Power load forecasting method and system based on association rule analysis

Abstract

The present application relates to the technical field of power systems, and provides a method and system for power load forecasting based on association rule analysis, which is used to solve the problems of delayed response to hidden faults caused by the lack of a dynamic coupling mechanism between equipment health status and load fluctuations in the prior art, and load forecast compensation deviation caused by insufficient quantification of fault propagation path weights. The method of the present application includes: generating equipment status data based on the vibration spectrum and insulation aging index of the power equipment; performing fusion analysis to generate equipment health assessment indicators, and establishing a dynamic impact model of equipment abnormal events on grid load fluctuations through association rules; identifying potential abnormal equipment for logical mapping to generate fault propagation path weights; and dynamically adjusting the weights to obtain adjusted power load forecast values. The technical solution provided by the present application integrates equipment vibration and insulation aging data, constructs health assessment and fault propagation models, and dynamically corrects load forecasts to prevent and control grid risks.

Description

Technical Field

[0001] The present application relates to the technical field of power systems, and in particular to a method and system for power load forecasting based on association rule analysis. Background Art

[0002] As the construction of new power systems progresses, grid load characteristics are becoming increasingly complex, and the high proportion of renewable energy integrated into the grid is exacerbating load volatility and uncertainty. Against this backdrop, there is an urgent need to integrate dynamic awareness of equipment health status into load forecasting to mitigate the risk of sudden load fluctuations caused by hidden equipment failures. Two core requirements must be addressed: first, establishing a real-time coupling analysis model between equipment status parameters and load fluctuations; second, quantifying the impact of equipment anomalies on the propagation of faults in the grid topology to enable dynamic compensation for load forecasting.

[0003] Existing solutions typically integrate device status with load forecasting using deep learning-based time series forecasting models, such as those combining stacked long short-term memory networks with federated learning. This approach captures the temporal characteristics of historical load data through multi-hidden-layer long short-term memory networks and uses a federated learning framework to aggregate load data from distributed regions, improving the model's ability to generalize complex time series patterns. For example, a federated long short-term memory network model proposed by a power grid optimizes load forecasting accuracy through distributed training, but does not incorporate dynamic correlation analysis between device health indicators and grid topology.

[0004] Although the existing solutions mentioned above can handle the nonlinear time series characteristics of load data, they lack the correlation between equipment status and load: the model cannot identify abnormal load fluctuations caused by the decline of equipment health. For example, a sudden increase in leakage current caused by insulation degradation may be misjudged as a normal load change; and the fault propagation path is insufficiently modeled: the existing model does not analyze the fault propagation weight in combination with the grid topology structure. When the equipment is abnormal, it is difficult to quantify its chain reaction effect on upstream and downstream nodes, resulting in an increase in the deviation of the predicted value in the fault scenario. Summary of the Invention

[0005] The present application provides a method and system for power load forecasting based on association rule analysis, which is used to solve the problems of delayed response to hidden faults caused by the lack of a dynamic coupling mechanism between equipment health status and load fluctuations in the prior art, and load forecast compensation deviation caused by insufficient quantification of fault propagation path weights.

[0006] In a first aspect, the present application provides a method for power load forecasting based on association rule analysis, comprising:

[0007] Generating equipment status data according to the vibration spectrum data and insulation aging index of the power equipment, wherein the equipment status data includes the vibration spectrum amplitude and the insulation aging index change rate;

[0008] Performing a fusion analysis on the vibration spectrum amplitude and the insulation aging index change rate to generate an equipment health assessment index;

[0009] Establishing a dynamic impact model of abnormal equipment events on grid load fluctuations based on the equipment health evaluation indicators and association rules, wherein the association rules are generated by analyzing the coupling relationship between the equipment health change trend and the load curve mutation within the same time window;

[0010] The dynamic impact model is used to identify potential abnormal equipment, and logical mapping is performed based on the operating parameters of the potential abnormal equipment and the power grid topology structure of the area where the power equipment is located to generate fault propagation path weights. The power load forecast value is dynamically adjusted according to the fault propagation path weights to obtain the adjusted power load forecast value.

[0011] Optionally, the performing a fusion analysis on the vibration spectrum amplitude and the insulation aging index change rate to generate an equipment health assessment index includes:

[0012] Based on the exponential decay of the vibration spectrum amplitude according to the vibration monitoring time, the exponentially decayed vibration spectrum amplitude is logarithmically transformed to generate a vibration degradation component, wherein the vibration degradation component is used to characterize the cumulative degradation degree of the power equipment;

[0013] The insulation aging index change rate is linearly amplified and corrected according to the insulation monitoring temperature, and the amplified and corrected insulation aging index change rate is subjected to hyperbolic tangent transformation to generate an aging degradation component, which is used to characterize the nonlinear degradation trend of the surface insulation material of the power equipment;

[0014] The vibration degradation component and the aging degradation component are fused to generate an equipment health assessment index.

[0015] Optionally, the fusing the vibration degradation component and the aging degradation component to generate an equipment health assessment index includes:

[0016] When the aging degradation component exceeds the vibration degradation component for three consecutive sampling periods and the average of the differences exceeds a preset threshold, activating a temperature compensation flag;

[0017] When the temperature compensation flag is activated, the fusion weight is adjusted according to the inverse proportional constraint relationship between the vibration spectrum amplitude and the insulation aging index change rate, and the fusion weight includes the vibration weight and the aging weight;

[0018] taking a sum of a first weighted value and a second weighted value as a fusion component, wherein the first weighted value is a weighted value of the vibration degradation component and the vibration weight, and the second weighted value is a weighted value of the aging degradation component and the aging weight;

[0019] Performing moving average processing on the fused components to generate a health sequence;

[0020] By presetting the fault mode characteristic codes in the equipment historical health status database, high-frequency fluctuation components are filtered from the health sequence to obtain the equipment health evaluation index.

[0021] Optionally, establishing a dynamic impact model of abnormal equipment events on power grid load fluctuations based on the equipment health evaluation index and association rules includes:

[0022] Classifying multiple device abnormal events into corresponding abnormality levels according to the numerical range of the device health assessment index to obtain an abnormality level classification result;

[0023] calibrating the confidence of the abnormality level classification results based on the Bayesian probability model to obtain the posterior probability distribution of each abnormality level, and assigning a confidence parameter to each abnormality level according to the posterior probability distribution;

[0024] Based on the confidence parameter, dynamically correct the initial impact coefficient corresponding to the abnormality level to generate a corrected impact coefficient;

[0025] Using the modified impact coefficient as input, a dynamic modified weight is generated through a Bayesian optimization algorithm, wherein the dynamic modified weight is constrained by the load disturbance safety margin in the association rule;

[0026] Extracting a dynamic correction weight corresponding to the device abnormal event according to the timestamp of the device abnormal event within a sliding time window, performing a point-by-point convolution operation on the dynamic correction weight and the instantaneous change value of the power grid load fluctuation to generate a real-time impact component of the device abnormal event;

[0027] According to the recovery trend of the equipment health assessment indicator, a time decay factor is calculated using a Bayesian recursive estimation method, wherein the decay rate of the time decay factor is negatively correlated with the recovery trend;

[0028] generating a residual of the historical impact component based on the time attenuation factor, and superimposing the residual and the real-time impact component to generate a total impact;

[0029] The residuals of the predicted value and the measured value of the grid load fluctuation are calculated in real time to form a residual sequence. If the peak-to-peak value of the residual in the residual sequence is not within the preset steady-state error range, the dynamic correction weight is adaptively adjusted according to the residual sign, and the attenuation rate is corrected based on the total impact amount until the peak-to-peak value of the residual converges to the preset steady-state error range, thereby obtaining a dynamic impact model of equipment abnormal events on grid load fluctuations.

[0030] Optionally, within the sliding time window, extracting a dynamic correction weight corresponding to the device abnormal event according to the timestamp of the device abnormal event, performing a point-by-point convolution operation on the dynamic correction weight and the instantaneous change value of the power grid load fluctuation to generate a real-time impact component of the device abnormal event includes:

[0031] Determine the starting boundary and the ending boundary of the sliding time window according to the timestamp of the device abnormal event, and determine the length between the ending boundary and the starting boundary as the sliding time window length;

[0032] Within the sliding time window, aligning the occurrence time of the device abnormal event according to the timestamp, and extracting the dynamic correction weight corresponding to the device abnormal event;

[0033] Based on the time decay characteristics of the dynamic correction weight, construct a convolution kernel with a length equal to the sliding time window;

[0034] Extracting instantaneous change values ​​of grid load fluctuations from a preset grid load fluctuation database according to the sampling frequency of the grid load fluctuations;

[0035] Performing a point-by-point convolution operation on the instantaneous change value and the convolution kernel to obtain a convolution result, wherein each sampling point in the convolution result corresponds to the local impact intensity of the device abnormal event;

[0036] The convolution result is amplitude-scaled within a sliding time window, and the scaled convolution result is superimposed with the baseline component of the power grid load fluctuation according to the timestamp to generate a real-time impact component of the equipment abnormal event.

[0037] Optionally, the generating of the fault propagation path weight based on the logical mapping of the operating parameters of the potential abnormal device and the power grid topology of the area where the power device is located includes:

[0038] Mapping the power grid topology into a directed graph, wherein vertices in the directed graph represent power devices, vertex attributes include vertex operating parameters, and edges represent electrical connection directions and physical coupling relationships between power devices, and edge attributes include initial weights and electrical distance values;

[0039] Based on the initial weight and in combination with vertex operating parameters of vertices at both ends of the edge, the fault propagation probability of the edge is calculated by a nonlinear compression function;

[0040] The power equipment with abnormal equipment health evaluation indicators is regarded as a potential abnormal equipment. Starting from the potential abnormal equipment, the adjacent vertices are expanded hop by hop along the edge direction. When the cumulative number of hops equals the preset propagation order, the expansion is stopped to generate an electrical connection subgraph.

[0041] In the electrical connection subgraph, all edges are traversed along the edge direction, the fault propagation probability of each edge is multiplied, and the total electrical distance value is superimposed to generate a comprehensive propagation path weight;

[0042] The path whose comprehensive weight of the propagation path exceeds the preset dynamic threshold is used as the fault propagation path, and the comprehensive weight of the fault propagation path is used as the fault propagation path weight.

[0043] Optionally, in the electrical connection subgraph, traversing all edges along the edge direction, multiplying the fault propagation probability of each edge, and superimposing the total electrical distance value to generate a comprehensive propagation path weight, including:

[0044] In the electrical connection subgraph, starting from the starting point, a breadth-first traversal is performed along the edge direction, and a set of traversal paths without repeated vertices is recorded layer by layer, where each path consists of a vertex sequence and a corresponding edge sequence;

[0045] For each edge sequence, sequentially multiply the fault propagation probability value of each edge to generate a path propagation coefficient, and accumulate the electrical distance values ​​of all edges in the path to obtain a total electrical distance value;

[0046] The path propagation coefficient is combined with the total electrical distance value to calculate the comprehensive propagation path weight.

[0047] In a second aspect, the present application provides a power load forecasting system based on association rule analysis, comprising:

[0048] A generating module, configured to generate device status data based on the vibration spectrum data and insulation aging index of the power equipment, wherein the device status data includes the vibration spectrum amplitude and the insulation aging index change rate;

[0049] An analysis module, configured to perform a fusion analysis on the vibration spectrum amplitude and the insulation aging index change rate to generate an equipment health assessment index;

[0050] A construction module is used to establish a dynamic impact model of abnormal equipment events on power grid load fluctuations based on the equipment health evaluation indicators and association rules, wherein the association rules are generated by analyzing the coupling relationship between the equipment health change trend and the load curve mutation amount within the same time window;

[0051] An adjustment module is used to use the dynamic impact model to identify potential abnormal equipment, perform logical mapping based on the operating parameters of the potential abnormal equipment and the power grid topology structure of the area where the power equipment is located, generate fault propagation path weights, and dynamically adjust the power load forecast value according to the fault propagation path weights to obtain an adjusted power load forecast value.

[0052] In a third aspect, the present application provides a computing device comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a power load forecasting method based on association rule analysis as described in any one of the first aspects.

[0053] In a fourth aspect, the present application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a method for power load forecasting based on association rule analysis as described in any one of the first aspects.

[0054] In an embodiment of the present application, a power load forecasting method based on association rule analysis is provided, the method comprising: generating equipment status data based on vibration spectrum data and insulation aging index of power equipment, the equipment status data comprising vibration spectrum amplitude and insulation aging index change rate; performing fusion analysis on the vibration spectrum amplitude and the insulation aging index change rate to generate an equipment health assessment index; establishing a dynamic impact model of equipment abnormal events on grid load fluctuations based on the equipment health assessment index and association rules, the association rules being generated by analyzing the coupling relationship between the equipment health change trend and the load curve mutation amount within the same time window; identifying potential abnormal equipment using the dynamic impact model, performing logical mapping based on the operating parameters of the potential abnormal equipment and the grid topology structure of the area where the power equipment is located, generating a fault propagation path weight, and dynamically adjusting the power load forecast value based on the fault propagation path weight to obtain an adjusted power load forecast value.

[0055] Beneficial effects of this application:

[0056] This application breaks through the limitations of traditional single-indicator monitoring by integrating vibration spectrum and multi-dimensional parameters of insulation aging, achieving simultaneous quantitative characterization of equipment mechanical wear and insulation performance degradation status, and solving the status blind spot problem caused by data dimension fragmentation in existing solutions. A multi-source degradation feature coupling analysis method is adopted to eliminate the evaluation bias caused by the time-varying characteristics of the vibration signal and the difference in insulation aging rate, and overcome the defect of the traditional linear weighted fusion model that is insufficiently adaptable to nonlinear degradation processes. By exploring the spatiotemporal correlation between health trends and load mutations, a causal reasoning framework for equipment anomalies and grid disturbances is constructed, filling the technical gap that the static threshold alarm mechanism cannot quantify the transmission effect of equipment degradation on system-level load fluctuations. Based on the real-time correction of the prediction model based on the weight of the fault propagation path, the accurate mapping of equipment-level risks to system-level load anomalies under the constraints of the grid topology is achieved, breaking the vicious cycle problem of "prediction inaccuracy-fault diffusion" caused by the traditional prediction method ignoring the evolution of equipment health status.

[0057] Furthermore, the exponential decay-logarithmic transformation of the vibration spectrum amplitude is used to extract the mechanical cumulative degradation characteristics, and the temperature-corrected hyperbolic transformation of the insulation aging exponent is combined to capture the nonlinear degradation law of the insulation material, and construct the vibration / aging degradation component. A temperature compensation mechanism based on the persistence of exceeding the standard is designed, and the fusion weight is dynamically adjusted using inverse proportional constraints. The anti-interference health index is generated through moving average and feature filtering to achieve coupled characterization of the multi-physical field degradation process. In response to the three major pain points in the existing technology, namely the mismatch between the time-varying characteristics of vibration and insulation parameters, the misjudgment of degradation trends under temperature interference, and the false health alarm caused by high-frequency noise, the technical bottleneck of the traditional fixed-weight fusion model is broken through. Through the enhanced extraction of nonlinear features and the dynamic compensation mechanism, the recognition sensitivity of the composite degradation mode is significantly improved. Combined with the fluctuation suppression strategy driven by the feature code, the true health status of the equipment can still be stably characterized in a strong electromagnetic interference environment, providing a reliable decision-making basis for risk warning of equipment with a high proportion of new energy connected to the power grid.

[0058] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0060] Figure 1 A flowchart of a method for power load forecasting based on association rule analysis provided in an embodiment of the present application;

[0061] Figure 2 A schematic diagram of the structure of a power load forecasting system based on association rule analysis provided in an embodiment of the present application;

[0062] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0063] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0064] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 11, 12, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0065] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0066] To address the problems of delayed response to latent faults caused by the lack of a dynamic coupling mechanism between equipment health status and load fluctuations in the prior art, as well as load forecast compensation deviation caused by insufficient quantification of fault propagation path weights, the present invention provides a method and system for power load forecasting based on association rule analysis. The method adopts the following concept: based on the physical characteristics of multi-source equipment status monitoring data, the mechanical cumulative degradation is first quantified through exponential decay processing and logarithmic transformation of the vibration spectrum amplitude, and the nonlinear insulation degradation trend is characterized by temperature correction of the insulation aging index change rate and hyperbolic tangent transformation, thereby constructing a two-dimensional degradation component. A dynamic weight fusion mechanism is then designed to activate the inverse proportional constraint to adjust the fusion weight when the aging component is persistently abnormal. The moving average and historical fault feature library are combined to filter out noise interference and generate a health index. Subsequently, association rules are used to explore the spatiotemporal coupling relationship between equipment health changes and load mutations, and a dynamic impact model is established to identify abnormal equipment. Finally, the fault propagation path weights are mapped based on the power grid topology structure, and closed-loop correction of the load forecast value is achieved through weighted compensation, forming a full-chain analysis framework of "equipment degradation, fault propagation, and load fluctuation."

[0067] Figure 1 A flowchart of a power load forecasting method based on association rule analysis is provided in an embodiment of the present application, such as Figure 1 As shown, the method includes:

[0068] S11. Generate equipment status data based on the vibration spectrum data and insulation aging index of the power equipment, where the equipment status data includes the vibration spectrum amplitude and the insulation aging index change rate.

[0069] Vibration spectrum data is the frequency domain amplitude distribution obtained by collecting the mechanical vibration signal of the equipment through a vibration sensor and performing a Fourier transform. The insulation aging index is a quantitative indicator reflecting the degree of deterioration of the equipment's insulation material, calculated using the dielectric loss tangent or polarization index. The vibration spectrum amplitude refers to the vibration signal generated by power equipment during operation, which is decomposed through spectral analysis to determine the amplitude value corresponding to each frequency component. The insulation aging index change rate refers to the rate at which the degree of aging of the insulation material of power equipment changes over time and is usually calculated based on the insulation aging index.

[0070] In an embodiment of the present application, vibration spectrum data of the power equipment is first collected, and the insulation aging index is monitored at the same time; the amplitude of the characteristic frequency point of the vibration spectrum data is extracted as the vibration spectrum amplitude, and the change rate of the insulation aging index is calculated according to the time window, and finally the equipment status data is generated, which includes two key parameters: vibration spectrum amplitude and insulation aging index change rate.

[0071] S12. Perform a fusion analysis on the vibration spectrum amplitude and the insulation aging index change rate to generate an equipment health assessment index.

[0072] Among them, the equipment health assessment index is a weighted comprehensive score that integrates vibration and insulation aging parameters, which is used to quantify the equipment operating status.

[0073] In this embodiment, a multi-source data fusion algorithm is used to jointly analyze the vibration spectrum amplitude and the rate of change of the insulation aging index. Weights are assigned based on the contribution of these two parameters, and a linear weighted evaluation index is used to calculate the equipment health assessment index. Lower health indicators indicate more severe equipment degradation, providing a quantitative basis for subsequent dynamic impact modeling.

[0074] S13. Based on the equipment health evaluation indicators and association rules, a dynamic impact model of equipment abnormal events on power grid load fluctuations is established. The association rules are generated by analyzing the coupling relationship between the equipment health change trend and the load curve mutation within the same time window.

[0075] Association rules are probabilistic rules that describe the correlation between changes in equipment health and load fluctuations. The dynamic impact model is a machine learning-based model that takes equipment health indicators as input and outputs predicted impact values ​​on grid load fluctuations. Load curve mutations are the amount by which load values ​​exceed historical fluctuation ranges over a short period of time, typically measured as multiples of the standard deviation. Equipment abnormal events occur when equipment health assessment indicators exceed preset thresholds or show a continuous deterioration trend, indicating that the equipment has entered a potential failure or performance degradation phase but has not yet completely failed.

[0076] In an embodiment of the present application, an association rule is trained based on historical data to establish a dynamic impact model: the health assessment index is used as input and the load fluctuation amount is used as output. The dynamic coupling relationship between the two is fitted through time series regression, and the probability impact matrix of abnormal equipment events on grid load fluctuations is output to quantify the chain effect of abnormal events on the grid.

[0077] S14. Use the dynamic impact model to identify potential abnormal equipment, perform logical mapping based on the operating parameters of the potential abnormal equipment and the power grid topology structure of the area where the power equipment is located, generate fault propagation path weights, and dynamically adjust the power load forecast value according to the fault propagation path weights to obtain the adjusted power load forecast value.

[0078] The fault propagation path weight is a probability weight calculated based on the grid topology and device parameters. The grid topology represents the physical and logical connections between device nodes within the grid, represented as a graph. The power load forecast is an estimate of power demand over a specific time period in the future, typically measured in power or energy. In step S14, this concept is expanded to a dynamic forecast that incorporates the influence of device health and grid topology.

[0079] In an embodiment of the present application, after using the dynamic impact model to identify potential abnormal equipment, the fault propagation path weight is calculated by combining the equipment operating parameters and the grid topology structure through the graph propagation algorithm, and the load forecast value is dynamically adjusted: the load node forecast value corresponding to the path with a weight higher than the threshold is reduced by a correction coefficient, and finally the adjusted power load forecast value is generated.

[0080] Here's a specific example:

[0081] A transformer at a substation collects time-domain vibration signals using a vibration sensor. Fast Fourier transform (FFT) is used to extract the vibration spectrum amplitude of 0.9 mm / s at 800 Hz. The transformer also monitors its insulation aging index and calculates the aging index change rate of 15% within the current cycle. This generates device status data, which includes two parameters: the vibration spectrum amplitude of 0.9 mm / s and the insulation aging index change rate of 15%. The vibration spectrum amplitude and insulation aging index change rate are then fed into a fusion model using the entropy weighting method and the near-ideal solution ranking method. The entropy weighting method is used to calculate the weights, resulting in a vibration amplitude weight of 0.6 and an aging change rate weight of 0.4. The proximity of these two parameters to the entropy weighting method and the near-ideal solution is then calculated, resulting in a weighted device health assessment index of 0.3. Association rules trained on historical data trigger an early warning. A dynamic impact model analyzes the correlation between the transformer health index of 0.3 and the current grid load curve, predicting an 82% probability that an abnormal event will cause a sudden increase in downstream load. This generates a load fluctuation risk probability matrix. After the system identified the transformer as a potentially abnormal device, it combined the operating parameters and the regional power grid topology, such as the transformer's downstream connected distribution stations A, B, and C, and calculated the fault propagation path weight through a graph propagation algorithm. The path weight coefficients were calculated based on the line impedance and load rate to be 0.7 for station A, 0.5 for station B, and 0.3 for station C. The load forecast value was dynamically adjusted: a weight of 0.7 corresponds to a 35% reduction in the forecast value for station A, a 25% reduction for station B, and a 15% reduction for station C. The final output was the adjusted power load forecast value to avoid cascading power outages caused by transformer overload.

[0082] By executing S11 to S14, the embodiment of the present application generates equipment health indicators through multi-dimensional fusion of vibration and insulation aging data, constructs a fault propagation model based on the power grid load fluctuation law, realizes accurate identification of abnormal equipment and dynamic prediction of fault paths, and finally corrects the load prediction value in real time based on the propagation weight, significantly improving the power grid risk warning capability and load scheduling reliability, and reducing power outage losses caused by equipment abnormalities.

[0083] In a possible embodiment, S11, performing a fusion analysis on the vibration spectrum amplitude and the insulation aging index change rate to generate an equipment health assessment index includes:

[0084] Step 111 : exponentially decay the vibration spectrum amplitude according to the vibration monitoring time, logarithmically transform the exponentially decayed vibration spectrum amplitude to generate a vibration degradation component, which is used to characterize the cumulative degradation degree of the power equipment.

[0085] The vibration monitoring duration is the cumulative vibration data collection time of the power equipment since it was put into operation, usually in hours. Exponential decay is based on the natural exponential function e (-λt)The mathematical process of applying time-decay correction to the vibration amplitude. The vibration degradation component is a quantitative indicator of the cumulative damage of the mechanical structure after time decay and logarithmic transformation.

[0086] In this embodiment, the vibration monitoring duration is divided by a preset time constant to obtain an attenuation index. The original value of the vibration spectrum amplitude is multiplied by the negative attenuation exponent e to achieve exponential attenuation of the amplitude. The attenuated amplitude is then transformed by a natural logarithm. This is done by adding 1 to the amplitude of each frequency point and taking the ln value. This eliminates dimensional differences and compresses the data range, ultimately generating a vibration degradation component. This component reflects the cumulative effects of mechanical wear by retaining the low-frequency energy ratio.

[0087] Step 112: linearly amplify and correct the insulation aging index change rate according to the insulation monitoring temperature, and perform hyperbolic tangent transformation on the amplified and corrected insulation aging index change rate to generate an aging degradation component. The aging degradation component is used to characterize the nonlinear degradation trend of the surface insulation material of the power equipment.

[0088] Insulation monitoring temperature refers to the real-time surface temperature of the insulation material, a key environmental parameter that influences the aging rate. Linear amplification correction is an adjustment method that proportionally amplifies the aging rate based on temperature deviation. The aging degradation component is an indicator that characterizes the nonlinear degradation of the insulation material after temperature correction and hyperbolic tangent transformation. The hyperbolic tangent transformation is a mathematical function called tanh(x), which compresses the input value to the (-1, 1) interval while preserving nonlinear characteristics.

[0089] In an embodiment of the present application, the insulation aging index change rate and the real-time insulation monitoring temperature are first obtained through a sensor. The temperature value is subtracted from the reference temperature of 25°C and multiplied by the temperature correction coefficient of 0.15 to linearly amplify the aging index change rate. For example, when the temperature is 60°C, the rate of change will be amplified to 1.525 times the original value. The amplified value is then input into the hyperbolic tangent function tanh(x), and the rate of change is mapped to the (-1,1) interval to eliminate the influence of extreme outliers. The aging degradation component finally generated characterizes the nonlinear acceleration characteristics of the insulation material performance degradation through the S-shaped curve characteristics.

[0090] Step 113: Fusing the vibration degradation component and the aging degradation component to generate an equipment health assessment index.

[0091] In this embodiment of the present invention, an entropy weighting method and a near-ideal solution sorting fusion algorithm are used to process the vibration degradation component and the aging degradation component. The entropy weighting method is used to calculate the weights of the two components. Then, based on the near-ideal solution sorting model, the proximity of each component to the ideal solution is calculated. Finally, the weighted summation is used to generate an equipment health assessment index. This index simultaneously reflects the characteristics of mechanical cumulative degradation and nonlinear insulation degradation, providing a unified quantitative basis for subsequent fault prediction.

[0092] Continuing with the above example, based on the currently collected 800Hz frequency point vibration spectrum amplitude of 0.9mm / s, combined with the 50-hour continuous vibration monitoring time, an exponential decay operation is performed according to the time decay coefficient of 0.02 to obtain the vibration amplitude after attenuation of 0.9×e （-0.02×50） =0.331mm / s, and then generate the vibration degradation component through natural logarithm transformation ln(0.331+1)=0.287 to quantitatively reflect the cumulative effect of mechanical wear of the equipment bearings and windings; at the same time, based on the insulation aging index change rate of 15%, combined with the current insulation material temperature monitoring value of 90℃, temperature linear correction is performed, and the change rate is amplified to 15%×(90 / 80)=16.875%, and then the aging degradation component is generated through the hyperbolic tangent function tanh(16.875%)=0.167 to characterize the high temperature The nonlinear degradation characteristics of thermal decomposition of insulating paper under normal environment are analyzed; finally, the entropy weight method is used to determine the vibration component weight of 0.6 and the aging component weight of 0.4, and the weighted values ​​of 0.287×0.6=0.172 and 0.167×0.4=0.067 are calculated respectively. After superposition, the comprehensive degradation degree of 0.239 is obtained. The Euclidean distance between the two types of parameters and the positive and negative ideal solutions is calculated by combining the approximate ideal solution sorting method. After normalization, the output equipment health assessment index is 0.3, which accurately reflects the coupling effect of mechanical vibration degradation and insulation aging process.

[0093] By executing steps 111 to 113, the embodiment of the present application extracts the mechanical cumulative degradation characteristics through the time-domain attenuation correction and logarithmic transformation of the vibration amplitude, captures the nonlinear degradation law of insulation in combination with the temperature-compensated hyperbolic tangent transformation, and finally uses the entropy weight method-approximate ideal solution sorting method to generate a comprehensive health index, thereby achieving a multi-dimensional and accurate evaluation of the mechanical and insulation status of the equipment, providing a scientific basis for the formulation of differentiated operation and maintenance strategies, and improving the accuracy compared with the traditional single parameter evaluation method.

[0094] In a possible embodiment, step 113 of fusing the vibration degradation component and the aging degradation component to generate an equipment health assessment index includes:

[0095] Step a1: When the aging degradation component exceeds the vibration degradation component for three consecutive sampling periods and the average of the differences exceeds a preset threshold, the temperature compensation flag is activated.

[0096] The sampling period is the interval between data collection and is used to periodically update the degradation component. The difference mean is the arithmetic mean of the differences between the aging component and the vibration component over multiple consecutive cycles. The temperature compensation flag is a binary status signal indicating a significant temperature impact on insulation aging.

[0097] In this embodiment of the application, the system monitors the relationship between the vibration degradation component and the aging degradation component in real time. When the aging degradation component exceeds the vibration degradation component for three consecutive sampling periods, and the average difference between the two exceeds a preset threshold, a temperature compensation flag is activated. This flag indicates that insulation aging is significantly affected by temperature and requires adjustment of subsequent fusion weights.

[0098] Step a2: When the temperature compensation flag is activated, the fusion weight is adjusted according to the inverse proportional constraint relationship between the vibration spectrum amplitude and the insulation aging index change rate. The fusion weight includes the vibration weight and the aging weight.

[0099] The inverse proportional constraint relationship refers to the distribution rule that the vibration and aging weights are inversely proportional to each other, ensuring that the sum of the weights is 1. The inverse proportional constraint relationship is defined as: , ,in, is the vibration degradation component, is the aging degradation component, vibration weight With aging weight The sum of is 1, that is .

[0100] In this embodiment of the present application, if the temperature compensation flag is activated, the fusion weight is adjusted according to the inverse proportional constraint relationship: vibration weight = aging degradation component / (vibration degradation component + aging degradation component), aging weight = vibration degradation component / (vibration degradation component + aging degradation component). For example, if the vibration degradation component is 1.2 and the aging degradation component is 1.5, then the vibration weight = 1.5 / (1.2 + 1.5) = 0.56, and the aging weight = 0.44, thereby reducing the interference of temperature anomalies on insulation aging.

[0101] Step a3: taking the sum of the first weighted value and the second weighted value as the fusion component, where the first weighted value is the weighted value of the vibration degradation component and the vibration weight, and the second weighted value is the weighted value of the aging degradation component and the aging weight.

[0102] The first weighted value is the product of the vibration degradation component and the vibration weight, reflecting the contribution of mechanical degradation. The second weighted value is the product of the aging degradation component and the aging weight, reflecting the contribution of insulation degradation. The fusion component is the sum of the first and second weighted values, representing the overall degradation level of the equipment.

[0103] In an embodiment of the present application, when calculating the fusion component, the vibration degradation component is multiplied by the vibration weight to obtain a first weighted value, and the aging degradation component is multiplied by the aging weight to obtain a second weighted value. The two are added together to obtain a fusion component, which comprehensively represents the superposition effect of mechanical and insulation degradation.

[0104] Step a4: Perform moving average processing on the fused components to generate a health sequence.

[0105] Moving average processing is a filtering method that smoothes data by calculating the mean of a sequence over a sliding window. A health series is continuous state trend data generated by smoothing the time dimension of equipment health assessment indicators.

[0106] In the embodiment of the present application, the fusion component is subjected to moving average processing: the fusion component of 5 consecutive sampling periods is taken, and the arithmetic mean is calculated as the current health sequence value, thereby eliminating short-term fluctuation interference and generating a smooth health time series.

[0107] Step a5: By presetting the fault mode characteristic codes in the historical health status database of the equipment, high-frequency fluctuation components are filtered from the health sequence to obtain the equipment health evaluation index.

[0108] Among them, the fault mode characteristic code refers to the frequency or amplitude characteristic code corresponding to a specific fault in the health sequence.

[0109] In an embodiment of the present invention, by presetting the historical health status database of the equipment, the health sequence is subjected to wavelet transform to extract the high-frequency components, the noise unrelated to the fault is filtered out, the low-frequency effective signal is retained, and finally the equipment health evaluation index is generated to directly match the fault warning threshold.

[0110] Continuing with the above example, a transformer in a substation monitored aging degradation components of 0.172, 0.185, and 0.198 in three consecutive sampling periods, and vibration degradation components of 0.165, 0.160, and 0.155, respectively. The average difference between the two was 0.015, and the preset threshold was 0.01, triggering the activation of the temperature compensation flag. Based on the inverse proportional relationship between the vibration spectrum amplitude of 0.9 mm / s and the insulation aging index change rate of 15%, the vibration amplitude weight formula is: vibration amplitude weight = aging change rate weight × aging change rate / vibration amplitude. The original entropy weights of 0.6 and 0.4 are dynamically adjusted to a vibration weight of 0.4 and an aging weight of 0.6. Subsequently, The vibration degradation component (0.287×0.4=0.115) and the aging degradation component (0.167×0.6=0.100) are calculated and superimposed to form a fusion component (0.215). The sequence [0.215, 0.220, 0.225] is smoothed using a three-day moving average window, and the output health sequence mean is 0.220. Finally, the historical fault mode feature code is called, and the fluctuation components with a frequency higher than 0.5 Hz in the health sequence are filtered out through wavelet transform. The low-frequency trend component (0.208) is extracted as the final equipment health assessment indicator. The transformer health parameters in the dynamic impact model are simultaneously updated to optimize the load surge probability prediction and path weight coefficient calculation.

[0111] By executing steps a1 to a5, the embodiment of the present application accurately distinguishes the dominant factors of mechanical and insulation degradation through dynamically activating the temperature compensation mechanism and inverse proportional adjustment of weights; combining moving average and fault feature filtering, it effectively suppresses noise interference and extracts key fault signals, so that the health assessment indicator has environmental adaptability, anti-interference and fault directionality at the same time, reducing the false alarm rate compared with the traditional static weight method.

[0112] In a possible embodiment, S13, based on the equipment health evaluation index and the association rules, a dynamic impact model of equipment abnormal events on power grid load fluctuations is established, including:

[0113] Step 131: Classify multiple device abnormal events into corresponding abnormality levels based on the numerical range of the device health evaluation index to obtain an abnormality level classification result. The abnormality level classification result is a device abnormality severity label classified according to the health index.

[0114] In this embodiment, we first obtain the numerical range of the device health assessment indicator. We then categorize historical device abnormal events into corresponding ranges based on the health values ​​at the time of their triggering, generating an abnormality level classification result. Specifically, we set a threshold divider for each range. When an abnormal event occurs, its associated health value is compared with a preset threshold. If the health value is 55, the abnormality level is determined to be severe. This result includes statistics and distribution characteristics for each abnormality level.

[0115] Step 132: calibrate the confidence of the abnormality level classification results based on the Bayesian probability model to obtain the posterior probability distribution of each abnormality level, and configure the confidence parameters for each abnormality level according to the posterior probability distribution.

[0116] The posterior probability distribution refers to the probability distribution calculated by combining prior knowledge and new evidence in a Bayesian model. The confidence parameter is a quantitative parameter (0-1) that represents the reliability of the anomaly level judgment.

[0117] In this embodiment, the confidence level of the abnormality classification results is calibrated based on a Bayesian probability model. First, a prior probability distribution is established. Combined with the frequency of abnormal events monitored in real time, the posterior probability distribution is calculated using the Bayesian formula: posterior probability = (prior probability × likelihood) / evidence factor. For example, the posterior probability of a severe abnormality is revised from 30% to 32.5%. Based on the posterior probability, a confidence parameter is assigned to each level. The parameter value is the product of the posterior probability and the historical accuracy rate.

[0118] Step 133: Based on the confidence parameter, dynamically modify the initial impact coefficient corresponding to the abnormality level to generate a modified impact coefficient. The modified impact coefficient refers to the impact intensity coefficient of the abnormal event on the power grid load after confidence calibration.

[0119] In this embodiment, the initial impact coefficient corresponding to the anomaly level is dynamically modified based on the confidence parameter. A linear interpolation algorithm is used: the modified impact coefficient = initial coefficient × (1 + confidence parameter × adjustment factor). For example, an initial coefficient of 1.2 for a severe anomaly is modified to 1.2 × (1 + 0.92 × 0.15) = 1.365 at a confidence level of 0.92. The adjustment factor of 0.15 is set based on expert experience to ensure that the correction coefficient does not exceed the tolerance limit of the equipment.

[0120] Step 134: Using the corrected impact coefficient as input, a Bayesian optimization algorithm is used to generate a dynamic correction weight. The dynamic correction weight is subject to the load disturbance safety margin in the association rule. The dynamic correction weight is the abnormal event impact weight generated by optimizing under the safety margin constraint.

[0121] In this embodiment of the present invention, a Bayesian optimization algorithm generates dynamic correction weights using the corrected impact coefficients as input. The objective function is defined as minimizing grid load forecast error, with the constraints being load disturbance safety margins. A surrogate model is constructed using Gaussian process regression. The objective function is calculated for each candidate weight value at each iteration, and the weight value that reduces the objective function while satisfying the constraints is selected. The final output of the dynamic correction weights is associated with the potential impact of device anomalies on the grid.

[0122] Step 135: Within the sliding time window, extract the dynamic correction weight corresponding to the device abnormal event according to the timestamp of the device abnormal event, perform point-by-point convolution operation on the dynamic correction weight and the instantaneous change value of the grid load fluctuation, and generate the real-time impact component of the device abnormal event.

[0123] In this embodiment of the present invention, within a sliding time window, the corresponding dynamic correction weights are extracted based on the timestamp of the device abnormal event. The weight sequence is convolved point-by-point with the instantaneous change value of the grid load fluctuation: the real-time impact component = ∑(weight_i × ΔP_t-i}), where i is the time offset within the sliding window and ΔP is the load change rate. For example, the impact at the current time t = 0.73 × ΔP_t + 0.68 × ΔP_{t-1} + ..., ultimately generating a real-time superposition effect reflecting the abnormal event on the load fluctuation.

[0124] Step 136: Based on the recovery trend of the equipment health evaluation indicator, a Bayesian recursive estimation method is used to calculate the time decay factor. The decay rate of the time decay factor is negatively correlated with the recovery trend.

[0125] Among them, the time decay factor is a parameter that reflects the decay rate of historical influence over time and is negatively correlated with the health recovery trend.

[0126] In this embodiment of the present application, a Bayesian recursive estimation method is used to calculate the time decay factor based on the recovery trend of the device health assessment indicator. A state-space model is constructed, and the health recovery rate is used as the observation variable. The decay factor is updated through Kalman filtering: decay rate = 1 / (1 + recovery rate × time constant). For example, a recovery rate of 3 minutes / day corresponds to a decay factor decrease of 0.25 per day, ensuring that historical impacts decay faster during rapid recovery.

[0127] Step 137 : Generate a residual of the historical impact component based on the time attenuation factor, and superimpose the residual and the real-time impact component to generate a total impact.

[0128] Among them, the total impact refers to the superposition of historical residual impact and real-time impact.

[0129] In the embodiment of the present application, the residual of the historical impact component is generated based on the time decay factor. Residual = historical impact component × e (-衰减速率×Δt) , where Δt is the time difference from the event to the present. The residual and real-time impact components are superimposed: Total Impact = Residual + Real-time Impact. For example, a fault's residual of 0.35 and its real-time impact of 0.62 add up to a total impact of 0.97, representing the sustained, comprehensive impact of the anomaly on the grid load.

[0130] Step 138: Calculate the residual between the predicted value and the measured value of the grid load fluctuation in real time to form a residual sequence. If the peak-to-peak value of the residual in the residual sequence is not within the preset steady-state error range, the dynamic correction weight is adaptively adjusted according to the residual sign, and the attenuation rate is corrected based on the total impact amount until the peak-to-peak value of the residual converges to the preset steady-state error range, thereby obtaining a dynamic impact model of equipment abnormal events on grid load fluctuations.

[0131] Among them, the residual sequence is a sequence composed of the differences between the load forecast value and the measured value sorted by time.

[0132] In an embodiment of the present invention, the residuals between the predicted value and the measured value of the grid load fluctuation are calculated in real time to form a residual sequence. If the peak-to-peak value of the residual exceeds the preset steady-state error range, the dynamic correction weight is adaptively adjusted according to the sign of the residual: the weight increases for positive residuals, for example, from 0.73 to 0.78, and decreases for negative residuals, for example, from 0.73 to 0.68). At the same time, the attenuation rate is corrected based on the total impact (for example, the attenuation rate decreases by 5% for every 0.1 increase in the total impact), until the peak-to-peak value of the residual converges to the allowable range, and finally the dynamic impact model is output.

[0133] Continuing with the above example, a transformer in a substation is judged to be a level 3 abnormality based on the numerical interval of the equipment health assessment index of 0.3, corresponding to the health interval of 0.2~0.4, and a "moderate degradation" abnormality level label is generated; the Bayesian probability model is used to combine the actual occurrence rate of level 3 abnormalities in historical fault data with the degree of deviation from the current health index, and the posterior probability distribution is calculated to obtain a confidence level of 85% for the level 3 abnormality, and a confidence parameter of 0.85 is assigned to it; the initial impact coefficient of 0.82 is dynamically corrected according to the confidence parameter, and the revised impact coefficient is generated using the confidence weighting formula 0.82×0.85=0.697; the revised impact coefficient is obtained through the Bayesian optimization algorithm The optimal weight is iteratively solved under the load disturbance safety boundary constraint with the number 0.697 as input, and a dynamic correction weight of 0.75 is generated. The timestamps of three consecutive abnormal events triggered by the transformer are extracted within a 30-minute sliding time window. The dynamic correction weight sequence [0.75, 0.73, 0.70] is convolved with the grid load fluctuation values ​​[0.9%, 1.2%, 0.8%] at the corresponding moments, and the real-time impact component is output as 0.75×0.9%+0.73×1.2%+0.70×0.8%=2.35%. Based on the trend of the health index rising from 0.25 to 0.3 in the past 6 hours, the time attenuation factor e is calculated using Bayesian recursive estimation. （-0.008×t） , t is the time, and the historical impact component 2.1% 3 hours ago is attenuated into a residual amount of: 2.1%×e （-0.008×3） =2.05%; finally, the residual amount of 2.05% and the real-time impact component of 2.35% are superimposed to generate a total impact of 4.4%, which is updated to the load fluctuation risk probability matrix. This drives the secondary revision of the predicted values ​​of distribution stations A, B, and C, and realizes the quantitative tracking of the impact of abnormal events throughout the entire life cycle.

[0134] By executing steps 131 to 138, the embodiment of the present application realizes refined modeling of the impact of equipment abnormal events on grid load fluctuations through dynamic calibration of abnormality levels, Bayesian optimization weight allocation and recursive estimation of attenuation factors; combined with the residual feedback adaptive adjustment mechanism, it ensures that the model tracks load changes in real time and converges to a stable state, reducing the prediction error by more than 40% compared with the traditional static model, significantly improving the robustness of grid dispatching decisions.

[0135] In one possible embodiment, step 135 extracts a dynamic correction weight corresponding to the device abnormal event according to the timestamp of the device abnormal event within the sliding time window, performs a point-by-point convolution operation on the dynamic correction weight and the instantaneous change value of the grid load fluctuation, and generates a real-time impact component of the device abnormal event, including:

[0136] Step b1: Determine the start and end boundaries of the sliding time window according to the timestamp of the device abnormal event, and determine the length between the end boundary and the start boundary as the sliding time window length.

[0137] The sliding time window length is the time span from the starting boundary to the ending boundary, which is used to limit the scope of data analysis.

[0138] In this embodiment, the boundaries of the sliding time window are first determined based on the timestamp of the device abnormal event. The starting boundary is set to a preset time period before the event, and the ending boundary is set to the event time. The difference between the two is the sliding time window length. In specific implementation, if the event occurs at 2:30 PM, the starting boundary is 2:30 PM of the previous day, and the ending boundary is the current 2:30 PM. The window length is calculated as the number of milliseconds from the ending boundary timestamp minus the starting boundary timestamp, ultimately obtaining the precise time period.

[0139] Step b2: Within the sliding time window, align the occurrence time of the device abnormal event according to the timestamp, and extract the dynamic correction weight corresponding to the device abnormal event.

[0140] In this embodiment, within a defined sliding time window, the occurrence time of each device abnormality event is aligned with the time axis within the window. Using a timestamp matching algorithm, the weight values ​​for the corresponding times are extracted from the dynamically corrected weight database. For example, if three abnormal events occur within the window at 2:00 PM, 8:00 PM, and 10:00 AM the following day, the weights corresponding to these three time points (0.75, 0.68, and 0.82) are extracted to form a chronologically ordered weight sequence.

[0141] Step b3: Based on the time decay characteristics of the dynamically modified weights, a convolution kernel with a length equal to the sliding time window is constructed. The convolution kernel, based on the weight sequence constructed based on the weight decay characteristics, is used to extract the time correlation characteristics of abnormal events.

[0142] In the embodiment of the present application, based on the time decay characteristics of the dynamic correction weight, a convolution kernel with the same length as the sliding time window is constructed. The specific process is: the window length is divided into 1440 points according to the sampling interval, and the convolution kernel value of each point i = weight_i × e (-衰减系数×i) For example, when the attenuation coefficient is set to 0.001, the kernel value at the 720th minute = 0.68×e (-0.001×720) =0.68×0.487≈0.331, and finally a 1440-dimensional attenuation convolution kernel is generated.

[0143] Step b4: extracting the instantaneous change value of the grid load fluctuation from the preset grid load fluctuation database according to the sampling frequency of the grid load fluctuation, wherein the instantaneous change value is the fluctuation amount of the grid load at a single sampling moment.

[0144] In this embodiment, instantaneous load change values ​​within a sliding time window are extracted from a preset database based on the sampling frequency of grid load fluctuations. A timestamp range query is performed to obtain a sequence of minute-by-minute load change rates (ΔP). For example, a 24-hour window contains 1440 ΔP values, forming the array [ΔP_1, ΔP_2, …, ΔP_1440].

[0145] Step b5: Perform a point-by-point convolution operation on the instantaneous change value and the convolution kernel to obtain a convolution result. Each sampling point in the convolution result corresponds to the local impact intensity of the device abnormal event.

[0146] Among them, the local impact intensity is the contribution of an abnormal event to the load fluctuation at a specific moment, which is obtained through convolution calculation.

[0147] In the embodiment of the present application, the extracted instantaneous change value is convolved with the generated convolution kernel point by point. For example, when k=100, the impact strength = convolution kernel_1×ΔP_100 + convolution kernel_2×ΔP_99 +…+ convolution kernel_1440×ΔP_1340, and 1440 local impact strength values ​​are output.

[0148] Step b6: perform amplitude scaling on the convolution result within the sliding time window, and superimpose the scaled convolution result with the baseline component of the grid load fluctuation according to the timestamp to generate the real-time impact component of the equipment abnormal event.

[0149] The baseline component refers to the baseline value of the grid load when there are no abnormal events, usually the historical average value for the same period. Amplitude scaling is the operation of adjusting the convolution result to match the actual load dimension.

[0150] In this embodiment, the convolution result is amplitude-scaled: Scaling factor = Maximum allowable load deviation / Maximum convolution result value. This scaled value is then overlaid with the baseline component by timestamp: Real-time impact component = Baseline_k + Scaling_result_k. For example, if the baseline at 2:30 PM is 50 MW and the scaling effect is +3.2 MW, the final output is 53.2 MW.

[0151] Continuing with the above example, for three consecutive abnormal events triggered by a transformer in a substation, the starting boundary of the sliding time window is set to 09:00 and the ending boundary is set to 09:15, and the window length is determined to be 15 minutes. The timestamps in the window are aligned to extract the dynamic correction weight sequence [0.75, 0.73, 0.70], and an exponential decay convolution kernel [0.75×e {-0.02×0} =0.75, 0.73×e {-0.02×5} =0.66, 0.70×e {-0.02×10}=0.57]; synchronously, the instantaneous change values ​​of grid load fluctuations at the corresponding moments [0.9%, 1.2%, 0.8%] are extracted from the load database at a 5-minute sampling frequency, and the convolution kernel is convolved with the fluctuation value point by point to calculate 0.75×0.9%+0.66×1.2%+0.57×0.8%=2.19%; the amplitude is scaled according to the maximum fluctuation value of 1.2% in the window to obtain the adjusted convolution result of 2.19%×0.8=1.75%, which is superimposed on the baseline load curve to generate a real-time impact component of 50MW×1.75%=0.875MW. This drives the load forecast values ​​of distribution stations A, B, and C to be additionally adjusted downward by 0.875MW×0.7=0.613MW, 0.875MW×0.5=0.438MW, and 0.875MW×0.3=0.263MW at 09:15, achieving precise spatiotemporal quantification of the impact intensity of the abnormal event.

[0152] By executing steps b1 to b6, the embodiment of the present application accurately aligns abnormal events and load fluctuation data through a sliding time window, combines the time domain attenuation characteristics of the event impact quantified by the attenuated convolution kernel, and realizes dynamic modeling of the local impact of abnormal events on the grid load; after superimposing the baseline component, a high-precision real-time impact component is generated, so that the load forecast value synchronously reflects the abnormal status of the equipment, and the error is reduced by more than 25% compared with the traditional static correction method, which significantly improves the real-time and reliability of grid scheduling.

[0153] In a possible embodiment, S14, performing logical mapping based on the operating parameters of the potential abnormal device and the power grid topology structure of the area where the power device is located to generate the fault propagation path weight includes:

[0154] Step 141: Map the power grid topology into a directed graph, where vertices in the directed graph represent power devices, vertex attributes include vertex operating parameters, and edges represent electrical connection directions and physical coupling relationships between power devices, and edge attributes include initial weights and electrical distance values.

[0155] A directed graph is a graph structure that uses vertices and directional edges to represent the connection relationships between power grid devices. Vertex operating parameters are real-time status parameters of power devices, such as load factor, temperature, and health. Initial weights are path-based risk values ​​(0 to 1) preset based on device type and connection relationship. Electrical distance values ​​are normalized parameters that characterize the strength of electrical coupling between devices, such as the inverse impedance.

[0156] Step 142: Based on the initial weight and in combination with the vertex operating parameters of the vertices at both ends of the edge, the fault propagation probability of the edge is calculated using a nonlinear compression function.

[0157] The nonlinear compression function maps the input to the (0,1) interval. The fault propagation probability refers to the probability that a fault on a certain edge will propagate from one device to another in the power grid topology. The formula for the fault propagation probability is: , where k is the adjustment coefficient.

[0158] Step 143: The power equipment with abnormal equipment health evaluation indicators is regarded as a potential abnormal equipment. Starting from the potential abnormal equipment, the adjacent vertices are expanded hop by hop along the edge direction. When the cumulative number of hops is equal to the preset propagation order, the expansion is stopped to generate an electrical connection subgraph.

[0159] The propagation order is the maximum number of hops a fault can propagate from its starting point. The electrical connection subgraph is the local topology of devices and connections that contain potential fault propagation paths. The propagation path weight is a quantitative value of the path risk that combines the fault probability and electrical distance.

[0160] Step 144: In the electrical connection subgraph, traverse all edges along the edge direction, multiply the fault propagation probability of each edge, and add the total electrical distance value to generate a comprehensive propagation path weight.

[0161] Among them, the comprehensive weight of the propagation path is an indicator used to quantify the overall risk of a fault propagation path.

[0162] The following two factors are considered: 1. The path fault propagation probability is the product of the fault propagation probabilities of each edge in the propagation path, reflecting the cumulative probability of the fault spreading step by step along the path. 2. The total electrical distance value is the sum of the electrical distance values ​​of all edges in the path, which represents the electrical coupling strength of the path. The formula for calculating the comprehensive weight of the propagation path is: .

[0163] Step 145: The path whose comprehensive weight of the propagation path exceeds the preset dynamic threshold is used as the fault propagation path, and the comprehensive weight of the fault propagation path is used as the fault propagation path weight.

[0164] Among them, the dynamic threshold is the path risk screening threshold value adjusted according to the real-time status of the power grid.

[0165] Continuing with the above example, a substation maps the power grid topology into a directed graph, with vertices including transformers and distribution stations A: load rate 90%, impedance 0.2Ω, B: load rate 80%, impedance 0.3Ω, and C: load rate 70%, impedance 0.4Ω. The initial weights of the edges are set to 0.7 for station A, 0.5 for station B, and 0.3 for station C. The electrical distances are 5km, 8km, and 10km, respectively. The propagation probability is calculated using the sigmoid function based on the initial edge weights and vertex load rates. For example, the initial weight of the edge from the transformer to station A is 0.7 is superimposed on the load rate of station A at 90%, and the input is σ(0.7+0.9×0.1)=0.82, generating a propagation probability of 0.7×0.82=0.57. Starting from the transformer, the propagation order is expanded according to the preset 3-hop propagation order, forming a subgraph containing the paths: transformer-ADG, transformer-BE, and transformer-CF. Traversing the transformer-A and D-G paths, the propagation probability is multiplied by 0.57×0.6×0.5=0.17, and the electrical distance is superimposed on 5+3+2=10km to generate a comprehensive weight of 0.17×e {-0.1×10} =0.063; filter the paths whose comprehensive weight exceeds the dynamic threshold of 0.15, and retain the path from the transformer to station A, whose weight is 0.57×e {-0.1×5} =0.35, retain the path from the transformer to station B, and its weight is 0.5×e {-0.1×8} =0.23, and input the path weights 0.35 and 0.23 into the load forecasting model, driving the forecast value of station A to be reduced by an additional 35%×0.35 / 0.7=17.5%, and the forecast value of station B to be reduced by 25%×0.23 / 0.5=11.5%, thus achieving quantitative risk control of the fault propagation path.

[0166] By executing steps 141 to 145, the embodiment of the present application accurately depicts the device connection relationship through directed graph modeling, combines nonlinear probability calculation with path comprehensive weight evaluation, and realizes quantitative prediction of fault propagation path; the dynamic threshold screening mechanism can adapt to changes in the operating status of the power grid, and reduces the false alarm rate by 35% compared with the traditional fixed threshold method, providing a high-credibility basis for the formulation of fault isolation strategy.

[0167] In a possible embodiment, step 144 traverses all edges in the electrical connection subgraph along the edge direction, multiplies the fault propagation probabilities of each edge, and adds the total electrical distance values ​​to generate a comprehensive propagation path weight, including:

[0168] Step c1: In the electrical connection subgraph, starting from the starting point, perform breadth-first traversal along the edge direction, and record the set of traversal paths without repeated vertices layer by layer. Each path consists of a vertex sequence and a corresponding edge sequence.

[0169] Breadth-first traversal is a graph traversal algorithm that expands from a starting point to visit adjacent vertices layer by layer, ensuring that the path contains no duplicate vertices. A set of traversal paths without duplicate vertices is a set of paths in which each vertex appears only once, avoiding cyclic paths. The vertex sequence is the ordered list of vertices in a path. The edge sequence is the ordered list of edges in a path.

[0170] Step c2: For each edge sequence, sequentially multiply the fault propagation probability value of each edge to generate a path propagation coefficient, and accumulate the electrical distance values ​​of all edges in the path to obtain a total electrical distance value.

[0171] The path propagation coefficient is the product of the fault propagation probabilities on each edge of the path, reflecting the cumulative likelihood of fault spread. The total electrical distance is the cumulative sum of the electrical distance values ​​on each edge of the path, representing the physical coupling strength of the path.

[0172] Step c3: Combine the path propagation coefficient with the total electrical distance value to calculate the comprehensive weight of the propagation path.

[0173] Among them, the higher the value of the comprehensive weight of the propagation path, the greater the risk of path failure.

[0174] Continuing with the above example, a substation performs a breadth-first traversal in the electrical connection subgraph starting from the transformer. The first-level traversal path is from transformer to station A, transformer to station B, and transformer to station C. The second-level traversal path is extended from station A to station D, station B to station E, and station C to station F. The third-level traversal path is extended from station D to station G. For the edge sequence of the path transformer, A, and D to G, the fault propagation probability of each edge is multiplied in sequence: 0.57 (transformer-A) × 0.6 (AD) × 0.5 (DG) = 0.17, and the accumulated electrical distance is 5km + 3km + 2km = 10km. Based on the formula, the comprehensive weight = path propagation coefficient × e {-0.1×总电气距离} , calculated to be 0.17×e {-0.1×10} =0.063, and similarly process the path transformer, B to E, to obtain the weight 0.5×0.4×e {-0.1×8} =0.23×e {-0.8} =0.10, and screen out the paths from transformer to A and transformer to B with a comprehensive weight exceeding the threshold of 0.15. These paths are used as the dominant propagation paths to drive the load forecast value of station A to be reduced by an additional 0.35 / 0.7×35%=17.5%, and that of station B to be reduced by 0.23 / 0.5×25%=11.5%, thereby achieving quantitative assessment of the electrical coupling strength of the fault propagation paths and precise risk control.

[0175] By executing steps c1 to c3, the embodiment of the present application exhaustively enumerates all possible fault propagation paths through breadth-first traversal, combines probability multiplication with electrical distance to quantify path risk, and achieves a systematic evaluation of fault propagation paths. The comprehensive weight calculation takes into account both the fault diffusion probability and the physical coupling strength, reducing the misjudgment rate compared to traditional single-index methods and providing a reliable basis for accurately isolating fault paths.

[0176] Figure 2 A schematic diagram of the structure of a power load forecasting system based on association rule analysis provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the system includes:

[0177] The generating module 21 is used to generate equipment status data according to the vibration spectrum data and insulation aging index of the power equipment, where the equipment status data includes the vibration spectrum amplitude and the insulation aging index change rate.

[0178] The analysis module 22 is used to perform a fusion analysis on the vibration spectrum amplitude and the insulation aging index change rate to generate an equipment health assessment index.

[0179] Construction module 23 is used to establish a dynamic impact model of equipment abnormal events on power grid load fluctuations based on equipment health evaluation indicators and association rules. The association rules are generated by analyzing the coupling relationship between the equipment health change trend and the load curve mutation within the same time window.

[0180] The adjustment module 24 is used to identify potential abnormal equipment using a dynamic impact model, perform logical mapping based on the operating parameters of the potential abnormal equipment and the power grid topology structure of the area where the power equipment is located, generate fault propagation path weights, and dynamically adjust the power load forecast value according to the fault propagation path weights to obtain an adjusted power load forecast value.

[0181] Figure 2 The power load forecasting system based on association rule analysis can be executed Figure 1 The implementation principle and technical effects of the power load forecasting method based on association rule analysis described in the illustrated embodiment will not be elaborated on here. The specific manner in which each module and unit performs operations in the power load forecasting system based on association rule analysis in the above embodiment has been described in detail in the embodiment of the method and will not be elaborated on here.

[0182] In one possible design, Figure 2 The power load forecasting system based on association rule analysis of the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32 .

[0183] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0184] The processing component 32 is used to: generate equipment status data based on the vibration spectrum data and insulation aging index of the power equipment, and the equipment status data includes the vibration spectrum amplitude and the insulation aging index change rate. The vibration spectrum amplitude and the insulation aging index change rate are fused and analyzed to generate equipment health assessment indicators. Based on the equipment health assessment indicators and association rules, a dynamic impact model of equipment abnormal events on grid load fluctuations is established. The association rules are generated by analyzing the coupling relationship between the equipment health change trend and the load curve mutation amount within the same time window. Potential abnormal equipment is identified using the dynamic impact model, and logical mapping is performed based on the operating parameters of the potential abnormal equipment and the grid topology structure of the area where the power equipment is located to generate fault propagation path weights, and the power load forecast value is dynamically adjusted according to the fault propagation path weights to obtain the adjusted power load forecast value.

[0185] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0186] The storage component 31 is configured to store various types of data to support operations on the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as random access memory (RAM), static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0187] Of course, a computing device may also include other components, such as input / output interfaces, display components, and communication components. The input / output interfaces provide interfaces between the processing components and peripheral interface modules, which may be output devices, input devices, and so on. The communication components are configured to facilitate wired or wireless communication between the computing device and other devices. The computing device may be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device may refer to a cloud server, and the processing components, storage components, and so on may be basic server resources rented or purchased from the cloud computing platform.

[0188] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The embodiment shown is a method for power load forecasting based on association rule analysis.

[0189] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0190] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0191] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0192] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A power load forecasting method based on association rule analysis, characterized in that: include: Generating equipment status data according to the vibration spectrum data and insulation aging index of the power equipment, wherein the equipment status data includes the vibration spectrum amplitude and the insulation aging index change rate; Performing a fusion analysis on the vibration spectrum amplitude and the insulation aging index change rate to generate an equipment health assessment index; Establishing a dynamic impact model of abnormal equipment events on grid load fluctuations based on the equipment health evaluation indicators and association rules, wherein the association rules are generated by analyzing the coupling relationship between the equipment health change trend and the load curve mutation within the same time window; Identifying potential abnormal devices using the dynamic impact model, performing logical mapping based on operating parameters of the potential abnormal devices and the power grid topology of the area where the power devices are located, generating fault propagation path weights, and dynamically adjusting the power load forecast value based on the fault propagation path weights to obtain an adjusted power load forecast value; The generating of the fault propagation path weight based on the logical mapping of the operating parameters of the potential abnormal device and the power grid topology structure of the area where the power device is located includes: Mapping the power grid topology into a directed graph, wherein vertices in the directed graph represent power devices, vertex attributes include vertex operating parameters, and edges represent electrical connection directions and physical coupling relationships between power devices, and edge attributes include initial weights and electrical distance values; Based on the initial weight and in combination with vertex operating parameters of vertices at both ends of the edge, the fault propagation probability of the edge is calculated by a nonlinear compression function; The power equipment with abnormal equipment health evaluation indicators is regarded as a potential abnormal equipment. Starting from the potential abnormal equipment, the adjacent vertices are expanded hop by hop along the edge direction. When the cumulative number of hops equals the preset propagation order, the expansion is stopped to generate an electrical connection subgraph. In the electrical connection subgraph, all edges are traversed along the edge direction, the fault propagation probability of each edge is multiplied, and the total electrical distance value is superimposed to generate a comprehensive propagation path weight; The path whose comprehensive weight of the propagation path exceeds the preset dynamic threshold is used as the fault propagation path, and the comprehensive weight of the fault propagation path is used as the fault propagation path weight.

2. The method according to claim 1, characterized in that The fusion analysis of the vibration spectrum amplitude and the insulation aging index change rate to generate the equipment health assessment index includes: Based on the exponential decay of the vibration spectrum amplitude according to the vibration monitoring time, the exponentially decayed vibration spectrum amplitude is logarithmically transformed to generate a vibration degradation component, wherein the vibration degradation component is used to characterize the cumulative degradation degree of the power equipment; The insulation aging index change rate is linearly amplified and corrected according to the insulation monitoring temperature, and the amplified and corrected insulation aging index change rate is subjected to hyperbolic tangent transformation to generate an aging degradation component, which is used to characterize the nonlinear degradation trend of the surface insulation material of the power equipment; The vibration degradation component and the aging degradation component are fused to generate an equipment health assessment index.

3. The method according to claim 2, characterized in that The fusing of the vibration degradation component and the aging degradation component to generate an equipment health evaluation index includes: When the aging degradation component exceeds the vibration degradation component for three consecutive sampling periods and the average of the differences exceeds a preset threshold, activating a temperature compensation flag; When the temperature compensation flag is activated, the fusion weight is adjusted according to the inverse proportional constraint relationship between the vibration spectrum amplitude and the insulation aging index change rate, and the fusion weight includes the vibration weight and the aging weight; taking a sum of a first weighted value and a second weighted value as a fusion component, wherein the first weighted value is a weighted value of the vibration degradation component and the vibration weight, and the second weighted value is a weighted value of the aging degradation component and the aging weight; Performing moving average processing on the fused components to generate a health sequence; By presetting the fault mode characteristic codes in the equipment historical health status database, high-frequency fluctuation components are filtered from the health sequence to obtain the equipment health evaluation index.

4. The method according to claim 1, wherein The method of establishing a dynamic impact model of abnormal equipment events on power grid load fluctuations based on the equipment health evaluation indicators and association rules includes: Classifying multiple device abnormal events into corresponding abnormality levels according to the numerical range of the device health assessment index to obtain an abnormality level classification result; calibrating the confidence of the abnormality level classification results based on the Bayesian probability model to obtain the posterior probability distribution of each abnormality level, and assigning a confidence parameter to each abnormality level according to the posterior probability distribution; Based on the confidence parameter, dynamically correct the initial impact coefficient corresponding to the abnormality level to generate a corrected impact coefficient; Using the modified impact coefficient as input, a dynamic modified weight is generated through a Bayesian optimization algorithm, wherein the dynamic modified weight is constrained by the load disturbance safety margin in the association rule; Extracting a dynamic correction weight corresponding to the device abnormal event according to the timestamp of the device abnormal event within a sliding time window, performing a point-by-point convolution operation on the dynamic correction weight and the instantaneous change value of the power grid load fluctuation to generate a real-time impact component of the device abnormal event; According to the recovery trend of the equipment health assessment indicator, a time decay factor is calculated using a Bayesian recursive estimation method, wherein the decay rate of the time decay factor is negatively correlated with the recovery trend; generating a residual of the historical impact component based on the time attenuation factor, and superimposing the residual and the real-time impact component to generate a total impact; The residuals of the predicted value and the measured value of the grid load fluctuation are calculated in real time to form a residual sequence. If the peak-to-peak value of the residual in the residual sequence is not within the preset steady-state error range, the dynamic correction weight is adaptively adjusted according to the residual sign, and the attenuation rate is corrected based on the total impact amount until the peak-to-peak value of the residual converges to the preset steady-state error range, thereby obtaining a dynamic impact model of equipment abnormal events on grid load fluctuations.

5. The method according to claim 4, characterized in that The method further comprises: extracting a dynamic correction weight corresponding to the device abnormal event according to the timestamp of the device abnormal event within the sliding time window, performing a point-by-point convolution operation on the dynamic correction weight and the instantaneous change value of the power grid load fluctuation to generate a real-time impact component of the device abnormal event, including: Determine the starting boundary and the ending boundary of the sliding time window according to the timestamp of the device abnormal event, and determine the length between the ending boundary and the starting boundary as the sliding time window length; Within the sliding time window, aligning the occurrence time of the device abnormal event according to the timestamp, and extracting the dynamic correction weight corresponding to the device abnormal event; Based on the time decay characteristics of the dynamic correction weight, construct a convolution kernel with a length equal to the sliding time window; Extracting instantaneous change values ​​of grid load fluctuations from a preset grid load fluctuation database according to the sampling frequency of the grid load fluctuations; Performing a point-by-point convolution operation on the instantaneous change value and the convolution kernel to obtain a convolution result, wherein each sampling point in the convolution result corresponds to the local impact intensity of the device abnormal event; The convolution result is amplitude-scaled within a sliding time window, and the scaled convolution result is superimposed with the baseline component of the power grid load fluctuation according to the timestamp to generate a real-time impact component of the equipment abnormal event.

6. The method according to claim 1, characterized in that In the electrical connection subgraph, all edges are traversed along the edge direction, the fault propagation probability of each edge is multiplied, and the total electrical distance value is superimposed to generate the comprehensive weight of the propagation path, including: In the electrical connection subgraph, starting from the starting point, a breadth-first traversal is performed along the edge direction, and a set of traversal paths without repeated vertices is recorded layer by layer, where each path consists of a vertex sequence and a corresponding edge sequence; For each edge sequence, sequentially multiply the fault propagation probability value of each edge to generate a path propagation coefficient, and accumulate the electrical distance values ​​of all edges in the path to obtain a total electrical distance value; The path propagation coefficient is combined with the total electrical distance value to calculate the comprehensive propagation path weight.

7. A power load forecasting system based on association rule analysis, characterized in that: include: A generating module, configured to generate device status data based on the vibration spectrum data and insulation aging index of the power equipment, wherein the device status data includes the vibration spectrum amplitude and the insulation aging index change rate; An analysis module, configured to perform a fusion analysis on the vibration spectrum amplitude and the insulation aging index change rate to generate an equipment health assessment index; A construction module is used to establish a dynamic impact model of abnormal equipment events on power grid load fluctuations based on the equipment health evaluation indicators and association rules, wherein the association rules are generated by analyzing the coupling relationship between the equipment health change trend and the load curve mutation amount within the same time window; an adjustment module, configured to identify potential abnormal devices using the dynamic impact model, perform logical mapping based on the operating parameters of the potential abnormal devices and the power grid topology of the area where the power devices are located, generate fault propagation path weights, and dynamically adjust the power load forecast value according to the fault propagation path weights to obtain an adjusted power load forecast value; The generating of the fault propagation path weight based on the logical mapping of the operating parameters of the potential abnormal device and the power grid topology structure of the area where the power device is located includes: Mapping the power grid topology into a directed graph, wherein vertices in the directed graph represent power devices, vertex attributes include vertex operating parameters, and edges represent electrical connection directions and physical coupling relationships between power devices, and edge attributes include initial weights and electrical distance values; Based on the initial weight and in combination with vertex operating parameters of vertices at both ends of the edge, the fault propagation probability of the edge is calculated by a nonlinear compression function; The power equipment with abnormal equipment health evaluation indicators is regarded as a potential abnormal equipment. Starting from the potential abnormal equipment, the adjacent vertices are expanded hop by hop along the edge direction. When the cumulative number of hops equals the preset propagation order, the expansion is stopped to generate an electrical connection subgraph. In the electrical connection subgraph, all edges are traversed along the edge direction, the fault propagation probability of each edge is multiplied, and the total electrical distance value is superimposed to generate a comprehensive propagation path weight; The path whose comprehensive weight of the propagation path exceeds the preset dynamic threshold is used as the fault propagation path, and the comprehensive weight of the fault propagation path is used as the fault propagation path weight.

8. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a power load forecasting method based on association rule analysis as described in any one of claims 1 to 6.

9. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, the method for predicting power load based on association rule analysis according to any one of claims 1 to 6 is implemented.

Images