[0165] The Bayesian network risk model obtained in step eight is a general model for general situations. When applied to engineering practice, the node variables can be supplemented and modified according to the specific context to make it more suitable for specific engineering cases.

[0166] In the field of dam engineering, errors often occur in early warning, forecasting, and emergency intervention stages. To facilitate engineering applications, the following two sub-models for dam risk analysis are established for each of these stages: (1) Bayesian network risk model in early warning, forecasting and prediction. Early warning, forecasting and prediction of the comprehensive safety performance of dams should include the following: 1) A certain number of hydrological stations (including water level and rainfall observation stations) should be set up according to the needs of hydrological forecasting and reservoir operation. The selection of stations should take into account representativeness, controllability, and transportation and communication conditions. In addition, the reservoir management unit should collect hydrological and meteorological information as well as downstream river water level information in a timely manner, and organize it as necessary as needed.

[0167] 2) A flood forecasting plan must be developed and implemented after approval by the reservoir management department. The flood forecasting plan should be re-evaluated at the end of each flood season or after a period of use, and supplemented, revised or updated as necessary.

[0168] 3) The reservoir management unit shall conduct routine and special inspections, keep detailed records, and report any abnormalities in a timely manner.

[0169] 4) Based on the dam instrument monitoring data, conduct observation data analysis to analyze and evaluate dam operation. Also, inspect, repair, and maintain the dam monitoring facilities.

[0170] 5) An emergency plan has been developed and an operational early warning system is in place.

[0171] The above analysis shows that in the early warning, forecasting, and prediction stage, risk prevention and control are mainly related to management, individuals, and systems, especially closely related to the "daily management" indicator at the higher level of "management." Therefore, three tertiary indicators were established for "daily management": inspection and patrol, operation, maintenance and monitoring, and analysis of observational data. (See details...) Figure 7 In the specific calculation process, the weight of each of the three tertiary indicators is 1 / 3. The evaluation criteria for the classification of human error can be found in Table 28, and the specific criteria will be determined by experts based on experience. Table 28 Evaluation Criteria for Classification of Level 3 Indicators

[0172]

[0173] (2) Bayesian network risk model in disaster relief intervention Emergency intervention refers to the urgent measures taken to address engineering hazards arising from sudden safety incidents at reservoir dams, aiming to control the situation and prevent dam failure. The success of emergency intervention depends primarily on the following factors: 1) Firstly, it is related to the environment. Dangerous situations usually occur alongside disasters, especially when a major earthquake or a flood exceeding the standard occurs. If the dam itself has engineering defects, the probability of successful manual rescue is low. 2) Secondly, it is related to the reservoir management. After the incident occurs, the management needs to examine and assess the phenomena and characteristics, nature and extent of the incident, as well as its possible development and harm, in order to formulate reasonable and effective emergency response measures. 3) Finally, it also relates to the individual. Reservoir site managers should understand and master some commonly used and advanced emergency repair methods suitable for the characteristics of the project, prepare materials and equipment, and obtain support from technical experts so that they can be used flexibly in emergency situations.

[0174] It is evident that the Bayesian network risk model in disaster relief intervention should primarily consider three levels of indicators: management, individuals, and the environment. For a detailed model, see [link to model details]. Figure 8 .

[0175] S10.2 Conduct risk analysis on engineering examples Based on historical flood disaster data of Hongze Lake, current engineering status, organizational structure of the management office, system construction, and the descriptions of inspection, dam safety monitoring, and emergency plans in the 2011 dam safety report, the 12 secondary indicators were classified and the probability of failure was determined according to Tables 16-27. The results are shown in Table 29.

[0176] Table 29 Classification of Secondary Indicators and Probability of Failure

[0177]

[0178] 1) Reliability analysis in early warning and forecasting This sub-model has three levels of indicators, so the level and error probability of the three levels of indicators should be determined first, as shown in Table 30.

[0179] Table 30 Classification of Level 3 Indicators and Probability of Failure

[0180] Calculate the probability of an error occurring in the "Daily Management" node using the formula in step eight: (1 / 3 × 0.05 + 1 / 3 × 0.05 + 1 / 3 × 0.20) / (1 / 3 + 1 / 3 + 1 / 3) = 0.10 The remaining nine secondary indicators all adopted the results determined in Table 29. The probabilities of each child node were calculated sequentially by the Bayesian network, and the final calculated risk probability was 0.1384.

[0181] 2) Human Factor Reliability Analysis in Disaster Relief Intervention Based on the error probability values ​​of the twelve secondary indicators determined in Table 29, the failure probability of each sub-node is calculated using the formula in step eight and the Bayesian network. The error probability in the emergency intervention is 0.1343.

[0182] S10.3 Considering the parent node description and error probability estimation table, and combining with the actual engineering situation, establish a risk assessment index system for human behavior factors of reservoir dams.

[0183] From the error probability estimation tables for human behavioral factors in Tables 16-27, the error probability of each parent node can be obtained. The probability of each child node can be calculated using the formula in step eight. The calculation continues to obtain the values ​​of the top-level indicators and the reliability of management personnel.

[0184] Based on the qualitative and quantitative probability tables and the conditional probability relationship between parent and child nodes, a risk assessment index system for human behavior errors in reservoir dams is established, as shown in Table 31.

[0185] Table 31 Risk Assessment Indicators for Human Behavior Errors in Reservoir Dams

[0186]

[0187] As shown in the table above, similar to the evaluation criteria for secondary indicators, each primary indicator and the top-level indicator are divided into four levels: "Very Good," "Good," "Average," and "Poor." The error probability of each node is calculated using the Bayesian network and Bayesian inference method in step eight. When the state of each node is "Poor," it indicates that this management indicator is unqualified. When the indicator state is "Average," a comprehensive analysis is required. It is recommended to take the critical result as 0.45. When the reliability calculation result is greater than 0.45, that is, the reliability is "General Deviation," the indicator is considered relatively unreliable, and the reservoir management level needs to be improved.

[0188] Obviously, those skilled in the art should understand that the steps of the Bayesian network risk model construction method for reservoir dam risk assessment described in the above embodiments of the present invention can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, which can then be stored in a storage device for execution by the computing device. Furthermore, in some cases, the steps shown or described can be performed in a different order than presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of the present invention are not limited to any particular hardware and software combination.

Claims

1. A method for constructing a Bayesian network risk model for reservoir dam risk assessment, characterized in that, The construction method includes the following steps: S1, Based on historical data on dam failure accidents, construct a historical dam failure database; S2, classify the causes of errors to obtain the root causes of risks in dam failure accidents, and conduct importance analysis; S3, analyze the factors affecting dam risks and establish an index system for error impact factors; S4. Factor correlation analysis was performed using the approximation matrix and Pearson correlation coefficient to obtain the cluster analysis phylogenetic diagram and the relationship between the influencing factors, which served as the basis for the connection of the Bayesian network directed acyclic graph. S5, based on expert scoring, uses the analytic hierarchy process (AHP) to determine the subjective weights of each level of indicators; S6 uses a BP neural network to calculate the objective weights between indicators; S7 uses empirical factors to weight the subjective and objective weights; S8. A directed acyclic graph of the Bayesian network is obtained from the clustering and correlation analysis results. The conditional probabilities between nodes are obtained based on expert evaluation scores and BP neural network results. A risk model based on the Bayesian network is then established. S9 combines the error impact factor index system with the actual situation of reservoir dam engineering, establishes evaluation standards for each index, and realizes the qualitative and quantitative conversion of the reliability of each index. S10: Perform risk assessment calculations on the obtained Bayesian network risk model, and use the results as an indicator to evaluate the risk of the reservoir dam.

2. The method for constructing a Bayesian network risk model for reservoir dam risk assessment according to claim 1, characterized in that, The error impact factor index system is divided into two aspects: internal factors and external factors. Internal factors include two primary indicators: management and individuals. External factors include two primary indicators: system and environment. Management is further divided into four secondary indicators: supervision intensity, risk awareness, situation judgment, and daily management. Individuals are divided into three secondary indicators: sense of responsibility, experience, and professional skills. System is divided into three secondary indicators: organizational configuration, system construction, and emergency plan. Environment is divided into two secondary indicators: external environment and internal environment.

3. The method for constructing a Bayesian network risk model for reservoir dam risk assessment according to claim 1, characterized in that, The specific implementation process of S4 is as follows: S4.1, let the impact factor score of the j-th error in the i-th reservoir be... Then, establish the evaluation characteristic matrix A: S4.2 introduces Pearson correlation coefficient for cluster analysis to measure the correlation among error-influencing factors; error-influencing factors and The Pearson correlation coefficient between them is calculated as follows: In the formula, m is the number of variables, that is, each error influencing factor has m reservoir scores; and Error influencing factors and The average value, yes and covariance; yes Standard deviation; yes Standard deviation; yes and The Pearson correlation coefficient; the Pearson correlation coefficient is between [-1, 1], when When, it means and A positive correlation; when When, it means and A negative correlation; when When, it means and A perfect positive correlation; express and A perfect negative correlation; when Time indicates and Irrelevant.

4. The method for constructing a Bayesian network risk model for reservoir dam risk assessment according to claim 1, characterized in that, The specific implementation process of S5 is as follows: S5.1, Establish a hierarchical structure model; S5.2, pairwise comparison between indicators and construction of the judgment matrix, where the indicators refer to error impact factors; after the hierarchical structure is established, judgments need to be made based on the relative importance of each indicator in each level to obtain the judgment matrix; For n error influencing factors, a judgment matrix is ​​obtained through pairwise comparisons. ; in The judgment matrix C represents the importance of error impact factors i and j relative to the target; it has the following properties: 1） 2） 3） 4) C is a matrix with positive and negative sides. If for any i, j, k, we have ; S5.3 Calculate the weights of each judgment matrix and perform a consistency check.

5. The method for constructing a Bayesian network risk model for reservoir dam risk assessment according to claim 1, characterized in that, The specific implementation process of S6 is as follows: S6.1, use a BP neural network to calculate the objective weights between indicators. By inputting dam failure data from the historical dam failure database into the BP neural network, the indicator weights are determined through multiple fittings and used as the objective weight values. S6.2, Input layer of BP neural network x 1 , x 2 , x 3 , ..., x n The output layer outputs the index scores for each dam failure case and control group in the historical dam failure database. y 1 , y 2 , y 3 , ..., y n , which represents the risk value of each case and control group in the historical dam failure database; if a dam fails, it is set to 1. S6.3, run the BP neural network, input several sets of data from the historical dam failure database, continuously fit, and obtain the connection weight coefficients of neurons in each layer. The weight transmission result of the multi-layer neural network connecting each indicator factor is the weight of each indicator factor. In the formula, g =1,2,3, where 1,2,3 represent the number of layers in the hidden neural network, and the objective weights are calculated. .

6. The method for constructing a Bayesian network risk model for reservoir dam risk assessment according to claim 1, characterized in that, The specific implementation process of S7 is as follows: Let the weights obtained by the analytic hierarchy process be subjective weights. The weights obtained by fitting a BP neural network are objective weights. Then, the comprehensive weighting using the combination of empirical factors is: In the formula, Subjective weight and objective weight Empirical factors at the time of combination The weighting is determined by the evaluator and reflects the evaluator's preference for subjective and objective weighting.

7. The method for constructing a Bayesian network risk model for reservoir dam risk assessment according to claim 1, characterized in that, The specific implementation process of S8 is as follows: Based on the error impact factor index system established in S3, and the cluster analysis phylogenetic diagram of the impact factor index obtained in S4 and the interrelationships between impact factors, combined with the analysis of human behavior mechanisms, the node association relationships in a Bayesian network directed acyclic graph are constructed; a hierarchical modeling method is adopted, with each layer only affected by its direct parent layer; related factors are placed in the same group; and the conditional distribution of each node in the Bayesian network is determined. 1) Each node has two values: yes or no; 2) If a node Y in the model has n parent nodes, represented as follows: ; Based on historical dam failure database records and combined with opinions from domain experts, each root node is assigned a weight ranging from 0 to 1, denoted as . This indicates the degree of influence of the root node on node Y; when k nodes in When the value of node Y is "yes" and the values ​​of all other nodes are "no", the probability that node Y is "yes" is: 。 8. A computer device, characterized in that: The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the Bayesian network risk model construction method for reservoir dam risk assessment as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program that performs the Bayesian network risk model construction method for reservoir dam risk assessment as described in any one of claims 1-7.

Images