Risk chain transmission prediction method and system for small hydropower station groups in a river basin

Abstract

The present invention relates to the technical field of hydropower station safety management, and specifically to a method and system for predicting the chain transmission of risks of a group of small hydropower stations in a river basin. The method includes acquiring real-time hydrological and meteorological data and historical operation data, constructing a hierarchical embedded risk transmission prediction architecture, constructing a complex network of a virtual small hydropower station group, and establishing a multi-domain coupling model using an embedded time series deep learning framework. The model processes multi-source heterogeneous data through a topology-aware input mechanism, captures time series dynamic characteristics through a risk state memory mechanism, and generates risk transmission prediction results through a multi-domain coupling output mechanism. The method significantly improves the accuracy of risk prediction, which is about 35% higher than traditional methods, and the prediction effect is particularly significant in complex chain transmission scenarios. At the same time, the method can discover potential risks 4-6 hours in advance, leaving sufficient preparation time for response measures.

Description

Technical Field

[0001] The present invention relates to the technical field of hydropower station safety management, and in particular to a method and system for predicting the chain transmission risk of a group of small hydropower stations in a river basin. Background Art

[0002] As a key component of renewable clean energy, small hydropower stations occupy a crucial position in my country's energy structure. However, due to their large number, wide distribution, and decentralized management, small hydropower stations in river basins face multiple risks during operation, including flood disasters, equipment failures, and ecological and environmental impacts. These risks have a clear chain-like transmission characteristic within the basin.

[0003] Existing technologies typically use isolated single-site analysis to assess small hydropower plant risks, making it difficult to capture the risk transmission relationships between sites. Some studies have attempted to analyze inter-site relationships from a hydrological perspective, but these analyses are often limited to a single risk type, such as focusing solely on flood risk or equipment failure risk. This makes it difficult to characterize the coupling and transmission mechanisms between different risk types. Furthermore, existing prediction methods generally suffer from insufficient timeliness and low prediction accuracy, making them unable to meet the practical needs of safety management for small hydropower plant clusters within a river basin.

[0004] With the development of artificial intelligence (AI), deep learning has made significant progress in time series forecasting. However, integrating complex network theory with deep learning to accurately predict the chain transmission of risk across a group of small hydropower stations in a river basin remains a pressing technical challenge. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for predicting the chain transmission of risks of a group of small hydropower stations in a river basin, aiming to solve the problem in the existing technology that it is difficult to characterize the laws of the chain transmission of multiple types of risks of a group of small hydropower stations, and to achieve accurate prediction of the chain transmission of flood-equipment failure-ecological pollution risks.

[0006] The present invention proposes a method for predicting chain transmission of risks of a group of small hydropower stations in a river basin, including:

[0007] Obtain real-time hydrological and meteorological data and historical operation data of small hydropower stations in the basin; including:

[0008] Based on the real-time hydrological and meteorological data and historical operational data, a hierarchical embedded risk transfer prediction architecture is constructed, which includes a physical connection layer, a risk mapping layer, and a multi-domain coupling layer;

[0009] Based on the hierarchical embedded risk transfer prediction architecture, a complex network of virtual small hydropower stations is constructed;

[0010] Based on the complex network of the virtual small hydropower station group, an embedded time series deep learning framework is used to establish a multi-domain coupling model of flood, equipment failure and ecological pollution. This model includes:

[0011] Use topology-aware input mechanisms to process multi-source heterogeneous data;

[0012] Capturing temporal dynamics through risk state memory mechanism;

[0013] Generate risk transfer prediction results through multi-domain coupling output mechanism;

[0014] The risk transfer prediction result is output to realize the prediction of risk chain transmission of small hydropower stations in the basin.

[0015] Preferably, the acquisition of real-time hydrological and meteorological data and historical operation data of a group of small hydropower stations in a river basin specifically includes:

[0016] Collect water level, flow, rainfall and temperature data from upstream and downstream power stations at a sampling frequency of 15 minutes.

[0017] Obtain operating status data of key equipment such as generators, transformers, and gates in a small hydropower station cluster, with a sampling frequency of 1 minute per time;

[0018] Collect water quality parameters such as dissolved oxygen, pH value, turbidity, etc. at the water quality monitoring section, with a sampling frequency of 1 hour / time;

[0019] Integrate at least 5 years of historical flood event data, equipment failure records, and ecological pollution event data.

[0020] Preferably, the construction of a complex network of virtual small hydropower stations specifically includes:

[0021] Taking each small hydropower station in the basin as a node, a node attribute vector is constructed, including geographical coordinates, installed capacity, regulation capacity and equipment parameters;

[0022] Directed weighted edges are established based on the hydraulic connections of the river channels, and the edge weights comprehensively consider hydraulic transmission time, flow correlation and geographical proximity;

[0023] Calibrate the hydraulic connection parameters and edge weights between nodes based on measured hydrological data;

[0024] The community detection algorithm is used to identify subgroups of small hydropower stations with close connections, forming a hierarchical network structure.

[0025] Preferably, the construction of a layered embedded risk transfer prediction architecture comprising a physical connection layer, a risk mapping layer, and a multi-domain coupling layer specifically includes:

[0026] In the physical connection layer, node connection relationships based on hydraulic characteristics are established to capture the dynamic transmission characteristics of water flow;

[0027] In the risk mapping layer, a risk measurement space is constructed, and the risk state vector and risk distance function are defined to characterize the spatial and temporal heterogeneity of risks.

[0028] At the multi-domain coupling layer, a bidirectional coupling mechanism of flood-equipment failure, equipment failure-ecological pollution, and ecological pollution-flood risks is established to characterize the interactive relationship between multiple types of risks.

[0029] Preferably, the processing of multi-source heterogeneous data using a topology-aware input mechanism specifically includes:

[0030] Achieve multi-source data fusion and unify data formats and time tags from different sources and sampling frequencies;

[0031] Extract local topological features, mid-range topological features and global topological features to form a multi-scale topological feature representation;

[0032] Identify short-term fluctuation characteristics, medium-term trend characteristics, and long-term cycle characteristics, and construct multi-scale time series characteristics;

[0033] The attention mechanism is used to selectively focus on feature changes at key nodes and key periods.

[0034] Preferably, the capturing of temporal dynamic characteristics through the risk state memory mechanism specifically includes:

[0035] Construct multi-scale state representations for single-station state representation, local network state representation, and global network state representation;

[0036] Achieve a dynamic balance between short-term memory updates and long-term memory maintenance to preserve key historical information;

[0037] Establish a nonlinear state transition mechanism based on threshold triggering to capture the rapid state changes caused by sudden events;

[0038] Smooth transition processing is implemented before and after state transition to ensure the continuity of prediction results.

[0039] Preferably, generating the risk transfer prediction result via the multi-domain coupling output mechanism specifically includes:

[0040] Decode the internal states into flood risk level, equipment failure probability and ecological pollution index respectively;

[0041] Generate short-term risk forecasts for the next 24 hours and medium- to long-term risk trend forecasts for the next 7 days;

[0042] Predict the temporal order, transmission intensity and coupling effects of risk transmission between different sites;

[0043] Determine the warning level, warning scope and response measures based on the forecast results.

[0044] As an option, the verification and optimization process of the risk chain transmission model is also included:

[0045] Verify the model's fitting ability by reproducing historical events;

[0046] Analyze the sources of forecast errors and biases by comparing actual observations with forecasts;

[0047] Optimize model parameters and structure based on validation results;

[0048] Establish a prediction accuracy evaluation mechanism to ensure that the prediction accuracy reaches over 90%.

[0049] Preferably, the application scenarios of the risk chain transmission prediction method for a group of small hydropower stations in a river basin include:

[0050] Flood risk warning during flood season, detecting potential risks 4-6 hours in advance;

[0051] Equipment failure prediction and preventive maintenance can reduce the failure rate by more than 30%;

[0052] Formulate ecological environmental pollution warning and protection measures to reduce the incidence of ecological pollution incidents by more than 50%;

[0053] The coordinated scheduling and optimization of small hydropower stations in the basin will increase water resource utilization efficiency by more than 15%.

[0054] A risk chain transmission prediction system for a group of small hydropower stations in a river basin for implementing the method includes:

[0055] Data acquisition module, used to obtain real-time hydrological and meteorological data and historical operation data of the small hydropower stations in the basin;

[0056] Network construction module, used to construct a complex network of virtual small hydropower stations;

[0057] Risk mapping module, used to represent the spatial and temporal heterogeneity of risks;

[0058] Multi-domain coupling module, used to establish the coupling relationship between flood, equipment failure and ecological pollution;

[0059] A predictive analytics module that generates risk transfer predictions using an embedded time-series deep learning framework;

[0060] Decision support module, used to generate early warning information and intervention measures based on prediction results;

[0061] Model optimization module, used to implement model verification and parameter optimization.

[0062] This paper constructs a hierarchical embedded risk transfer prediction architecture, organically combining complex network theory with deep learning technology to achieve coupled analysis and transfer prediction of multiple types of risks, with the following beneficial effects:

[0063] 1. Significantly improved the accuracy of risk prediction, by approximately 35% compared to traditional methods, especially in complex chain transmission scenarios;

[0064] 2. Extended warning time enables detection of potential risks 4-6 hours in advance, leaving ample time for preparation of response measures;

[0065] 3. Improved water resource utilization efficiency. Through accurate prediction and guidance of coordinated dispatching of power station groups, water resource utilization efficiency has increased by more than 15%;

[0066] 4. Enhanced the effectiveness of ecological environmental protection. By accurately predicting pollution risks, the incidence of ecological pollution incidents has been reduced by more than 50%;

[0067] 5. The scalability of the system has been improved, and the architecture design supports small hydropower stations of different sizes and types with strong adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a flow chart of the method for predicting the chain transmission of risk of small hydropower stations in a river basin according to the present invention;

[0069] Figure 2 This is a schematic diagram of the layered embedded risk transfer prediction architecture of the present invention;

[0070] Figure 3 This is a flow chart of complex network construction of virtual small hydropower station group of the present invention;

[0071] Figure 4 This is a structural diagram of the embedded time series deep learning prediction framework of the present invention. DETAILED DESCRIPTION

[0072] Please refer to the attached Figures 1-4 The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0073] Reference Figure 1 As shown, the risk chain transmission prediction method of a group of small hydropower stations in a river basin proposed by the present invention includes the following steps:

[0074] 1) Obtain real-time hydrological and meteorological data and historical operation data of small hydropower stations in the basin;

[0075] 2) Based on the real-time hydrological and meteorological data and historical operational data, a layered embedded risk transfer prediction architecture is constructed, which includes a physical connection layer, a risk mapping layer, and a multi-domain coupling layer;

[0076] 3) Constructing a complex network of virtual small hydropower stations based on the hierarchical embedded risk transfer prediction architecture;

[0077] 4) Based on the complex network of virtual small hydropower stations, an embedded time-series deep learning framework was used to establish a multi-domain coupling model of flood, equipment failure, and ecological pollution;

[0078] 5) Output the risk transfer prediction results to realize the prediction of risk chain transmission of small hydropower stations in the basin.

[0079] Preferably, the acquisition of real-time hydrological and meteorological data and historical operating data of a group of small hydropower stations in a river basin specifically includes: collecting water level, flow, rainfall and temperature data of upstream and downstream power stations in the river basin, with a sampling frequency of 15 minutes / time; obtaining operating status data of key equipment such as generator sets, transformers, gates, etc. of the small hydropower station group, with a sampling frequency of 1 minute / time; collecting water quality parameters such as dissolved oxygen, pH value, turbidity, etc. of water quality monitoring sections, with a sampling frequency of 1 hour / time; and integrating at least 5 years of historical flood events, equipment failure records and ecological pollution event data.

[0080] In practical applications, water level and flow data can be obtained through the automatic data collection system of river hydrological stations, while rainfall and temperature data can be obtained from meteorological stations. Generator set operating status, including parameters such as speed, power, and vibration, can be obtained from the power plant automation system; transformer status includes oil temperature and winding temperature; and gate status includes opening and operating status. Water quality monitoring data can be obtained through fixed monitoring stations or mobile monitoring equipment. Historical data can be integrated and obtained from the databases of power plant management systems, river basin management departments, and environmental protection departments.

[0081] In a certain embodiment, for 15 small hydropower stations distributed on a tributary of the Yangtze River Basin, water level and flow data were collected from three upstream hydrological stations and four downstream hydrological stations at a sampling frequency of 15 minutes per time; operating status data of 35 units of the 15 power stations were collected, including parameters such as speed, power, and vibration, at a sampling frequency of 1 minute per time; water quality parameters such as dissolved oxygen, pH value, and turbidity were collected from 5 water quality monitoring sections at a sampling frequency of 1 hour per time; historical flood events, equipment failure records, and ecological pollution event data from 2017 to 2022 were integrated.

[0082] Preferably, the construction of a complex network of virtual small hydropower station groups specifically includes: taking each small hydropower station in the basin as a node, constructing a node attribute vector, including geographic coordinates, installed capacity, regulation capacity and equipment parameters; establishing directed weighted edges based on the hydraulic connection of the river channel, and the edge weights comprehensively consider the hydraulic transmission time, flow correlation and geographical proximity; calibrating the hydraulic connection parameters and edge weights between nodes according to the measured hydrological data; and using a community detection algorithm to identify subgroups of small hydropower stations with close connections to form a hierarchical network structure.

[0083] Specifically, the node attribute vector can be expressed as:

[0084] ,

[0085] in: is the attribute vector of the first node, is the geographic coordinate of the node, is the installed capacity (megawatt, MW), is the regulating capacity (hours), is the equipment parameter vector (including the number of units, transformer capacity, etc.).

[0086] The weight of a directed weighted edge can be calculated using the following formula:

[0087] ,

[0088] in: is the edge weight from node i to node j (dimensionless), is the hydraulic transmission time (hours), is the flow correlation (a dimensionless number between 0 and 1), is the geographical proximity (km), are the weight coefficients of the corresponding factors, and According to practical experience, it is usually set .

[0089] The hydraulic transfer time can be calculated from the length of the river and the average flow velocity:

[0090] ,

[0091] in: is the length of the river from node i to node j (km), is the average flow velocity (km / h). The average flow velocity can be calculated using Manning's equation or determined from field measurements. In practice, for a 10-kilometer river with an average flow velocity of 3 km / h, the hydraulic transfer time is approximately 3.33 hours.

[0092] The flow correlation can be calculated by the Pearson correlation coefficient:

[0093] ,

[0094] in: is the flow rate of node i at time t (cubic meters / second), is the average flow rate of node i (cubic meters per second), For node j at time Flow rate (cubic meters per second), is the average flow rate at node j (cubic meters per second), is the time delay of flow transmission (hours), n is the number of samples (units), and t is a discrete time point in the time series. In practical applications, flow correlation is typically between 0.6 and 0.9, and node pairs with a correlation greater than 0.7 are considered to have significant hydraulic connections.

[0095] Geographic proximity can be calculated using Euclidean distance:

[0096] ,

[0097] In order to make the edge weights within a reasonable range, it is usually 、 、 Perform normalization processing.

[0098] The community detection algorithm uses the Louvain algorithm to identify small hydropower station subgroups by maximizing modularity. The calculation formula of modularity Q is as follows:

[0099] ,

[0100] in: is the edge weight between nodes i and i (dimensionless), is the degree of node i (dimensionless), is the sum of the weights of all edges in the network (dimensionless), is the community to which node i belongs (dimensionless integer identifier), is the Kronecker function, when The value is 1 when the and is the node index.

[0101] In one embodiment of the present invention, a virtual complex network was constructed for 15 small hydropower stations on a tributary of the Yangtze River. Three communities were identified using the Louvain algorithm, corresponding to the upstream, midstream, and downstream regions, respectively. The modularity Q value reached 0.68, indicating that the network has obvious community structure characteristics.

[0102] Preferably, the construction of a hierarchical embedded risk transfer prediction architecture comprising a physical connection layer, a risk mapping layer and a multi-domain coupling layer specifically includes: in the physical connection layer, establishing a node connection relationship based on hydraulic characteristics to capture the hydrodynamic transfer characteristics; in the risk mapping layer, constructing a risk measurement space, defining a risk state vector and a risk distance function, and characterizing the spatiotemporal heterogeneity of risks; in the multi-domain coupling layer, establishing a bidirectional coupling mechanism of flood-equipment failure, equipment failure-ecological pollution and ecological pollution-flood risks to characterize the interactive relationship between multiple types of risks.

[0103] like Figure 2 As shown, the layered embedded risk transfer prediction architecture of the present invention includes three tightly coupled layers: a physical connection layer, a risk mapping layer, and a multi-domain coupling layer.

[0104] At the physical connection layer, node connection relationships are established based on hydraulic characteristics. The hydrodynamic transfer characteristics are mainly described by the Saint-Venant equations:

[0105] ,

[0106] ,

[0107] in: is the cross-sectional area of the river (square meters), is the flow rate (cubic meters per second), is the time (seconds), is the distance along the river (meters), is the lateral inflow per unit length (cubic meters per second per meter), is the acceleration due to gravity (9.8 m / s²), is the integral of the water pressure in the x direction (cubic meters), is the riverbed slope (dimensionless), is the friction slope (dimensionless), and denote the partial derivatives with respect to time and space, respectively.

[0108] In practical applications, appropriate simplified models, such as the kinematic wave model or the diffusion wave model, can be selected according to the characteristics of the river channel to improve computational efficiency.

[0109] In the risk mapping layer, a risk measurement space is constructed to map the observation data of the physical layer to the risk state space. The risk state vector is defined as:

[0110] ,

[0111] in: is the risk state vector (dimensionless) at time t, is the flood risk (a dimensionless number between 0 and 1), is the risk of equipment failure (a dimensionless number between 0 and 1), is the ecological pollution risk (a dimensionless number between 0 and 1).

[0112] The risk distance function is used to characterize the spatiotemporal characteristics of risk transmission between different nodes and is defined as:

[0113] ,

[0114] in: For nodes The risk state at time t is the same as the risk state of node j at time The distance between risk states (dimensionless), is the risk transfer time (hours), For the The weight coefficient of the class risk (dimensionless), , For nodes At the moment No. Class risk value (a dimensionless number between 0 and 1), For nodes At the moment The k-th risk value (a dimensionless number between 0 and 1) is the risk type index, and its values are 1, 2, and 3, corresponding to the three risks of flood, equipment failure, and ecological pollution, respectively.

[0115] The spatial and temporal heterogeneity of risk can be characterized by the following indicators:

[0116] ,

[0117] ,

[0118] ,

[0119] in: is the spatial heterogeneity index (dimensionless), is the proportion of the risk of node i to the total risk (dimensionless); is the temporal heterogeneity index (dimensionless), is the proportion of risk at time t to total risk (dimensionless); is the type heterogeneity index (dimensionless), is the proportion of the kth risk to the total risk (dimensionless); is the total number of nodes in the network, is the total length of the time series (time steps), These indicators are based on the Shannon entropy principle and are used to quantify the uneven distribution of risk in different dimensions.

[0120] At the multi-domain coupling layer, the coupling relationship between different types of risks is established, which can be represented by the following coupling matrix:

[0121] ,

[0122] in: represents the impact coefficient of the i-th risk on the j-th risk (dimensionless), , corresponding to three types of risks: flood, equipment failure and ecological pollution. For example, represents the impact coefficient of flood on equipment failure, Represents the impact coefficient of equipment failure on ecological pollution.

[0123] According to the analysis of measured data, the impact coefficient of flood on equipment failure in a typical small hydropower station group is The impact coefficient of equipment failure on ecological pollution is about 0.65-0.85. The feedback coefficient of ecological pollution on floods is about 0.50-0.70. About 0.20-0.30.

[0124] In one embodiment of the present invention, by analyzing the historical data of a group of small hydropower stations in a tributary of the Yangtze River, the coupling matrix obtained is:

[0125] ,

[0126] The matrix shows that floods have a strong impact on equipment failure (0.78), and equipment failure has a significant impact on ecological pollution (0.63), while the feedback effects of ecological pollution on floods and equipment failure are relatively weak (0.22 and 0.18).

[0127] Preferably, the use of a topology-aware input mechanism to process multi-source heterogeneous data specifically includes: realizing multi-source data fusion, unifying data formats and time labels from different sources and with different sampling frequencies; extracting local topological features, mid-range topological features, and global topological features to form a multi-scale topological feature representation; identifying short-term fluctuation features, medium-term trend features, and long-term cycle features to construct multi-scale time series features; and selectively focusing on feature changes at key nodes and key time periods through an attention mechanism.

[0128] In the multi-source data fusion process, it is necessary to solve the problem of inconsistent sampling frequencies of different data sources. This paper uses time interpolation to align data to ensure that all data have a unified time label. The specific method is:

[0129] ,

[0130] in: is the data value at time t that needs to be interpolated (consistent with the original data unit), and are the two most recent sampling times before and after t (hours), and Separate moments and The measured data value (the unit is consistent with the original data), is the time point (hour) at which interpolation is required, is the time weighting factor (dimensionless, range 0-1). This linear interpolation method is suitable for parameters with relatively gradual changes, such as water level and flow. For data with significantly different sampling frequencies, such as hydrological data (15-minute interval) and water quality data (1-hour interval), the system first interpolates all data to the minimum time interval (15 minutes) before performing subsequent processing to ensure accurate time alignment.

[0131] The Z-score method is used for data standardization:

[0132] ,

[0133] in: is the standardized data (dimensionless), X is the original data (the unit is consistent with the original data), is the mean (consistent with the original data unit), is the standard deviation (same as the original data unit).

[0134] Topological feature extraction is one of the core innovations of this invention. Local topological features mainly characterize the first-order neighborhood structure of nodes and can be calculated using the following indicators:

[0135] ,

[0136] in: is the local topological eigenvector of node i (dimensionless), is the degree of node i (dimensionless), is the clustering coefficient of node i (dimensionless), is the betweenness centrality of node i (dimensionless).

[0137] The medium-distance topology mainly characterizes the network structure within 2-3 hops:

[0138] ,

[0139] in: is the mid-range topological eigenvector of node i (dimensionless), is the number of 2-hop neighbors of node i, is the number of 3-hop neighbors of node i, is the effective radius (number of hops) of node i.

[0140] The global topological features are mainly based on spectral graph theory and are extracted through the eigenvalues and eigenvectors of the Laplace matrix:

[0141] ,

[0142] ,

[0143] in: is the Laplace matrix (dimensionless), is the degree matrix (dimensionless), is the adjacency matrix (dimensionless), is the normalized Laplace matrix (dimensionless), is the identity matrix (dimensionless), It represents the diagonal matrix formed by raising the elements of the diagonal matrix D to the power of -1 / 2.

[0144] The global topological features can be expressed as the eigenvectors corresponding to the first k smallest non-zero eigenvalues of the Laplacian matrix:

[0145] ,

[0146] in: is the global topological characteristic matrix (dimensionless), is the dimensionless eigenvector corresponding to the i-th eigenvalue. In practical applications, k is usually set to 5-10, that is, the first 5-10 eigenvectors are extracted as global topological features.

[0147] Time series feature extraction also uses a multi-scale strategy. Short-term fluctuation characteristics are mainly captured through moving window statistics:

[0148] ,

[0149] in: is the short-term fluctuation characteristic vector at time t (dimensionless), is the data in the window centered at time t (original data), is the mean of the data in the window, is the standard deviation of the data in the window, is the maximum value of the data in the window, is the minimum value of the data in the window, The range of the data within the window. The window width is usually 6-24 hours.

[0150] The medium-term trend characteristics are mainly extracted through the trend decomposition method:

[0151] ,

[0152] in: is the original time series (original data), is the trend component (original data), is the seasonal component (raw data), is the residual component (original data). The medium-term trend characteristics are .

[0153] Long-term periodic features are mainly extracted through Fourier transform or wavelet transform:

[0154] ,

[0155] in: for The Fourier transform of represents the amplitude of different frequency components (complex number), is the imaginary unit ( is the angular frequency (radians per second), is the base of natural logarithms (approximately equal to 2.71828), Represents the integral differential over time and space.

[0156] The attention mechanism is used to selectively focus on feature changes at key nodes and key periods, which can be expressed as:

[0157] ,

[0158] ,

[0159] in: is the attention coefficient of node i to node i (a dimensionless number between 0 and 1), is the similarity coefficient between nodes i and j (dimensionless), is the similarity function, is the weight matrix (dimensionless), and are the dimensionless eigenvectors of nodes i and j, is an exponential function with the natural logarithm base e as the base, is the total number of nodes in the network.

[0160] In one embodiment of the present invention, a topology-aware input mechanism was applied to data from a cluster of small hydropower stations on a tributary of the Yangtze River. The system extracted local, mid-range, and global topological features of 15 nodes, as well as short-term fluctuations, medium-term trends, and long-term cyclical features within a year. Through the attention mechanism, the system automatically identified characteristic changes at key nodes (such as large hydropower stations with strong regulatory capabilities) and during key periods (such as the flood season), improving the targetedness and accuracy of forecasts.

[0161] Preferably, the capture of temporal dynamic characteristics through the risk state memory mechanism specifically includes: constructing a multi-scale state representation of single-station state representation, local network state representation and global network state representation; achieving a dynamic balance between short-term memory update and long-term memory maintenance to maintain key historical information; establishing a nonlinear state transition mechanism based on threshold triggering to capture the sharp changes in state caused by emergencies; and implementing smooth transition processing before and after state transition to ensure the continuity of prediction results.

[0162] The risk state memory mechanism is another core innovation of this invention, mainly based on the improved design of the long short-term memory network (LSTM). The multi-scale state representation includes:

[0163] 1. Single station status representation: represents the risk status of a single hydropower station, with three dimensions (flood, equipment failure, and ecological pollution);

[0164] 2. Local network state representation: represents the overall state of the small hydropower station sub-network, with a dimension of 3×K, where K is the number of nodes in the sub-network;

[0165] 3. Global network state representation: represents the system state of the small hydropower station group in the entire basin, with a dimension of 3×N, where N is the total number of nodes in the network.

[0166] The dynamic balance between short-term memory updating and long-term memory maintenance is achieved through the LSTM gating mechanism. The specific formula is as follows:

[0167] ,

[0168] ,

[0169] ,

[0170] ,

[0171] ,

[0172] ,

[0173] in: It is the forget gate, which controls the retention ratio of the memory at the previous moment (the vector between O-1); It is the input gate, which controls the ratio of new input information received (the vector between O-1); is the candidate memory unit (vector); is the current memory unit (vector); is the output gate, which controls the output ratio of the current memory (the vector between O-1); is the hidden state (vector); is the input at the current moment (vector); is the corresponding weight matrix (dimensionless); is the corresponding bias vector (dimensionless); is the sigmoid activation function, defined as , the output range is (0,1); is the hyperbolic tangent activation function, defined as , the output range is (-1,1); is the Hadamard product (element-wise multiplication); Indicates that the vector and Concatenate columns into a new vector.

[0174] This paper improves the standard LSTM and adds a nonlinear state transition mechanism to capture the rapid state changes caused by sudden events:

[0175] ,

[0176] ,

[0177] in: is the transition indicator (a dimensionless number with a value of 0 or 1), is the Heaviside step function (output 1 when the input is greater than 0, otherwise output 0), is the state cumulative value (dimensionless), is the transition threshold (dimensionless), is the original candidate memory unit (vector, calculated by standard LSTM), is the updated candidate memory unit (vector, as the final use), is the candidate memory unit after the transition (vector, representing the memory unit value in the burst state). The core idea of the state transition mechanism is: when the system state accumulates When the transition mechanism is triggered, the transition state is used Replace the original state ; otherwise, keep the original state unchanged. It can be generated by a sub-model trained specifically for sudden events, or by adding preset transition vectors to the original state:

[0178] ,

[0179] in: is the transition vector (a vector representing the additional memory change under the sudden state), which is learned from historical sudden event data.

[0180] In practical applications, appropriate transition thresholds can be set based on historical data analysis. For example, for flood risk, a flow exceeding 1.2 times the 10-year flood flow could be set as the transition threshold; for equipment failure risk, an equipment abnormality indicator exceeding 1.5 times the safety value could be set as the transition threshold; for ecological pollution risk, a water quality parameter exceeding 1.3 times the standard limit could be set as the transition threshold.

[0181] In order to ensure the continuity of the prediction results, the present invention implements smooth transition processing before and after the state transition:

[0182] ,

[0183] in: is the memory unit (vector) after smoothing, is the original memory unit (vector), is the prediction memory unit (vector), is the smoothing coefficient, ranging from [0,1] (dimensionless). It can be adjusted dynamically over time, gradually increasing before the transition and gradually decreasing after the transition to achieve a smooth transition.

[0184] In one embodiment of the present invention, a risk state memory mechanism was applied to a cluster of small hydropower stations on a tributary of the Yangtze River, successfully capturing a flood risk jump event triggered by sudden heavy rainfall upstream. The system issued an early warning 4.5 hours before the flood arrived, approximately two hours earlier than traditional methods, saving valuable time for flood control operations at downstream power stations. Furthermore, through smooth transition processing, the prediction results maintained good continuity, avoiding frequent fluctuations in the warning signal and improving the reliability of the warning.

[0185] Preferably, the generation of risk transfer prediction results via the multi-domain coupling output mechanism specifically includes: decoding the internal state into flood risk level, equipment failure probability and ecological pollution index respectively; generating short-term risk forecasts for the next 24 hours and medium- and long-term risk trend forecasts for the next 7 days; predicting the time sequence, transmission intensity and coupling effect of risk transmission between different sites; and determining the warning level, warning range and response measures based on the prediction results.

[0186] The multi-domain coupling output mechanism is the key to achieving risk chain transmission prediction. First, the internal state needs to be decoded into various risk indicators:

[0187] Flood risk levels decoded:

[0188] ,

[0189] in: is the probability distribution of flood risk level (5-dimensional vector, corresponding to the probability of 5 risk levels), which is usually divided into 5 levels (no risk, slight, moderate, severe, extreme); and are the corresponding weight matrices and bias vectors (dimensionless); is the hidden state (vector) of LSTM; Softmax is the softmax function, defined as Softmax( ) = ,in is a vector The i-th element of K is a vector The function converts the input into a probability distribution so that the sum of all elements is 1.

[0190] Equipment failure probability decoding:

[0191] ,

[0192] in: is the probability of equipment failure (a dimensionless number between 0 and 1), and is the corresponding weight matrix and bias vector (dimensionless); Sigmoid is the sigmoid function, defined as Sigmoid( ) = , the output range is (0,1).

[0193] Decoding of ecological pollution indicators:

[0194] ,

[0195] in: is an ecological pollution index vector (multiple water quality parameter values), which can include multiple water quality parameters (such as dissolved oxygen, pH value, turbidity, etc.); and are the corresponding weight matrices and bias vectors (dimensionless).

[0196] Short-term risk forecast (24 hours) and medium- to long-term risk trend forecast (7 days) use a recursive forecasting method:

[0197] ,

[0198] ,

[0199] in: is the hidden state (vector) at the next moment, is the prediction input (vector) for the next moment, As the input prediction function, LSTM represents the long short-term memory network calculation process defined above. is the input (vector) at the current moment. For longer-term predictions, a multi-step recursive strategy is used, where the prediction result of the previous step is used as the input for the next step to continue the prediction.

[0200] Enter the prediction function The specific construction method is a multi-layer perceptron (MLP) network:

[0201] ,

[0202] in: and are the weight matrices of the first and second layers respectively (dimensionless), and are the bias vectors (dimensionless) of the first and second layers respectively, and ReLU is the rectified linear unit activation function, defined as ReLU Indicates that the vector and Concatenate columns into a new vector.

[0203] This construction method enables the system to predict the input at the next moment based on the input and hidden state at the current moment, thereby achieving multi-step recursive prediction. In practical applications, to reduce error accumulation, actual observation data will be introduced for correction after a certain number of steps:

[0204] ,

[0205] in: is the predicted input (vector) at time t+n, is the actual observation data (vector) at time t+n, is the hidden state (vector) at time t+n-1. The correction interval is usually set to 3 to 6 time steps. This method of combining recursive prediction with periodic correction is adjusted according to data characteristics and prediction requirements. It can avoid the error accumulation problem in long-term prediction while maintaining prediction continuity. It is particularly suitable for systems with complex dynamic characteristics such as small hydropower stations in a river basin.

[0206] Risk transfer prediction is the core output of this invention, including transfer time, transfer intensity and coupling effect:

[0207] Delivery time prediction:

[0208] ,

[0209] in: is the predicted time (hours) for risk to be transferred from node i to node j, is the benchmark transfer time under normal hydrological conditions (hours), is the current flow rate of node i (cubic meters per second), is the baseline flow rate of node i (cubic meters per second), is the adjustment coefficient (dimensionless), usually taking a value of 0.2-0.5.

[0210] Transfer Strength Prediction:

[0211] ,

[0212] in: is the intensity of risk transferred from node i to node j (a dimensionless number between 0 and 1), is the risk intensity of node i (a dimensionless number between 0 and 1), is the distance between nodes i and j (km), is the attenuation coefficient (1 / km), is the coupling coefficient between risk types k and l (dimensionless), is the coupling adjustment coefficient (dimensionless), is the base of the natural logarithm (approximately equal to 2.71828). k and l are the type indexes of the risk source and target risk, respectively, with values of 1, 2, and 3 corresponding to the three types of risks: flood, equipment failure, and ecological pollution.

[0213] Coupling effect prediction:

[0214] ,

[0215] in: is the coupling effect between nodes i and j (dimensionless), is the coupling weight between risk types k and l (dimensionless), is the k-th risk of node i (a dimensionless number between 0 and 1), is the lth type of risk at node j (a dimensionless number between 0 and 1). The coupling effect describes the interaction strength between different types of risks.

[0216] Based on the forecast results, the warning level, warning scope, and response recommendations are determined. Warning levels are generally divided into four levels: blue (general), yellow (severe), orange (critical), and red (extremely severe). The warning scope is determined by the risk transmission path, and response recommendations are formulated based on the risk type and level.

[0217] In one embodiment of the present invention, the system performed a risk chain transmission forecast for a cluster of small hydropower stations on a tributary of the Yangtze River, successfully predicting a risk transmission process starting from an upstream power station, spreading through midstream power stations, and then to downstream power stations. The forecast showed that the flood risk from the upstream power station would be transmitted to the midstream power station six hours later, increasing the equipment failure risk at the midstream power station from 0.15 to 0.68. The risk would then be transmitted to the downstream power station four hours later, increasing the ecological pollution risk at the downstream power station from 0.08 to 0.52. Based on this, the system issued a yellow alert, preemptively arranged equipment maintenance at the midstream power station and water quality monitoring at the downstream power station, successfully averting a potential equipment failure and water pollution incident.

[0218] In another embodiment of the present invention, the verification and optimization process of the risk chain transmission model is also included: verifying the fitting ability of the model by reproducing historical events; analyzing the sources of prediction errors and deviations by comparing actual observations with predictions; optimizing the model parameters and structure based on the verification results; and establishing a prediction accuracy evaluation mechanism, so that the prediction accuracy reaches more than 90%.

[0219] Model validation and optimization are key steps to ensure the reliability of the forecasting system. Backtesting is used to reproduce historical events. Typical historical events are selected for simulation and forecasting. The forecast results are compared with actual observations to calculate the goodness of fit:

[0220] ,

[0221] in: is the goodness of fit (dimensionless), is the actual observed value (original unit), is the predicted value (original unit), is the mean of the observations (in original units), is the number of samples. Ideally, It should be close to 1, indicating that the model can fit the actual data well. The prediction error analysis uses the root mean square error (RMSE) and mean absolute percentage error (MAPE):

[0222] ,

[0223] ,

[0224] Where: RMSE is the root mean square error (same as the original data unit), MAPE is the mean absolute percentage error (%).

[0225] According to the error analysis results, the gradient descent method is used to optimize the model parameters:

[0226] ,

[0227] in: is the current parameter (dimensionless), is the updated parameter (dimensionless), is the learning rate (dimensionless), is the loss function About parameters The gradient of (dimensionless).

[0228] The prediction accuracy evaluation is based on the confusion matrix calculation:

[0229] ,

[0230] in: is the prediction accuracy (%), For real examples, is a true negative example, is a false positive example (), is a false negative example.

[0231] In the embodiment of the present invention, the risk chain transmission prediction system of a group of small hydropower stations in a tributary of the Yangtze River Basin was verified, and 15 typical risk events between 2017 and 2022 were selected for backtesting. The results showed that the model's goodness of fit for flood risk R 2 The accuracy of the 24-hour short-term forecast reached 94.3%, and the accuracy of the 7-day medium- and long-term forecast reached 91.5%, meeting the requirements of practical applications.

[0232] Preferably, the application scenarios of the risk chain transmission prediction method for a group of small hydropower stations in a river basin include: flood risk warning during flood season, discovering potential risks 4-6 hours in advance; equipment failure prediction and preventive maintenance, reducing the failure rate by more than 30%; ecological environmental pollution warning and protection measures formulation, reducing the incidence of ecological pollution incidents by more than 50%; coordinated scheduling and optimization of a group of small hydropower stations in a river basin, improving water resource utilization efficiency by more than 15%.

[0233] The risk chain transmission prediction system for a group of small hydropower stations in a river basin proposed in the present invention is used to implement the above method, and includes: a data acquisition module, a network construction module, a risk mapping module, a multi-domain coupling module, a prediction and analysis module, a decision support module and a model optimization module.

[0234] The data acquisition module is used to obtain real-time hydrological and meteorological data and historical operating data for the small hydropower stations in the basin. It specifically includes a hydrological data acquisition unit, an equipment status acquisition unit, a water quality monitoring unit, and a historical data management unit. The hydrological data acquisition unit collects water level, flow, rainfall, and temperature data from upstream and downstream power stations in the basin; the equipment status acquisition unit acquires operating status data for key equipment such as generators, transformers, and gates at the small hydropower stations; the water quality monitoring unit collects water quality parameters such as dissolved oxygen, pH, and turbidity at the monitoring sections; and the historical data management unit integrates historical flood events, equipment failure records, and ecological pollution incident data.

[0235] The network construction module is used to construct a complex network of virtual small hydropower stations. Specifically, it includes a node construction unit, an edge establishment unit, a weight calculation unit, and a community identification unit. The node construction unit is responsible for constructing node attribute vectors based on each small hydropower station in the basin; the edge establishment unit is responsible for establishing directed weighted edges based on the hydraulic connections of the river channel; the weight calculation unit is responsible for calculating edge weights, taking into account hydraulic transmission time, flow correlation, and geographic proximity; and the community identification unit is responsible for identifying closely connected small hydropower station subgroups using a community detection algorithm.

[0236] The risk mapping module is used to characterize the spatiotemporal heterogeneity of risk. It specifically includes a risk measurement unit, a risk distance calculation unit, and a risk heterogeneity analysis unit. The risk measurement unit is responsible for constructing the risk measurement space and defining the risk state vector; the risk distance calculation unit is responsible for calculating the spatiotemporal characteristics of risk transmission between different nodes; and the risk heterogeneity analysis unit is responsible for characterizing the heterogeneity of risk in space, time, and type.

[0237] The multi-domain coupling module establishes the coupling relationship between floods, equipment failures, and ecological pollution. Specifically, it includes a flood-equipment coupling unit, an equipment-ecology coupling unit, and an ecology-flood coupling unit. The flood-equipment coupling unit establishes the impact mechanism of floods on equipment failures; the equipment-ecology coupling unit establishes the impact mechanism of equipment failures on ecological pollution; and the ecology-flood coupling unit establishes the feedback mechanism of ecological pollution on floods.

[0238] The predictive analysis module utilizes an embedded time-series deep learning framework to generate risk transfer predictions. Specifically, it comprises a data preprocessing unit, a topological feature extraction unit, a time-series feature extraction unit, an LSTM prediction unit, and a result decoding unit. The data preprocessing unit is responsible for data cleaning, standardization, and alignment; the topological feature extraction unit extracts multi-scale topological features; the time-series feature extraction unit extracts multi-scale time-series features; the LSTM prediction unit generates predictions based on a risk state memory mechanism; and the result decoding unit decodes the predictions into flood risk levels, equipment failure probabilities, and ecological pollution indicators.

[0239] The decision support module generates warning information and intervention measures based on the prediction results. Specifically, it comprises a warning generation unit, a scope determination unit, and an intervention suggestion unit. The warning generation unit is responsible for generating graded warning information based on the prediction results; the scope determination unit is responsible for determining the scope and targets of the warning; and the intervention suggestion unit is responsible for generating risk prevention and control recommendations.

[0240] The model optimization module is used for model validation and parameter optimization. It specifically includes a validation unit, an error analysis unit, and a parameter optimization unit. The validation unit verifies the model's fit by reproducing historical events; the error analysis unit analyzes sources of prediction errors and deviations; and the parameter optimization unit optimizes model parameters and structure based on the validation results.

[0241] During the actual operation of the system, there is a close data flow relationship between the various modules. The data acquisition module first acquires real-time and historical data and transmits the processed data to both the network construction module and the predictive analysis module. The network construction module constructs a virtual complex network based on the received data and transmits the network structure information to the risk mapping module. The risk mapping module maps the physical space to the risk space and transmits the risk representation information to the multi-domain coupling module. The multi-domain coupling module establishes coupling relationships for different risk types and transmits the coupling matrix to the predictive analysis module. The predictive analysis module integrates all information to generate risk transfer prediction results, which are then passed to the decision support module and the model optimization module. The decision support module generates warning information and intervention recommendations based on the prediction results and provides them to the user. The model optimization module then feeds back the optimization parameters to the predictive analysis module based on the discrepancies between the prediction results and actual observations, forming a closed-loop optimization.

[0242] The foregoing is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It will be apparent to those skilled in the art that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A method for predicting the chain transmission of risk of a group of small hydropower stations in a river basin, characterized by: include: Obtain real-time hydrological and meteorological data and historical operation data of small hydropower stations in the basin; including: Based on the real-time hydrological and meteorological data and historical operational data, a hierarchical embedded risk transfer prediction architecture is constructed, which includes a physical connection layer, a risk mapping layer, and a multi-domain coupling layer; Based on the hierarchical embedded risk transfer prediction architecture, a complex network of virtual small hydropower stations is constructed; Based on the complex network of the virtual small hydropower station group, an embedded time series deep learning framework is used to establish a multi-domain coupling model of flood, equipment failure and ecological pollution. This model includes: Use topology-aware input mechanisms to process multi-source heterogeneous data; Capturing temporal dynamics through risk state memory mechanism; Generate risk transfer prediction results through multi-domain coupling output mechanism; The risk transfer prediction result is output to realize the prediction of risk chain transmission of small hydropower stations in the basin.

2. The method according to claim 1, characterized in that The acquisition of real-time hydrological and meteorological data and historical operation data of the small hydropower stations in the basin specifically includes: Collect water level, flow, rainfall and temperature data from upstream and downstream power stations at a sampling frequency of 15 minutes. Obtain operating status data of key equipment such as generators, transformers, and gates of small hydropower stations, with a sampling frequency of 1 minute per time; Collect dissolved oxygen, pH value, and turbidity water quality parameters at the water quality monitoring section with a sampling frequency of 1 hour / time; Integrate at least 5 years of historical flood event data, equipment failure records, and ecological pollution event data.

3. The method according to claim 1, characterized in that The construction of a complex network of virtual small hydropower stations specifically includes: Taking each small hydropower station in the basin as a node, a node attribute vector is constructed, including geographical coordinates, installed capacity, regulation capacity and equipment parameters; Directed weighted edges are established based on the hydraulic connections of the river channels, and the edge weights comprehensively consider hydraulic transmission time, flow correlation and geographical proximity; Calibrate the hydraulic connection parameters and edge weights between nodes based on measured hydrological data; The community detection algorithm is used to identify subgroups of small hydropower stations with close connections, forming a hierarchical network structure.

4. The method according to claim 1, wherein The construction of a layered embedded risk transfer prediction architecture comprising a physical connection layer, a risk mapping layer, and a multi-domain coupling layer specifically includes: In the physical connection layer, node connection relationships based on hydraulic characteristics are established to capture the dynamic transmission characteristics of water flow; In the risk mapping layer, a risk measurement space is constructed, and the risk state vector and risk distance function are defined to characterize the spatial and temporal heterogeneity of risks. At the multi-domain coupling layer, a bidirectional coupling mechanism of flood-equipment failure, equipment failure-ecological pollution, and ecological pollution-flood risks is established to characterize the interactive relationship between multiple types of risks.

5. The method according to claim 1, wherein The processing of multi-source heterogeneous data using the topology-aware input mechanism specifically includes: Achieve multi-source data fusion and unify data formats and time tags from different sources and sampling frequencies; Extract local topological features, mid-range topological features and global topological features to form a multi-scale topological feature representation; Identify short-term fluctuation characteristics, medium-term trend characteristics, and long-term cycle characteristics, and construct multi-scale time series characteristics; The attention mechanism is used to selectively focus on feature changes at key nodes and key periods.

6. The method according to claim 1, characterized in that The capture of temporal dynamic characteristics through the risk state memory mechanism specifically includes: Construct multi-scale state representations for single-station state representation, local network state representation, and global network state representation; Achieve a dynamic balance between short-term memory updates and long-term memory maintenance to preserve key historical information; Establish a nonlinear state transition mechanism based on threshold triggering to capture the rapid state changes caused by sudden events; Smooth transition processing is implemented before and after state transition to ensure the continuity of prediction results.

7. The method according to claim 1, characterized in that Generating the risk transfer prediction result via the multi-domain coupling output mechanism specifically includes: Decode the internal states into flood risk level, equipment failure probability and ecological pollution index respectively; Generate short-term risk forecasts for the next 24 hours and medium- to long-term risk trend forecasts for the next 7 days; Predict the temporal order, transmission intensity and coupling effects of risk transmission between different sites; Determine the warning level, warning scope and response measures based on the forecast results.

8. The method according to claim 7, characterized in that It also includes the verification and optimization process of the risk chain transmission model: Verify the model's fitting ability by reproducing historical events; Analyze the sources of forecast errors and biases by comparing actual observations with forecasts; Optimize model parameters and structure based on validation results; Establish a prediction accuracy evaluation mechanism.

9. The method according to claim 8, characterized in that The application scenarios of the risk chain transmission prediction method for small hydropower stations in a river basin include: Flood risk warning during flood season; Equipment failure prediction and preventive maintenance; Early warning and protection measures for ecological and environmental pollution; Coordinated scheduling and optimization of small hydropower stations within the basin.

10. A risk chain transmission prediction system for a group of small hydropower stations in a river basin that implements the method according to any one of claims 1 to 9, characterized in that: include: Data acquisition module, used to obtain real-time hydrological and meteorological data and historical operation data of the small hydropower stations in the basin; Network construction module, used to construct a complex network of virtual small hydropower stations; Risk mapping module, used to represent the spatial and temporal heterogeneity of risks; Multi-domain coupling module, used to establish the coupling relationship between flood, equipment failure and ecological pollution; A predictive analytics module that generates risk transfer predictions using an embedded time-series deep learning framework; Decision support module, used to generate early warning information and intervention measures based on prediction results; Model optimization module, used to implement model verification and parameter optimization.

Images