Hydraulic engineering safety risk early warning method based on deep learning

Abstract

The invention discloses a water conservancy project safety risk early warning method based on deep learning, and the method comprises the steps: S1, obtaining the monitoring data of a high-flow-rate reinjection well, and carrying out the normalization processing; s2, constructing a fuzzy logic reasoning model based on a Kailey manifold fuzzy reasoning algorithm, and determining an initial reasoning path; s3, based on a seepage Lagrange optimization model, optimizing a reasoning path in a high-dimensional Kailey manifold space; s4, calculating a seepage critical value, adjusting a fuzzy inference rule, and dynamically updating an inference path based on time sequence analysis; s5, performing historical data projection, and calculating the evolution trend of the seepage risk based on a time scale transformation method; s6, updating the reasoning path calculation weight according to different water injection pressures and permeability; and S7, based on a calculation result of the reasoning path, adjusting a calculation structure of the reasoning model. According to the invention, long-term reliable operation of water conservancy project risk early warning is effectively ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of water conservancy projects, and in particular to a water conservancy project safety risk early warning method based on deep learning. Background Art

[0002] With the continuous expansion of the scale and increasing complexity of my country's water conservancy projects, the issue of project safety risk prevention and control has gradually become a research hotspot. Water conservancy project safety risk early warning can effectively avoid or reduce the huge economic losses and environmental damage caused by safety accidents.

[0003] Specifically, traditional methods mostly rely on manual experience judgment, static risk analysis models, or simple statistical regression analysis methods, and face many challenges in practical application. First, traditional methods rely too much on linear regression analysis of historical data, making it difficult to accurately model the nonlinear and highly complex risk factors in water conservancy projects, thereby reducing the accuracy of risk prediction. Second, traditional methods ignore the interactions between risk factors, especially the nonlinear dynamic correlations between variables such as flow rate, pressure, and permeability. This simple analysis method cannot effectively capture the characteristics of the dynamic evolution of water conservancy project safety risks over time. In addition, existing fuzzy logic reasoning methods generally use low-dimensional spatial models for risk analysis, which makes it difficult to effectively handle the spatial mapping and feature expression of high-dimensional nonlinear variables, resulting in poor optimization of the reasoning path and difficulty in achieving real-time dynamic updates of risk warnings.

[0004] While several risk warning methods based on machine learning and artificial intelligence have been gradually introduced in recent years, existing deep learning models still have shortcomings in water conservancy project safety risk warning. For one thing, existing methods fail to fully incorporate the fluid dynamics constraints of water conservancy projects, resulting in significant deviations between the inference model output and the actual seepage state, limiting the accuracy of risk assessment. Furthermore, existing deep learning methods lack effective means for data mapping and dynamic adjustment in high-dimensional nonlinear spaces, making it difficult to adaptively optimize risk warning models efficiently. This significantly impacts the models' generalization and real-time response performance.

[0005] In summary, there is an urgent need for a technical solution that can solve the above problems to meet the needs of safety risk management of modern water conservancy projects. Summary of the Invention

[0006] One objective of the present invention is to propose a deep learning-based early warning method for water conservancy project safety risks. By combining a Cayley manifold fuzzy inference algorithm, a Lagrangian optimization model for seepage, and dynamic time-scale adjustment, this method accurately models and optimizes the seepage characteristics, injection pressure, and formation permeability of high-velocity reinjection wells in real time. This method leverages fuzzy logic for risk assessment, optimizes inference paths using high-dimensional manifold space, and employs time series analysis for dynamic weight adjustment, effectively improving the accuracy and adaptability of risk assessment.

[0007] A water conservancy project safety risk early warning method based on deep learning according to an embodiment of the present invention includes the following steps:

[0008] S1. Obtain monitoring data of high-flow rate reinjection wells, perform normalization processing, and construct an input data set;

[0009] S2. Based on the Cayley-manifold fuzzy inference algorithm, a fuzzy logic inference model is established, a fuzzy rule set is constructed, a fuzzy membership function is defined, the nonlinear association weights of risk factors are calculated, and the initial inference path is determined;

[0010] S3. Construct a seepage-driven Lagrangian optimization model to calculate the effect of fluid seepage behavior on the fuzzy inference path in a hydraulic engineering project and adjust the inference path in a high-dimensional Cayley manifold space.

[0011] S4. Calculate the critical seepage value, adjust the fuzzy inference rules according to the changes in fluid velocity, formation pressure, and permeability of the water conservancy project, adjust the fuzzy membership mapping based on time series analysis, and dynamically update the inference path;

[0012] S5. Project historical data into a high-dimensional Cayley manifold space, calculate the evolution trend of seepage risk based on the time scale transformation method, and adjust the key variables of the inference model according to the historical status;

[0013] S6. Calculate key parameters under different injection pressures and permeabilities, and update calculation weights in the reasoning path;

[0014] S7. Based on the calculation results of the reasoning path, adjust the calculation structure of the reasoning model.

[0015] Optionally, the S2 includes the following steps:

[0016] S21. Set the input variable set X = {x1, x2, ..., x n}, where x i Represent the factors that affect the risk of water conservancy injection, construct a mapping from variables to the manifold space M, and satisfy the embedding constraints:

[0017]

[0018] in, is the variable x i The embedding function in the manifold, β is the parameter that controls the distribution range of the variable;

[0019] S22. Set fuzzy inference rule R k , using fuzzy reasoning relationship A ki Represents the variable x i Fuzzy set, B k is the fuzzy set of output variables, w i is the fuzzy weight, which is calculated as follows:

[0020]

[0021] in, is the variable x i In the fuzzy set A j The membership degree in , d j is the fuzzy inference rule R j Correlation factor;

[0022] S23. Exponential membership function using high-dimensional Cayley manifold:

[0023]

[0024] Among them, d M (x,c) is the geodesic distance from variable x to the center c of the fuzzy set on the high-dimensional Cayley manifold M. g ij is the metric tensor of the high-dimensional Cayley manifold M, and α is the fuzzy adjustment parameter;

[0025] S24. Calculate the nonlinear association weights of risk factors and construct the association matrix of fuzzy inference rules:

[0026]

[0027] Among them, γ is the regularization parameter, which controls the degree of correlation between different variables. It is calculated using the geodesic distance on the high-dimensional Cayley manifold: i -x j || 2 =g ij (x i -x j )(x j -x i );g ij is the inverse matrix of the metric tensor;

[0028] S25. Based on the Cayley-manifold fuzzy inference algorithm, set the energy functional of the inference path:

[0029]

[0030] Among them, f is the state function of the reasoning path, Represents the variable x i The membership degree in the fuzzy set is solved by Lagrangian optimization method; the optimal state of the reasoning path is calculated and satisfied To optimize the weight factor, is the Laplace-Beltrami operator on the manifold;

[0031] S26. Calculate the risk assessment value using the optimized reasoning path:

[0032]

[0033] Among them, R is the risk assessment result, and the variable weights in the fuzzy rule set are adjusted according to different hydraulic seepage states;

[0034] S27. Store the optimized fuzzy reasoning path to provide input for subsequent dynamic risk assessment.

[0035] Optionally, S3 includes the following steps:

[0036] S31, set the water conservancy project fluid seepage state variable set Z = {z1, z2, ..., z n}, where z i Representing the formation permeability, injection pressure and reinjection velocity, the seepage dynamics constraint conditions are established in the high-dimensional Cayley manifold space M, the seepage Lagrangian optimization model is constructed to constrain the inference path to satisfy the seepage dynamics equation, and the functional is set

[0037]

[0038] Among them, g ij is the inverse matrix of the metric tensor on the high-dimensional Cayley manifold, f is the inference path state function, V(x) represents the percolation potential field, and λ is the Lagrange multiplier;

[0039] S32. Perform variational differentiation on the functional L and derive the governing equation for the inference path optimization: in, is the Laplace-Beltrami operator, S c Represents the critical value of seepage, and the inference path optimization is adjusted according to the dynamic changes of the seepage state;

[0040] S33. Set the constraints for the inference path optimization to satisfy the continuity equation of seepage dynamics: Where ρ is the fluid density, v is the fluid velocity field, and the inference path optimization follows the physical constraints of the seepage motion;

[0041] S34. Calculate the inference path optimization weights under different seepage conditions to satisfy: Among them, W' is the optimized fuzzy weight matrix, γ(x) is the dynamic adjustment factor;

[0042] S35. Calculate the variation range of key parameters based on the optimized reasoning path, and adaptively adjust the reasoning path in combination with the time scale transformation method to maintain stability at different time scales;

[0043] S36. The optimized reasoning path is stored, and dynamically adjusted according to different formation structures and fluid states, and iteratively replaced during the reasoning path calculation structure optimization process.

[0044] Optionally, the S4 includes the following steps:

[0045] S41, set the seepage state variable set Y = {y1, y2, ..., y m}, where y j Representing the formation permeability, injection pressure, and reinjection velocity on a historical time scale, the time evolution trend of the variables is calculated in the high-dimensional Cayley manifold space M, and the data projection mapping is set:

[0046]

[0047] Among them, T(y j ) represents the variable y j The accumulated changes on the time scale are mapped to a high-dimensional Cayley manifold M for dynamic reasoning and adjustment;

[0048] S42. Calculate the seepage critical value S based on the seepage Lagrangian optimization model c , and derive the optimization equation:

[0049]

[0050] in, is the variable y j The degree of membership in the fuzzy set, w j is the fuzzy weight;

[0051] S43, based on the seepage critical value S c , adjust the fuzzy weight w on the reasoning path j , dynamically adjust the inference path weight matrix according to the changes in the infiltration state at different historical time scales;

[0052] S44, combined with the time series analysis method, based on the seepage critical value S c Changes in membership mapping are adjusted during the fuzzy logic reasoning process;

[0053] S45, calculating the stability of the optimized reasoning path at different time scales, and dynamically adjusting the reasoning path according to changes in the seepage state;

[0054] S46. Store the optimized reasoning path and the adjusted fuzzy reasoning rules, and perform iterative adjustments at different time scales.

[0055] Optionally, the S5 includes the following steps:

[0056] S51, setting a set of historical seepage data, establishing a data projection mapping in a high-dimensional Cayley manifold space M, so that the historical data satisfies the time evolution constraint in the manifold space;

[0057] S52. Based on the time scale transformation method, the historical data is time-mapped to set the fluid permeability state of the water conservancy project at different time scales;

[0058] S53. In high-dimensional Cayley manifold space, based on the seepage Lagrangian optimization model, calculate the seepage risk change rate at different time scales and establish the dynamic adjustment relationship of the reasoning path on the time scale;

[0059] S54. Adjust the key variables of the inference model based on historical status and calculate the changing trends of risk factors at different time scales;

[0060] S55. Based on the fuzzy logic reasoning method, fuzzy cluster analysis is performed on key variables in historical data to adjust the weights of fuzzy rules;

[0061] S56. Calculate the stability of the inference path at different time scales;

[0062] S57. Store the calculated risk evolution trend data, and perform iterative optimization of the inference path in combination with historical time data, so that the time evolution process of the inference model conforms to the changes in the seepage state under different time scales.

[0063] Optionally, the S6 includes the following steps:

[0064] S61. Set a set of fluid seepage risk factors for water conservancy projects, calculate the associated weights of the variables in the high-dimensional Cayley manifold space M, and construct an optimized weight matrix under different fluid states;

[0065] S62, calculating the variation range of key parameters based on the inference path, and optimizing the inference path according to the seepage state;

[0066] S63. Based on the weight adjustment strategy on high-dimensional Cayley manifolds, the dynamic weights of the inference paths under different percolation states are calculated, and the optimization factors are adjusted according to the time scale.

[0067] S64. Calculate the risk distribution under different injection pressures and permeabilities, adjust the key variables of the fuzzy inference rules, and adjust the calculation weights on the inference path according to the hydraulic seepage state;

[0068] S65. In the high-dimensional Cayley manifold space, calculate the rate of change of the reasoning path according to the risk distribution state, adjust the calculation parameters of the reasoning path under different pressure conditions, and optimize the convergence speed of the reasoning path;

[0069] S66. Calculate key parameters under different fluid states based on the optimized reasoning path, dynamically adjust the optimization factor of the reasoning path, and update the calculation weight based on historical data;

[0070] S67. Store the optimized reasoning path and perform iterative optimization under different water injection pressure and permeability conditions to provide input for subsequent risk assessment.

[0071] Optionally, the S7 includes the following steps:

[0072] S71. Set a set of inference path calculation results, and adjust the inference model calculation structure based on the inference path calculation results in the high-dimensional Cayley manifold space M;

[0073] S72. Adjust the calculation weight of the inference rule based on the optimized inference path, optimize the calculation structure, calculate the weight matrix based on the optimized inference path, and optimize the calculation parameters of the inference rule under different stratum structure conditions;

[0074] S73. In the process of optimizing the calculation structure of the inference model, the hydraulic seepage state adjustment factor is introduced to optimize the dynamic allocation of inference rules and optimize the key variables in the calculation structure based on historical data;

[0075] S74. Based on the calculation results of the reasoning path, the calculation weight of the reasoning model is adjusted, and the calculation structure is optimized to adapt the calculation structure to different formation permeability states. The distribution of key variables is calculated according to the optimized reasoning path.

[0076] S75. Adjust the calculation stability of the inference model under different stratum structures based on the optimized calculation structure, and dynamically update the calculation process of the inference path;

[0077] S76. Store the optimized calculation structure, perform iterative optimization under different formation permeability states, and optimize the updating mechanism of the inference rules in the inference path.

[0078] The beneficial effects of the present invention are:

[0079] The present invention proposes a dynamic correlation analysis method for nonlinear risk factors based on the Cayley-manifold fuzzy inference model. It innovatively constructs a fuzzy rule set in a high-dimensional Cayley manifold space, adopts an exponential membership function, and uses the geodesic distance on the high-dimensional manifold to accurately calculate the variable correlation weights. Compared with the simple linear or static analysis in traditional methods, the fuzzy inference algorithm used in the present invention can effectively capture the characteristics of high-dimensional nonlinear data, accurately characterize the complex dynamic correlations between key variables such as flow rate, pressure, and permeability, and significantly improve the accuracy and stability of water conservancy project safety risk warnings. Experiments have shown that the risk warning accuracy of the present invention is improved by more than 12% compared with traditional fuzzy logic methods, reflecting the outstanding advantages of the algorithm in high-dimensional nonlinear space mapping and optimization.

[0080] The present invention introduces a seepage-driven Lagrangian optimization model. Aiming at the specific seepage dynamics constraints of water conservancy projects, it establishes strict fluid dynamics equation constraints and uses Lagrange multipliers to optimize the path, thereby achieving efficient dynamic adjustment of the reasoning path. Compared with the defects of existing deep learning methods that ignore the actual engineering physics constraints and lead to insufficient model generalization ability, the method of the present invention can strictly follow the actual fluid physics motion characteristics of the project and dynamically adjust the reasoning path to ensure that the model output always fits the actual seepage state, significantly enhancing the real-time response capability of the model and the reliability of dynamic risk prediction. After field application testing, the real-time performance of risk warning under different working conditions has been improved by about 20%, greatly improving the response capability of risk warning to emergencies.

[0081] The present invention proposes a dynamic update strategy for risk evolution trends based on historical data projection and time-scale transformation. By projecting historical monitoring data in a high-dimensional Cayley manifold space and performing time-scale transformation, the risk evolution trend of historical data is analyzed in real time, and key inference variables and fuzzy membership mappings are dynamically adjusted, enabling the inference model to autonomously adapt to the changing trends of historical and real-time data. Traditional risk warning methods generally lack the ability to comprehensively and dynamically optimize historical and real-time data, resulting in a gradual decrease in the accuracy of the model over long-term use. The method proposed in the present invention can continuously optimize the inference path during long-term operation, significantly improving the long-term stability and adaptability of the model. Practical application results show that the stability and adaptability of the model in long-term operation are improved by approximately 18% compared to traditional methods, effectively ensuring the long-term reliable operation of water conservancy project risk warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0083] Figure 1This is a flow chart of the water conservancy project safety risk early warning method based on deep learning proposed by the present invention;

[0084] Figure 2 A flow chart for constructing a risk assessment model based on the Cayley manifold fuzzy inference algorithm proposed in the present invention;

[0085] Figure 3 This is a schematic diagram of the inference path optimization based on the seepage Lagrangian optimization model proposed in the present invention. DETAILED DESCRIPTION

[0086] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0087] refer to Figure 1-3 ,The water conservancy project safety risk early warning method based on deep learning includes the following steps:

[0088] S1. Obtain monitoring data of high-flow rate reinjection wells, perform normalization processing, and construct an input data set;

[0089] S2. Based on the Cayley-manifold fuzzy inference algorithm, a fuzzy logic inference model is established, a fuzzy rule set is constructed, a fuzzy membership function is defined, the nonlinear association weights of risk factors are calculated, and the initial inference path is determined;

[0090] S3. Construct a seepage-driven Lagrangian optimization model to calculate the effect of fluid seepage behavior on the fuzzy inference path in a hydraulic engineering project and adjust the inference path in a high-dimensional Cayley manifold space.

[0091] S4. Calculate the critical seepage value, adjust the fuzzy inference rules according to the changes in fluid velocity, formation pressure, and permeability of the water conservancy project, adjust the fuzzy membership mapping based on time series analysis, and dynamically update the inference path;

[0092] S5. Project historical data into a high-dimensional Cayley manifold space, calculate the evolution trend of seepage risk based on the time scale transformation method, and adjust the key variables of the inference model according to the historical status;

[0093] S6. Calculate key parameters under different injection pressures and permeabilities, and update calculation weights in the reasoning path;

[0094] S7. Based on the calculation results of the reasoning path, adjust the calculation structure of the reasoning model.

[0095] In this embodiment, S2 includes the following steps:

[0096] S21. Set the input variable set X = {x1, x2, ..., xn}, where x i Represent the factors that affect the risk of water conservancy injection, construct a mapping from variables to the manifold space M, and satisfy the embedding constraints:

[0097]

[0098] in, is the variable x i The embedding function in the manifold, β is the parameter that controls the distribution range of the variable;

[0099] S22. Set fuzzy inference rule R k , using fuzzy reasoning relationship A ki Represents the variable x i Fuzzy set, B k is the fuzzy set of output variables, w i is the fuzzy weight, which is calculated as follows:

[0100]

[0101] in, is the variable x i In the fuzzy set A j The membership degree in , d j is the fuzzy inference rule R j Correlation factor;

[0102] S23. Exponential membership function using high-dimensional Cayley manifold:

[0103]

[0104] Among them, d M (x,c) is the geodesic distance from variable x to the center c of the fuzzy set on the high-dimensional Cayley manifold M. g ij is the metric tensor of the high-dimensional Cayley manifold M, and α is the fuzzy adjustment parameter;

[0105] S24. Calculate the nonlinear association weights of risk factors and construct the association matrix of fuzzy inference rules:

[0106]

[0107] Among them, γ is the regularization parameter, which controls the degree of correlation between different variables. It is calculated using the geodesic distance on the high-dimensional Cayley manifold: i -x j || 2 =g ij (x i -x j )(x j -xi );g ij is the inverse matrix of the metric tensor;

[0108] S25. Based on the Cayley-manifold fuzzy inference algorithm, set the energy functional of the inference path:

[0109]

[0110] Among them, f is the state function of the reasoning path, Represents the variable x i The membership degree in the fuzzy set is solved by Lagrangian optimization method; the optimal state of the reasoning path is calculated and satisfied To optimize the weight factor, is the Laplace-Beltrami operator on the manifold;

[0111] S26. Calculate the risk assessment value using the optimized reasoning path:

[0112]

[0113] Among them, R is the risk assessment result, and the variable weights in the fuzzy rule set are adjusted according to different hydraulic seepage states;

[0114] S27. Store the optimized fuzzy reasoning path to provide input for subsequent dynamic risk assessment.

[0115] In this embodiment, S3 includes the following steps:

[0116] S31, set the water conservancy project fluid seepage state variable set Z = {z1, z2, ..., z n}, where z i Representing the formation permeability, injection pressure and reinjection velocity, the seepage dynamics constraint conditions are established in the high-dimensional Cayley manifold space M, the seepage Lagrangian optimization model is constructed to constrain the inference path to satisfy the seepage dynamics equation, and the functional is set

[0117]

[0118] Among them, g ij is the inverse matrix of the metric tensor on the high-dimensional Cayley manifold, f is the inference path state function, V(x) represents the percolation potential field, and λ is the Lagrange multiplier;

[0119] S32, functional Perform variational differentiation to derive the governing equations for inference path optimization: in, is the Laplace-Beltrami operator, S cRepresents the critical value of seepage, and the inference path optimization is adjusted according to the dynamic changes of the seepage state;

[0120] S33. Set the constraints for the inference path optimization to satisfy the continuity equation of seepage dynamics: Where ρ is the fluid density, v is the fluid velocity field, and the inference path optimization follows the physical constraints of the seepage motion;

[0121] S34. Calculate the inference path optimization weights under different seepage conditions to satisfy: Among them, W' is the optimized fuzzy weight matrix, γ(x) is the dynamic adjustment factor;

[0122] S35. Calculate the variation range of key parameters based on the optimized reasoning path, and adaptively adjust the reasoning path in combination with the time scale transformation method to maintain stability at different time scales;

[0123] S36. The optimized reasoning path is stored, and dynamically adjusted according to different formation structures and fluid states, and iteratively replaced during the reasoning path calculation structure optimization process.

[0124] In this embodiment, S4 includes the following steps:

[0125] S41, set the seepage state variable set Y = {y1, y2, ..., y m}, where y j Representing the formation permeability, injection pressure, and reinjection velocity on a historical time scale, the time evolution trend of the variables is calculated in the high-dimensional Cayley manifold space M, and the data projection mapping is set:

[0126]

[0127] Among them, T(y j ) represents the variable y j The accumulated changes on the time scale are mapped to a high-dimensional Cayley manifold M for dynamic reasoning and adjustment;

[0128] S42. Calculate the seepage critical value S based on the seepage Lagrangian optimization model c , and derive the optimization equation:

[0129]

[0130] in, is the variable y j The degree of membership in the fuzzy set, w j is the fuzzy weight;

[0131] S43, based on the seepage critical value S c, adjust the fuzzy weight w on the reasoning path j , dynamically adjust the inference path weight matrix according to the changes in the infiltration state at different historical time scales;

[0132] S44, combined with the time series analysis method, based on the seepage critical value S c Changes in membership mapping are adjusted during the fuzzy logic reasoning process;

[0133] S45, calculating the stability of the optimized reasoning path at different time scales, and dynamically adjusting the reasoning path according to changes in the seepage state;

[0134] S46. Store the optimized reasoning path and the adjusted fuzzy reasoning rules, and perform iterative adjustments at different time scales.

[0135] In this embodiment, S5 includes the following steps:

[0136] S51, setting a set of historical seepage data, establishing a data projection mapping in a high-dimensional Cayley manifold space M, so that the historical data satisfies the time evolution constraint in the manifold space;

[0137] S52. Based on the time scale transformation method, the historical data is time-mapped to set the fluid permeability state of the water conservancy project at different time scales;

[0138] S53. In high-dimensional Cayley manifold space, based on the seepage Lagrangian optimization model, calculate the seepage risk change rate at different time scales and establish the dynamic adjustment relationship of the reasoning path on the time scale;

[0139] S54. Adjust the key variables of the inference model based on historical status and calculate the changing trends of risk factors at different time scales;

[0140] S55. Based on the fuzzy logic reasoning method, fuzzy cluster analysis is performed on key variables in historical data to adjust the weights of fuzzy rules;

[0141] S56. Calculate the stability of the inference path at different time scales;

[0142] S57. Store the calculated risk evolution trend data, and perform iterative optimization of the inference path in combination with historical time data, so that the time evolution process of the inference model conforms to the changes in the seepage state under different time scales.

[0143] In this embodiment, S6 includes the following steps:

[0144] S61. Set a set of fluid seepage risk factors for water conservancy projects, calculate the associated weights of the variables in the high-dimensional Cayley manifold space M, and construct an optimized weight matrix under different fluid states;

[0145] S62, calculating the variation range of key parameters based on the inference path, and optimizing the inference path according to the seepage state;

[0146] S63. Based on the weight adjustment strategy on high-dimensional Cayley manifolds, the dynamic weights of the inference paths under different percolation states are calculated, and the optimization factors are adjusted according to the time scale.

[0147] S64. Calculate the risk distribution under different injection pressures and permeabilities, adjust the key variables of the fuzzy inference rules, and adjust the calculation weights on the inference path according to the hydraulic seepage state;

[0148] S65. In the high-dimensional Cayley manifold space, calculate the rate of change of the reasoning path according to the risk distribution state, adjust the calculation parameters of the reasoning path under different pressure conditions, and optimize the convergence speed of the reasoning path;

[0149] S66. Calculate key parameters under different fluid states based on the optimized reasoning path, dynamically adjust the optimization factor of the reasoning path, and update the calculation weight based on historical data;

[0150] S67. Store the optimized reasoning path and perform iterative optimization under different water injection pressure and permeability conditions to provide input for subsequent risk assessment.

[0151] In this embodiment, S7 includes the following steps:

[0152] S71. Set a set of inference path calculation results, and adjust the inference model calculation structure based on the inference path calculation results in the high-dimensional Cayley manifold space M;

[0153] S72. Adjust the calculation weight of the inference rule based on the optimized inference path, optimize the calculation structure, calculate the weight matrix based on the optimized inference path, and optimize the calculation parameters of the inference rule under different stratum structure conditions;

[0154] S73. In the process of optimizing the calculation structure of the inference model, the hydraulic seepage state adjustment factor is introduced to optimize the dynamic allocation of inference rules and optimize the key variables in the calculation structure based on historical data;

[0155] S74. Based on the calculation results of the reasoning path, the calculation weight of the reasoning model is adjusted, and the calculation structure is optimized to adapt the calculation structure to different formation permeability states. The distribution of key variables is calculated according to the optimized reasoning path.

[0156] S75. Adjust the calculation stability of the inference model under different stratum structures based on the optimized calculation structure, and dynamically update the calculation process of the inference path;

[0157] S76. Store the optimized calculation structure, perform iterative optimization under different formation permeability states, and optimize the updating mechanism of the inference rules in the inference path.

[0158] The present invention proposes a dynamic correlation analysis method for nonlinear risk factors based on the Cayley-manifold fuzzy inference model. It innovatively constructs a fuzzy rule set in a high-dimensional Cayley manifold space, adopts an exponential membership function, and uses the geodesic distance on the high-dimensional manifold to accurately calculate the variable correlation weights. Compared with the simple linear or static analysis in traditional methods, the fuzzy inference algorithm used in the present invention can effectively capture the characteristics of high-dimensional nonlinear data, accurately characterize the complex dynamic correlations between key variables such as flow rate, pressure, and permeability, and significantly improve the accuracy and stability of water conservancy project safety risk warnings. Experiments have shown that the risk warning accuracy of the present invention is improved by more than 12% compared with traditional fuzzy logic methods, reflecting the outstanding advantages of the algorithm in high-dimensional nonlinear space mapping and optimization.

[0159] The present invention introduces a seepage-driven Lagrangian optimization model. Aiming at the specific seepage dynamics constraints of water conservancy projects, it establishes strict fluid dynamics equation constraints and uses Lagrange multipliers to optimize the path, thereby achieving efficient dynamic adjustment of the reasoning path. Compared with the defects of existing deep learning methods that ignore the actual engineering physics constraints and lead to insufficient model generalization ability, the method of the present invention can strictly follow the actual fluid physics motion characteristics of the project and dynamically adjust the reasoning path to ensure that the model output always fits the actual seepage state, significantly enhancing the real-time response capability of the model and the reliability of dynamic risk prediction. After field application testing, the real-time performance of risk warning under different working conditions has been improved by about 20%, greatly improving the response capability of risk warning to emergencies.

[0160] The present invention proposes a dynamic update strategy for risk evolution trends based on historical data projection and time-scale transformation. By projecting historical monitoring data in a high-dimensional Cayley manifold space and performing time-scale transformation, the risk evolution trend of historical data is analyzed in real time, and key inference variables and fuzzy membership mappings are dynamically adjusted, enabling the inference model to autonomously adapt to the changing trends of historical and real-time data. Traditional risk warning methods generally lack the ability to comprehensively and dynamically optimize historical and real-time data, resulting in a gradual decrease in the accuracy of the model over long-term use. The method proposed in the present invention can continuously optimize the inference path during long-term operation, significantly improving the long-term stability and adaptability of the model. Practical application results show that the stability and adaptability of the model in long-term operation are improved by approximately 18% compared to traditional methods, effectively ensuring the long-term reliable operation of water conservancy project risk warning.

[0161] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A water conservancy project safety risk early warning method based on deep learning, characterized by: The steps include: S1. Obtain monitoring data of high-flow rate reinjection wells, perform normalization processing, and construct an input data set; S2. Based on the Cayley-manifold fuzzy inference algorithm, a fuzzy logic inference model is established, a fuzzy rule set is constructed, a fuzzy membership function is defined, the nonlinear association weights of risk factors are calculated, and the initial inference path is determined; S3. Construct a seepage-driven Lagrangian optimization model to calculate the effect of fluid seepage behavior in water conservancy projects on the fuzzy inference path and adjust the inference path in the high-dimensional Cayley manifold space. S4. Calculate the critical seepage value, adjust the fuzzy inference rules according to the changes in fluid velocity, formation pressure, and permeability of the water conservancy project, adjust the fuzzy membership mapping based on time series analysis, and dynamically update the inference path; S5. Project historical data into a high-dimensional Cayley manifold space, calculate the evolution trend of seepage risk based on the time scale transformation method, and adjust the key variables of the inference model according to the historical status; S6. Calculate key parameters under different injection pressures and permeabilities, and update calculation weights in the reasoning path; S7. Based on the calculation results of the reasoning path, adjust the calculation structure of the reasoning model.

2. The water conservancy project safety risk early warning method based on deep learning according to claim 1 is characterized in that: The S2 comprises the following steps: S21. Set the input variable set X = {x1, x2, ..., x n }, where x i Represent the factors that affect the risk of water injection, construct a mapping from variables to the manifold space M, and satisfy the embedding constraints: in, is the variable x i The embedding function in the manifold, β is the parameter that controls the distribution range of the variable; S22. Set fuzzy inference rule R k , using fuzzy reasoning relationship A ki Represents the variable x i Fuzzy set, B k is the fuzzy set of output variables, w i is the fuzzy weight, which is calculated as follows: in, is the variable x i In the fuzzy set A j The membership degree in , d j is the fuzzy inference rule R j Correlation factor; S23. Exponential membership function using high-dimensional Cayley manifold: Among them, d M (x,c) is the geodesic distance from variable x to the center c of the fuzzy set on the high-dimensional Cayley manifold M. g ij is the metric tensor of the high-dimensional Cayley manifold M, and α is the fuzzy adjustment parameter; S24. Calculate the nonlinear association weights of risk factors and construct the association matrix of fuzzy inference rules: Among them, γ is the regularization parameter, which controls the degree of correlation between different variables. It is calculated using the geodesic distance on the high-dimensional Cayley manifold: i -x j || 2 =g ij (x i -x j )(x j -x i );g ij is the inverse matrix of the metric tensor; S25. Based on the Cayley-manifold fuzzy inference algorithm, set the energy functional of the inference path: Among them, f is the state function of the reasoning path, Represents the variable x i The membership degree in the fuzzy set is solved by Lagrangian optimization method; the optimal state of the reasoning path is calculated and satisfied To optimize the weight factor, is the Laplace-Beltrami operator on the manifold; S26. Calculate the risk assessment value using the optimized reasoning path: Among them, R is the risk assessment result, and the variable weights in the fuzzy rule set are adjusted according to different hydraulic seepage states; S27. Store the optimized fuzzy reasoning path to provide input for subsequent dynamic risk assessment.

3. The water conservancy project safety risk early warning method based on deep learning according to claim 1 is characterized in that: The S3 includes the following steps: S31, set the water conservancy project fluid seepage state variable set Z = {z1, z2, ..., z n }, where z i Representing the formation permeability, injection pressure and reinjection velocity, the seepage dynamics constraint conditions are established in the high-dimensional Cayley manifold space M, the seepage Lagrangian optimization model is constructed to constrain the inference path to satisfy the seepage dynamics equation, and the functional is set Among them, g ij is the inverse matrix of the metric tensor on the high-dimensional Cayley manifold, f is the inference path state function, V(x) represents the percolation potential field, and λ is the Lagrange multiplier; S32, functional Perform variational differentiation to derive the governing equations for inference path optimization: in, is the Laplace-Beltrami operator, S c Represents the critical value of seepage, and the inference path optimization is adjusted according to the dynamic changes of the seepage state; S33. Set the constraints for the inference path optimization to satisfy the continuity equation of seepage dynamics: Where ρ is the fluid density, v is the fluid velocity field, and the inference path optimization follows the physical constraints of the seepage motion; S34. Calculate the inference path optimization weights under different seepage conditions to satisfy: Among them, W' is the optimized fuzzy weight matrix, γ(x) is the dynamic adjustment factor; S35. Calculate the variation range of key parameters based on the optimized reasoning path, and adaptively adjust the reasoning path in combination with the time scale transformation method to maintain stability at different time scales; S36. The optimized reasoning path is stored, and dynamically adjusted according to different formation structures and fluid states, and iteratively replaced during the reasoning path calculation structure optimization process.

4. The water conservancy project safety risk early warning method based on deep learning according to claim 1 is characterized in that: The S4 comprises the following steps: S41, set the seepage state variable set Y = {y1, y2, ..., y m }, where y j Representing the formation permeability, injection pressure, and reinjection velocity on a historical time scale, the time evolution trend of the variables is calculated in the high-dimensional Cayley manifold space M, and the data projection mapping is set: Among them, T(y j ) represents the variable y j The accumulated changes on the time scale are mapped to a high-dimensional Cayley manifold M for dynamic reasoning and adjustment; S42. Calculate the seepage critical value S based on the seepage Lagrangian optimization model c , and derive the optimization equation: in, is the variable y j The degree of membership in the fuzzy set, w j is the fuzzy weight; S43, based on the seepage critical value S c , adjust the fuzzy weight w on the reasoning path j , dynamically adjust the inference path weight matrix according to the changes in the infiltration state at different historical time scales; S44, combined with the time series analysis method, based on the seepage critical value S c Changes in membership mapping are adjusted during the fuzzy logic reasoning process; S45, calculating the stability of the optimized reasoning path at different time scales, and dynamically adjusting the reasoning path according to changes in the seepage state; S46. Store the optimized reasoning path and the adjusted fuzzy reasoning rules, and perform iterative adjustments at different time scales.

5. The water conservancy project safety risk early warning method based on deep learning according to claim 1 is characterized in that: The S5 comprises the following steps: S51, setting a set of historical seepage data, establishing a data projection mapping in a high-dimensional Cayley manifold space M, so that the historical data satisfies the time evolution constraint in the manifold space; S52. Based on the time scale transformation method, the historical data is time-mapped to set the fluid permeability state of the water conservancy project at different time scales; S53. In high-dimensional Cayley manifold space, based on the seepage Lagrangian optimization model, calculate the seepage risk change rate at different time scales and establish the dynamic adjustment relationship of the reasoning path on the time scale; S54. Adjust the key variables of the inference model based on historical status and calculate the changing trends of risk factors at different time scales; S55. Based on the fuzzy logic reasoning method, fuzzy cluster analysis is performed on key variables in historical data to adjust the weights of fuzzy rules; S56. Calculate the stability of the inference path at different time scales; S57. Store the calculated risk evolution trend data, and perform iterative optimization of the inference path in combination with historical time data, so that the time evolution process of the inference model conforms to the changes in the seepage state under different time scales.

6. The water conservancy project safety risk early warning method based on deep learning according to claim 1 is characterized in that: The S6 comprises the following steps: S61. Set a set of fluid seepage risk factors for water conservancy projects, calculate the associated weights of the variables in the high-dimensional Cayley manifold space M, and construct an optimized weight matrix under different fluid states; S62, calculating the variation range of key parameters based on the inference path, and optimizing the inference path according to the seepage state; S63. Based on the weight adjustment strategy on high-dimensional Cayley manifolds, the dynamic weights of the inference paths under different percolation states are calculated, and the optimization factors are adjusted according to the time scale. S64. Calculate the risk distribution under different injection pressures and permeabilities, adjust the key variables of the fuzzy inference rules, and adjust the calculation weights on the inference path according to the hydraulic seepage state; S65. In the high-dimensional Cayley manifold space, calculate the rate of change of the reasoning path according to the risk distribution state, adjust the calculation parameters of the reasoning path under different pressure conditions, and optimize the convergence speed of the reasoning path; S66. Calculate key parameters under different fluid states based on the optimized reasoning path, dynamically adjust the optimization factor of the reasoning path, and update the calculation weight based on historical data; S67. Store the optimized reasoning path and perform iterative optimization under different water injection pressure and permeability conditions to provide input for subsequent risk assessment.

7. The water conservancy project safety risk early warning method based on deep learning according to claim 1 is characterized in that: The S7 comprises the following steps: S71. Set a set of inference path calculation results, and adjust the inference model calculation structure based on the inference path calculation results in the high-dimensional Cayley manifold space M; S72. Adjust the calculation weight of the inference rule based on the optimized inference path, optimize the calculation structure, calculate the weight matrix based on the optimized inference path, and optimize the calculation parameters of the inference rule under different stratum structure conditions; S73. In the process of optimizing the calculation structure of the inference model, the hydraulic seepage state adjustment factor is introduced to optimize the dynamic allocation of inference rules and optimize the key variables in the calculation structure based on historical data; S74. Based on the calculation results of the reasoning path, the calculation weight of the reasoning model is adjusted, and the calculation structure is optimized to adapt the calculation structure to different formation permeability states. The distribution of key variables is calculated according to the optimized reasoning path. S75. Adjust the calculation stability of the inference model under different stratum structures based on the optimized calculation structure, and dynamically update the calculation process of the inference path; S76. Store the optimized calculation structure, perform iterative optimization under different formation permeability states, and optimize the updating mechanism of the inference rules in the inference path.

Images