Short-circuit parameter change trend prediction method, system, equipment and medium

Abstract

The invention discloses a short-circuit parameter change trend prediction method, system, device and medium, and relates to the technical field of power grid operation and maintenance, and the system comprises a data acquisition and processing module, a model construction module, a feature analysis module, a fusion prediction module and a monitoring deduction module. The method comprises the steps of data acquisition and processing, model construction, feature analysis, fusion prediction and monitoring deduction. Through a multi-model fusion strategy, the short-circuit parameter nonlinear characteristics and the complex time sequence dependency relationship are effectively captured, and the prediction precision and robustness are remarkably improved. And a multi-level anomaly detection mechanism is also established, various anomaly modes are accurately identified, equipment hidden dangers are pre-warned, and preventive maintenance is assisted. And meanwhile, an online learning and dynamic weight adjustment algorithm is utilized to ensure that the model has excellent environmental adaptability and real-time calculation capability. Scientific basis can be provided for power grid operation and maintenance decisions, maintenance strategies are optimized, operation cost is reduced, and equipment reliability is improved.

Description

Technical Field

[0001] This invention relates to the field of power grid operation and maintenance technology, and in particular to a method, system, device and medium for predicting the trend of short-circuit parameter changes. Background Technology

[0002] As power systems expand and age, factors such as aging grid equipment and changes in network topology cause short-circuit parameters to change slowly, significantly impacting the accuracy of grid protection setpoints and the safe and stable operation of the system. Traditional short-circuit parameter calculation methods, primarily based on theoretical calculations and periodic testing, cannot track parameter change trends in real time and are ill-equipped to predict future parameter evolution. Existing parameter monitoring systems mainly focus on instantaneous conditions and lack the ability to analyze and predict long-term trends.

[0003] Existing technologies have several shortcomings when processing time series data of short-circuit parameters. First, traditional statistical methods such as linear regression cannot effectively capture the nonlinear characteristics and complex time dependencies of parameter changes. Second, existing methods lack the ability to identify abnormal change patterns, making it difficult to detect potential equipment failures in a timely manner. Moreover, the prediction accuracy and robustness of a single model are limited, making it difficult to cope with the complex and ever-changing power grid operating environment.

[0004] In terms of data processing, existing technologies are insufficient in handling noisy data and missing values, affecting the accuracy of prediction models. Regarding model selection, there is a lack of dedicated prediction models tailored to the characteristics of short-circuit parameters, and the applicability of general time series models is limited. In terms of system integration, existing methods lack effective integration with power grid operation and management systems, making it difficult to achieve real-time early warning and decision support. Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is that traditional methods cannot track changes in short-circuit parameters in real time, have low prediction accuracy, and weak anomaly detection capabilities.

[0006] The above-mentioned technical problems are solved by the following technical solutions: A method for predicting the changing trend of short-circuit parameters includes data acquisition and processing, model construction, feature analysis, fusion prediction, and monitoring and deduction.

[0007] In a preferred embodiment of the short-circuit parameter change trend prediction method of the present invention: short-circuit parameter data of power grid equipment are acquired, and a multidimensional feature time series is constructed; the multidimensional feature time series is processed to obtain a stationary input sample; based on the stationary input sample, a heterogeneous feature prediction model system including a linear statistical evolution model and a nonlinear intelligent mapping model is constructed; the linear autocorrelation feature of the multidimensional feature time series is analyzed through the linear statistical evolution model, and the long- and short-term dependencies and high-dimensional nonlinear features of the multidimensional feature time series are captured through the nonlinear intelligent mapping model; based on the linear autocorrelation feature, the long- and short-term dependencies, and the high-dimensional nonlinear features, the following is obtained: A basic prediction set is generated; the real-time prediction residuals of the linear statistical evolution model and the nonlinear intelligent mapping model when outputting the basic prediction set are calculated; the weight coefficients of the corresponding linear statistical evolution model and the nonlinear intelligent mapping model are dynamically updated based on the real-time prediction residuals; and the basic prediction set is weighted and fused using the updated weight coefficients to obtain high-precision trend prediction values; based on the deviation between the high-precision trend prediction values ​​and the actual observed values, anomaly scores are calculated using statistical process control theory to quantify parameter deviation and identify anomaly patterns; and by combining the high-precision trend prediction values ​​and the anomaly scores, a panoramic view of parameter evolution is generated to infer the future trend of short-circuit parameters and output prediction information.

[0008] In a preferred embodiment of the short-circuit parameter change trend prediction method of the present invention: processing the multidimensional feature time series to obtain a stationary input sample includes: calculating the statistical feature parameters of the multidimensional feature time series, and determining the stationarity of the series based on the comparison results of the statistical feature parameters and the theoretical critical values; constructing a difference iteration logic, and if the determination result is non-stationary, performing difference operations of increasing order on the multidimensional feature time series; checking the statistical characteristics of the series after the difference operation until a stationary series that meets the stationarity determination condition is obtained, and using the stationary series as the stationary input sample.

[0009] In a preferred embodiment of the short-circuit parameter change trend prediction method of the present invention: the step of analyzing the linear autocorrelation characteristics of the multidimensional feature time series through the linear statistical evolution model includes: constructing an autoregressive polynomial logic to quantify the linear dependence strength between the current time value and the historical time value of the multidimensional feature time series; constructing a moving average polynomial logic to describe the lag effect of the prediction error and the linear correction effect of the lag effect of the prediction error on the current time value; and solving for the coefficients of the autoregressive polynomial logic and the moving average polynomial logic to establish the mathematical expression describing the linear autocorrelation characteristics.

[0010] In a preferred embodiment of the short-circuit parameter change trend prediction method of the present invention: the step of capturing the long-term and short-term dependencies and high-dimensional nonlinear features of the multidimensional feature time series through the nonlinear intelligent mapping model includes: constructing an information filtering mechanism, processing historical state information and current input information through nonlinear transformation, and calculating the memory retention strength to regulate the transmission ratio of historical information; constructing a state update mechanism, combining the filtered current input features to update the temporal memory state inside the nonlinear intelligent mapping model to accumulate long-term dependency information; and constructing an output mapping mechanism, generating a high-dimensional feature output to characterize the high-dimensional nonlinear features based on the updated temporal memory state and mapping weights.

[0011] In a preferred embodiment of the short-circuit parameter change trend prediction method of the present invention: the step of dynamically updating the weight coefficients corresponding to the linear statistical evolution model and the nonlinear intelligent mapping model according to the real-time prediction residual includes: constructing an error feedback mechanism to calculate the weight adjustment step size according to the magnitude of the real-time prediction residual; performing a dynamic correction operation to update the weight coefficients of the previous time step by penalty or reward based on the weight adjustment step size; and implementing normalization constraint processing to map the updated weight coefficients of the linear statistical evolution model and the nonlinear intelligent mapping model to a unit interval to ensure that the probability sum of the fused weights is one.

[0012] In a preferred embodiment of the short-circuit parameter change trend prediction method of the present invention: the calculation of the anomaly score based on the deviation between the high-precision trend prediction value and the actual observed value, combined with statistical process control theory, includes: calculating the residual sequence between the actual observed value and the high-precision trend prediction value, and statistically analyzing the distribution characteristic value of the residual sequence; constructing a dynamic control interval, and setting a fluctuation range including a statistical upper limit and a statistical lower limit based on the distribution characteristic value; performing deviation quantification calculation, measuring the degree of deviation of the current residual value relative to the dynamic control interval, and mapping it to the normalized anomaly score.

[0013] In a preferred embodiment of the short-circuit parameter change trend prediction method of the present invention: the inference of the future trend of the short-circuit parameter includes: constructing a nonparametric trend test statistic and analyzing the rank relationship of the high-precision trend prediction value in the time dimension; calculating a standardized trend index and performing a standardized transformation on the data distribution in combination with the variance of the nonparametric trend test statistic; determining the significance of the trend and determining the monotonic direction and confidence level of the short-circuit parameter change based on the positive and negative signs and the magnitude of the standardized trend index.

[0014] A short-circuit parameter change trend prediction system includes a data acquisition and processing module, a model building module, a feature parsing module, a fusion prediction module, and a monitoring and inference module.

[0015] In a preferred embodiment of the short-circuit parameter change trend prediction system of the present invention: a data acquisition and processing module is used to acquire short-circuit parameter data of power grid equipment, construct a multi-dimensional feature time series, and process it to obtain stationary input samples; a model construction module is used to construct a heterogeneous feature prediction model system including a linear statistical evolution model and a nonlinear intelligent mapping model; a feature analysis module is used to analyze linear autocorrelation features through the linear statistical evolution model and capture long-term and short-term dependencies and high-dimensional nonlinear features through the nonlinear intelligent mapping model; a fusion prediction module is used to generate a basic prediction set, calculate real-time prediction residuals and dynamically update weight coefficients, and obtain high-precision trend prediction values ​​using the updated weight coefficients; a monitoring and inference module is used to calculate anomaly scores to identify anomaly patterns, and infer future trends of short-circuit parameters by combining the panoramic view of parameter evolution and output prediction information.

[0016] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of a short-circuit parameter change trend prediction method as described above.

[0017] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a short-circuit parameter change trend prediction method as described above.

[0018] The beneficial effects of this invention are as follows: By employing a multi-model fusion strategy, it effectively captures the nonlinear characteristics and complex temporal dependencies of short-circuit parameters, significantly improving prediction accuracy and robustness. It also establishes a multi-level anomaly detection mechanism to accurately identify various anomaly patterns and provide early warnings of potential equipment hazards, assisting in preventative maintenance. Simultaneously, it utilizes online learning and dynamic weight adjustment algorithms to ensure the model possesses excellent environmental adaptability and real-time computing capabilities. This provides a scientific basis for power grid operation and maintenance decisions, optimizes maintenance strategies, reduces operating costs, and improves equipment reliability. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Obviously, the drawings described below only relate to some embodiments of the present invention and are not intended to limit the present invention.

[0020] Figure 1 A flowchart of the method for predicting the trend of short-circuit parameter changes is shown; Figure 2 A flowchart of a short-circuit parameter change trend prediction system is shown. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0022] The terminology used in this invention is that which is currently widely used in the art in consideration of the function of the invention; however, these terms may vary according to the intent of those skilled in the art, precedent, or new technology in the art. Furthermore, specific terms may be chosen by the applicant, and in such cases, their detailed meanings will be described in the detailed description of the invention. Therefore, the terms used in this specification should not be construed as simple names, but rather based on their meanings and the overall description of the invention.

[0023] Example 1, referring to Figure 1 This embodiment provides a method for predicting the trend of short-circuit parameter changes, including data acquisition and processing, model construction, feature parsing, fusion prediction, and monitoring and deduction.

[0024] S1, Data Acquisition and Processing: Obtain short-circuit parameter data of power grid equipment and construct a multidimensional feature time series; process the multidimensional feature time series to obtain a stationary input sample.

[0025] Standardized data acquisition and stationarity processing can eliminate noise interference and irregular fluctuations in the raw data, ensuring a high-quality data foundation that serves as a reliable input source for subsequent model analysis.

[0026] S2, Model Building: Based on stationary input samples, a heterogeneous feature prediction model system is constructed, which includes a linear statistical evolution model and a nonlinear intelligent mapping model.

[0027] This establishes the framework of the prediction system, which leverages the complementary advantages of different models to lay a structural foundation for the subsequent comprehensive extraction of simple linear laws and complex nonlinear dynamics in parameter changes.

[0028] S3, Feature Analysis: The linear autocorrelation characteristics of multidimensional feature time series are analyzed by a linear statistical evolution model, and the long-short-term dependencies and high-dimensional nonlinear characteristics of multidimensional feature time series are captured by a nonlinear intelligent mapping model. Based on the linear autocorrelation characteristics, long-short-term dependencies and high-dimensional nonlinear characteristics, a basic prediction set is obtained.

[0029] This avoids the problem that a single model cannot fully characterize the evolution of power grid parameters and enriches the dimensions of prediction information.

[0030] S4, Fusion Prediction: The real-time prediction residuals of the basic prediction set output by the linear statistical evolution model and the nonlinear intelligent mapping model are calculated. The weight coefficients of the corresponding linear statistical evolution model and the nonlinear intelligent mapping model are dynamically updated based on the real-time prediction residuals. The updated weight coefficients are then used to perform weighted fusion on the basic prediction set to obtain high-precision trend prediction values.

[0031] This mechanism enables adaptive adjustment of prediction weights, which can automatically select the best weights based on the model's error performance at the current moment. This significantly reduces the impact of environmental changes on prediction results, thereby greatly improving the overall prediction accuracy and robustness of the system under complex working conditions.

[0032] S5, Monitoring and Simulation: Based on the deviation between high-precision trend predictions and actual observations, anomaly scores are calculated using statistical process control theory to quantify parameter deviation and identify abnormal patterns. By combining high-precision trend predictions and anomaly scores, a panoramic view of parameter evolution is generated, the future trend of short-circuit parameters is deduced, and prediction information is output.

[0033] Transforming numerical predictions into intuitive decision-making basis can sensitively identify potential hazards such as equipment aging, and also provide maintenance personnel with clear guidance on future trends, realizing the transformation from passive monitoring to proactive early warning.

[0034] It should be noted that existing technologies have many shortcomings when processing short-circuit parameter time series data. For example, traditional statistical methods often cannot effectively capture the nonlinear characteristics and complex time dependencies of parameter changes, and the prediction accuracy and robustness of a single model are limited, making it difficult to cope with the complex and ever-changing power grid operating environment. At the same time, they lack the ability to identify abnormal change patterns, resulting in the inability to detect potential equipment failures in a timely manner.

[0035] Therefore, to address the aforementioned problems, this invention constructs a heterogeneous feature prediction model system and introduces an adaptive dynamic fusion mechanism, combining the advantages of linear and nonlinear models to achieve accurate prediction of short-circuit parameter variation trends. This method not only significantly improves prediction accuracy and effectively captures the complex characteristics of parameter changes, but also enhances the ability to identify potential equipment faults through an abnormal state measurement model, providing scientific and real-time decision support for power grid operation and maintenance.

[0036] Example 2, refer to Figure 1 This embodiment, based on the previous embodiment, provides a method for predicting the changing trend of short-circuit parameters.

[0037] Specifically, the process involves acquiring short-circuit parameter data from power grid equipment and constructing a multidimensional feature time series. This multidimensional feature time series is then processed to obtain stationary input samples. Based on these stationary input samples, a heterogeneous feature prediction model system is constructed, incorporating a linear statistical evolution model and a nonlinear intelligent mapping model. The linear autocorrelation characteristics of the multidimensional feature time series are analyzed using the linear statistical evolution model, while the long-term and short-term dependencies and high-dimensional nonlinear characteristics of the multidimensional feature time series are captured using the nonlinear intelligent mapping model. A basic prediction set is obtained based on the linear autocorrelation characteristics, long-term and short-term dependencies, and high-dimensional nonlinear characteristics. The real-time prediction residuals of the basic prediction set output by the linear statistical evolution model and the nonlinear intelligent mapping model are calculated. The weight coefficients of the corresponding linear statistical evolution model and the nonlinear intelligent mapping model are dynamically updated based on the real-time prediction residuals. The updated weight coefficients are then used to perform weighted fusion of the basic prediction set to obtain high-precision trend prediction values. Based on the deviation between the high-precision trend prediction values ​​and the actual observed values, anomaly scores are calculated using statistical process control theory to quantify parameter deviation and identify abnormal patterns. Finally, by combining the high-precision trend prediction values ​​and the anomaly scores, a panoramic view of parameter evolution is generated, predicting future trends of short-circuit parameters and outputting prediction information.

[0038] This invention overcomes the limitations of single models in handling complex short-circuit parameters of power grids by constructing a heterogeneous system that includes linear and nonlinear models. It can simultaneously capture the linear laws and deep nonlinear characteristics of parameter changes. Combined with dynamic weight fusion and anomaly detection mechanisms, it significantly improves prediction accuracy and robustness, and realizes comprehensive perception and accurate early warning of the health status of power grid equipment, providing a scientific and reliable decision-making basis for proactive maintenance of the power grid.

[0039] Specifically, processing multidimensional feature time series to obtain stationary input samples includes: calculating the statistical feature parameters of the multidimensional feature time series, and determining the stationarity of the series based on the comparison results of the statistical feature parameters and the theoretical critical values; constructing a difference iteration logic, and if the determination result is non-stationary, performing difference operations of increasing order on the multidimensional feature time series; and examining the statistical characteristics of the series after the difference operation until a stationary series that meets the stationarity determination conditions is obtained, and using the stationary series as the stationarization input sample.

[0040] Through rigorous statistical testing and adaptive differencing, the interference of non-stationary trend terms and seasonal terms in the original data was eliminated, ensuring that the input data met the mathematical assumptions of the time series analysis model and effectively avoiding spurious regression. This laid the foundation for the stability of model training and the reliability of prediction results.

[0041] Specifically, the analysis of linear autocorrelation characteristics of multidimensional feature time series using linear statistical evolution models includes: constructing an autoregressive polynomial logic to quantify the linear dependence between the current value and historical values ​​of the multidimensional feature time series; constructing a moving average polynomial logic to describe the lag effect of prediction error and the linear correction effect of the lag effect of prediction error on the current value; and solving for the coefficients of the autoregressive polynomial logic and the moving average polynomial logic to establish a mathematical expression describing the linear autocorrelation characteristics.

[0042] By utilizing classical statistical principles, the linear inertial characteristics and error lag effects of short-circuit parameters over time were precisely quantified. This effectively extracted the gradual change patterns of parameters when power grid equipment is in a stable operating state, providing a robust benchmark value with statistical interpretability for overall prediction.

[0043] Specifically, capturing the long-term and short-term dependencies and high-dimensional nonlinear features of multidimensional feature time series through a nonlinear intelligent mapping model includes: constructing an information filtering mechanism, processing historical state information and current input information through nonlinear transformation, and calculating the memory retention strength to regulate the transmission ratio of historical information; constructing a state update mechanism, combining the filtered current input features to update the temporal memory state inside the nonlinear intelligent mapping model to accumulate long-term dependency information; and constructing an output mapping mechanism, generating high-dimensional feature outputs based on the updated temporal memory state and mapping weights to characterize high-dimensional nonlinear features.

[0044] By simulating memory filtering and state update mechanisms, the problem of traditional models being unable to handle long-sequence dependencies is solved. This enables the system to store long-term equipment aging trend information and to keenly capture short-term nonlinear disturbance characteristics, greatly improving the model's fitting ability and generalization performance for complex power grid nonlinear operating conditions.

[0045] Specifically, the dynamic updating of the weight coefficients of the corresponding linear statistical evolution model and nonlinear intelligent mapping model based on the real-time predicted residuals includes: constructing an error feedback mechanism to calculate the weight adjustment step size based on the magnitude of the real-time predicted residuals; performing dynamic correction operations to update the weight coefficients of the previous time step with penalties or rewards based on the weight adjustment step size; and implementing normalization constraint processing to map the updated weight coefficients of the linear statistical evolution model and nonlinear intelligent mapping model to a unit interval to ensure that the probability sum of the fused weights is one.

[0046] A competitive model selection mechanism was established, which can adjust the weight of each sub-model in real time according to the error performance of each sub-model at the current moment, automatically reduce the weight of poorly performing models and increase the influence of high-accuracy models, thereby ensuring that the fusion prediction results are always in the optimal state and significantly enhancing the system's ability to adapt to environmental changes.

[0047] Specifically, the calculation of outlier scores based on the deviation between high-precision trend predictions and actual observations, combined with statistical process control theory, includes: calculating the residual sequence between actual observations and high-precision trend predictions, and statistically analyzing the distribution characteristics of the residual sequence; constructing a dynamic control interval, and setting a fluctuation range including statistical upper and lower limits based on the distribution characteristics; performing deviation quantification calculation, measuring the degree of deviation of the current residual value from the dynamic control interval, and mapping it to a normalized outlier score.

[0048] By converting absolute error values ​​into relative statistical risk measures and dynamically adjusting control limits, the system effectively distinguishes between normal random fluctuations and genuine equipment anomalies. While ensuring a high detection rate, it significantly reduces the false alarm rate, providing maintenance personnel with an intuitive and reliable quantitative assessment of equipment health status.

[0049] Specifically, extrapolating the future trend of short-circuit parameters includes: constructing a nonparametric trend test statistic to analyze the rank relationship of high-precision trend prediction values ​​over time; calculating standardized trend indices and standardizing the data distribution by combining the variance of the nonparametric trend test statistic; determining the significance of the trend and, based on the positive and negative signs and numerical values ​​of the standardized trend indices, determining the monotonic direction and confidence level of the changes in short-circuit parameters.

[0050] Rigorous statistical testing methods are introduced, no longer relying on subjective experience and judgment, but quantitatively providing conclusions on the significance and confidence level of parameter changes. This can help maintenance personnel accurately identify whether there is an irreversible degradation trend in the equipment, thus providing strong theoretical support for the formulation of scientific preventive maintenance strategies.

[0051] It should be noted that existing technologies have many shortcomings when processing short-circuit parameter time series data. For example, traditional statistical methods often cannot effectively capture the nonlinear characteristics and complex time dependencies of parameter changes, and the prediction accuracy and robustness of a single model are limited, making it difficult to cope with the complex and ever-changing power grid operating environment. At the same time, they lack the ability to identify abnormal change patterns, resulting in the inability to detect potential equipment failures in a timely manner.

[0052] Therefore, to address the aforementioned problems, this invention constructs a heterogeneous feature prediction model system and introduces an adaptive dynamic fusion mechanism, combining the advantages of linear and nonlinear models to achieve accurate prediction of short-circuit parameter variation trends. This method not only significantly improves prediction accuracy and effectively captures the complex characteristics of parameter changes, but also enhances the ability to identify potential equipment faults through an abnormal state measurement model, providing scientific and real-time decision support for power grid operation and maintenance.

[0053] Example 3, referring to Figure 1 and Figure 2This embodiment provides a short-circuit parameter change trend prediction system, including a data acquisition and processing module, a model building module, a feature parsing module, a fusion prediction module, and a monitoring and inference module.

[0054] Specifically, the data acquisition and processing module is used to acquire short-circuit parameter data of power grid equipment, construct a multi-dimensional feature time series, and process the multi-dimensional feature time series to obtain a stable input sample.

[0055] The data acquisition and processing module eliminates noise interference and non-stationary fluctuations in the original data through standardized data collection and strict stationarity processing, establishing a high-quality data foundation and thus providing a reliable input source that conforms to mathematical assumptions for subsequent model analysis.

[0056] Specifically, the model building module is used to construct a heterogeneous feature prediction model system that includes linear statistical evolution models and nonlinear intelligent mapping models.

[0057] The model building module establishes the core architecture of the prediction system. By leveraging the complementary advantages of models with different properties, it lays a solid structural foundation for the system to comprehensively extract simple linear laws and complex nonlinear dynamics in parameter changes.

[0058] Specifically, the feature parsing module is used to analyze linear autocorrelation features through a linear statistical evolution model, and to capture long- and short-term dependencies and high-dimensional nonlinear features through a nonlinear intelligent mapping model.

[0059] The feature parsing module solves the technical challenge of a single model being unable to fully characterize the evolution characteristics of power grid parameters, enriches the dimensions of prediction information, and ensures a deep understanding of the complex operating state of the power grid.

[0060] Specifically, the fusion prediction module is used to generate a basic prediction set, calculate the real-time prediction residuals and dynamically update the weight coefficients, and use the updated weight coefficients to obtain high-precision trend prediction values.

[0061] The fusion prediction module enables adaptive adjustment of prediction weights, which can automatically select the best weights based on the model's error performance at the current moment. This significantly reduces the impact of environmental changes on the prediction results, thereby greatly improving the overall prediction accuracy and robustness of the system under complex working conditions.

[0062] Specifically, the monitoring and simulation module is used to calculate anomaly scores to identify anomaly patterns, and combines the panoramic view of parameter evolution to infer the future trend of short-circuit parameters and output prediction information.

[0063] The monitoring and simulation module transforms numerical predictions into intuitive decision-making basis, which can not only sensitively identify potential hidden dangers such as equipment aging, but also provide clear guidance on future trends for operation and maintenance personnel, realizing the transformation from passive monitoring to proactive early warning.

[0064] It should be noted that existing technical solutions usually rely on theoretical calculations or periodic testing, making it difficult to track parameter changes in real time. They also focus more on instantaneous states and lack the ability to analyze long-term trends. At the same time, traditional statistical methods cannot effectively capture nonlinear characteristics, and single models have significant deficiencies in prediction accuracy and anomaly detection capabilities when dealing with complex power grid environments.

[0065] Therefore, to address the aforementioned problems, this invention constructs a prediction system incorporating heterogeneous models and combines it with an adaptive dynamic fusion mechanism, achieving accurate prediction of short-circuit parameter variation trends and real-time measurement of abnormal states. This system not only significantly improves prediction accuracy and environmental adaptability but also effectively enhances the ability to identify potential equipment faults, providing a scientific and reliable decision support platform for proactive power grid maintenance and safe operation.

[0066] Example 4, refer to Figure 1 This embodiment, based on the first embodiment, provides a method for predicting the changing trend of short-circuit parameters. The data acquisition and processing includes short-circuit parameter time series modeling; the model construction includes ARIMA model construction, LSTM neural network model construction, and support vector regression model construction; the feature analysis corresponds to the autoregressive part and moving average part in the ARIMA model, the LSTM unit structure in the LSTM neural network model, and the kernel function mapping in the support vector regression model; the fusion prediction includes model fusion strategy and evaluation index system; and the monitoring and inference includes anomaly detection mechanism and trend analysis method.

[0067] Specifically: 1. Short-circuit parameter time series modeling Short-circuit parameter time series definition: (1) In the formula, Let be the short-circuit parameter vector at time t. For resistance, For reactance, For electrical conductivity, It is the susceptance.

[0068] Parameter change rate calculation: (2) In the formula, Let be the rate of change of the parameter at time t. The parameter value is from the previous moment.

[0069] Time series stationarity test: (3) In the formula, To enhance the Dickie-Fowler statistic, For the estimated regression coefficients, This is the theoretical value. This represents the standard error.

[0070] Differential operations: (4) In the formula, Let L be a difference sequence of order d, where L is the lag operator and d is the difference order.

[0071] 2. ARIMA Model Construction ARIMA model expression: (5) In the formula, It is an autoregressive polynomial. It is a moving average polynomial. It is a white noise sequence.

[0072] Autoregressive component: (6) In the formula, is the autoregressive coefficient, and p is the autoregressive order.

[0073] Moving average section: (7) In the formula, q represents the moving average coefficient, and q represents the moving average order.

[0074] Maximum likelihood estimation: (8) In the formula, Let n be the log-likelihood function, and n be the sample size. This represents the error variance.

[0075] 3. LSTM Neural Network Model LSTM cell structure: (9) In the formula, Output for the forget gate. It is the sigmoid activation function. Here is the forget gate weight matrix. This is the hidden state from the previous moment. For the current input, This is the bias vector.

[0076] Input Gate: (10) In the formula, For input gate output, The input gate weight matrix, This is the input gate bias.

[0077] Candidate values: (11) In the formula, Candidate memory cell state, The candidate value weight matrix, For candidate value bias, It is the hyperbolic tangent activation function.

[0078] Cell status update: (12) In the formula, The current cell state, This represents the cell state at the previous moment. This represents element-wise product.

[0079] Output gate: (13) In the formula, For output gate output, This is the output gate weight matrix. This is used to bias the output gate.

[0080] Hidden state output: (14) In the formula, Output the current hidden state.

[0081] 4. Support Vector Regression Model SVR optimization issues: (15) In the formula, Let C be the weight vector and C be the penalty parameter. , is a slack variable, and n is the number of samples.

[0082] Constraints: (16) In the formula, For the target value, Here, b is the feature map, and b is the bias term. This is an insensitive loss parameter.

[0083] (17) Kernel function: (18) In the formula, For radial basis kernel functions, For kernel parameters.

[0084] 5. Model fusion strategy Weighted average fusion: (19) In the formula, To merge the predicted values, The weights of the k-th model are... Let be the predicted value of the k-th model, where K is the total number of models.

[0085] Weight optimization: (20) In the formula, Let be the mean square error of the k-th model.

[0086] Dynamic weight adjustment: (twenty one) In the formula, Let be the dynamic weights of the k-th model at time t. To adjust the coefficient, This represents the prediction error.

[0087] 6. Anomaly detection mechanism Control chart methods: (twenty two) In the formula, UCL is the upper control limit, and LCL is the lower control limit. The mean, The standard deviation is denoted as .

[0088] Residual analysis: (twenty three) In the formula, Let be the residual at time t. These are predicted values.

[0089] Anomaly Measurement: (twenty four) In the formula, These are abnormal scores. This represents the standard deviation of the residuals.

[0090] 7. Evaluation Index System Root mean square error: (25) In the formula, This is the root mean square error.

[0091] Mean absolute percentage error: (26) In the formula, The mean absolute percentage error.

[0092] Symmetrical mean absolute percentage error: (27) In the formula, It represents the symmetrical average absolute percentage error.

[0093] Coefficient of determination: (28) In the formula, As the coefficient of determination, This represents the mean of the observed values.

[0094] 8. Trend Analysis Methods Linear trend test: (29) In the formula, T is the t-statistic. This is an estimate of the slope. This represents the standard error.

[0095] Mann-Kendall trend test: (30) In the formula, S is the Mann-Kendall statistic, and sign is the sign function.

[0096] Standardized statistics: (31) In the formula, Z is the standardized statistic. Let S be the variance of S.

[0097] 9. Algorithm Flow 1. Data Acquisition and Preprocessing Stage A short-circuit parameter data acquisition system was established to obtain long-term historical data through online monitoring equipment and periodic testing. The raw data was quality assessed to identify and process outliers, missing values, and noisy data. Interpolation methods were used to complete missing data, and filtering techniques were employed to remove high-frequency noise.

[0098] Implement data standardization to eliminate dimensional differences between different parameters. Establish a data storage and management system to ensure data integrity and consistency. Design data security protection mechanisms to safeguard data security and privacy.

[0099] 2. Time series modeling stage The stationarity of the short-circuit parameter time series was tested using the ADF and KPSS tests. Non-stationary series were differencing until a stationary series was obtained. The autocorrelation and partial autocorrelation functions of the series were analyzed to determine the order of the ARIMA model.

[0100] Various time series models were established, including ARIMA, exponential smoothing, and state-space models. Information criteria such as AIC and BIC were used for model selection and parameter estimation. The applicability and effectiveness of the models were verified through residual analysis.

[0101] 3 Machine Learning Model Building Stage Design an LSTM neural network architecture and determine hyperparameters such as the number of hidden layers and neurons. Construct a support vector regression model and select appropriate kernel functions and parameters. Build ensemble learning models such as random forest and gradient boosting.

[0102] Hyperparameter tuning was performed using methods such as grid search and Bayesian optimization. Cross-validation was implemented to evaluate model performance and select the optimal model structure. Regularization mechanisms were designed to prevent overfitting and improve the model's generalization ability.

[0103] 4. Model Fusion and Optimization Stage Design multi-model fusion strategies, including simple averaging, weighted averaging, and stacking methods. Establish a dynamic weight adjustment mechanism to dynamically adjust the fusion weights based on model performance. Implement model ensemble optimization to improve prediction accuracy and robustness.

[0104] Develop online learning algorithms to enable the model to continuously learn and update from new data. Design a model performance monitoring mechanism to promptly detect model performance degradation and perform retraining. Establish a model version management system to ensure model traceability.

[0105] 5. Anomaly Detection and Early Warning Stage Establish a multi-layered anomaly detection mechanism, including statistical methods, machine learning methods, and deep learning methods. Design an anomaly pattern recognition algorithm to distinguish between normal and abnormal changes. Establish a dynamic adjustment mechanism for early warning thresholds, adjusting early warning parameters based on historical data and operating conditions.

[0106] Develop a real-time monitoring system to track and analyze changes in short-circuit parameters in real time. Design a multi-level early warning mechanism to provide different levels of warning signals based on the severity of the anomaly. Establish an early warning information push system to promptly send early warning information to operations personnel.

[0107] 6. System Integration and Application Phase Develop a user-friendly visual interface to display forecast results, trend analysis, and anomaly detection results. Establish an interface with the power grid dispatching system to enable automatic transmission and application of forecast results. Design decision support functions to provide maintenance suggestions and response strategies for operators.

[0108] Establish a system performance evaluation mechanism to regularly assess prediction accuracy and system reliability. Design system expansion interfaces to support the integration of new models and functions. Establish a user training and technical support system to ensure the effective application of the system.

[0109] 10. Simulation Verification A simulation verification platform based on 10 years of historical data from a 500kV substation in a provincial power grid was established, including short-circuit parameters of equipment such as main transformers, lines, and circuit breakers. The dataset contains monthly measurements of parameters such as resistance, reactance, conductance, and susceptance, totaling 1200 data points. The simulation environment uses the MATLAB platform, integrating machine learning libraries such as TensorFlow and Scikit-learn.

[0110] Table 1: Performance Comparison of Different Prediction Models

[0111] Table 2: Prediction accuracy analysis for different parameter types

[0112] Table 3: Anomaly Detection Performance Evaluation

[0113] Simulation results demonstrate that the short-circuit parameter variation trend prediction method based on time series analysis proposed in this patent achieves excellent performance across all performance indicators. In terms of prediction accuracy, the RMSE of the model fusion method is reduced to 0.0176, MAPE to 2.67%, and R² reaches 0.942, significantly outperforming the single model. Prediction results for different parameter types show that reactance parameters have the highest prediction accuracy, with a trend prediction accuracy of 91.7%. Regarding anomaly detection, the detection rate for topology changes is the highest, reaching 96.5%, with a false alarm rate controlled below 2.3%. The response time is controlled within 1.2 seconds, meeting real-time monitoring requirements. Sensitivity analysis indicates that the model is sensitive to data quality; when the data missing rate exceeds 15%, the prediction accuracy decreases significantly. Robustness testing shows that the model performance remains stable under different operating conditions, verifying the reliability of the method. Long-term tracking experiments demonstrate that the prediction model can effectively track seasonal changes and long-term trends of parameters, providing strong support for power grid maintenance decisions. Overall simulation verification fully demonstrates the significant advantages of this method in improving prediction accuracy, enhancing anomaly detection capabilities, and providing decision support.

[0114] Finally, it should be noted that the methods and devices described in detail above are merely embodiments, and those skilled in the art can modify these embodiments in different ways as long as they do not depart from the scope of the present invention.

Claims

1. A method for predicting the changing trend of short-circuit parameters, characterized in that, include: Obtain short-circuit parameter data of power grid equipment and construct a multidimensional feature time series; The multidimensional feature time series is processed to obtain a stable input sample; Based on the stabilized input samples, a heterogeneous feature prediction model system including a linear statistical evolution model and a nonlinear intelligent mapping model is constructed. The linear autocorrelation characteristics of the multidimensional feature time series are analyzed by the linear statistical evolution model, and the long-term and short-term dependencies and high-dimensional nonlinear characteristics of the multidimensional feature time series are captured by the nonlinear intelligent mapping model. Based on the linear autocorrelation characteristics, the long-term and short-term dependencies, and the high-dimensional nonlinear characteristics, a basic prediction set is obtained; The real-time prediction residuals of the linear statistical evolution model and the nonlinear intelligent mapping model when outputting the basic prediction set are calculated. The weight coefficients of the corresponding linear statistical evolution model and the nonlinear intelligent mapping model are dynamically updated according to the real-time prediction residuals. The updated weight coefficients are then used to perform weighted fusion on the basic prediction set to obtain a high-precision trend prediction value. Based on the deviation between the high-precision trend prediction value and the actual observed value, anomaly score is calculated in combination with statistical process control theory to quantify parameter deviation and identify anomaly patterns. By combining the high-precision trend prediction value and the anomaly score, a panoramic map of parameter evolution is generated, the future trend of short-circuit parameters is inferred, and prediction information is output.

2. The method for predicting the trend of short-circuit parameter changes as described in claim 1, characterized in that: The process of processing the multidimensional feature time series to obtain the stable input sample includes: Calculate the statistical characteristic parameters of the multidimensional feature time series, and determine the stationarity of the series based on the comparison results of the statistical characteristic parameters and the theoretical critical values; Construct a difference iteration logic; if the determination result is non-stationary, perform difference operations of increasing order on the multidimensional feature time series. The statistical properties of the sequence after the difference operation are examined until a stationary sequence that meets the stationarity determination condition is obtained, and the stationary sequence is used as the input sample for the stationarization.

3. The method for predicting the trend of short-circuit parameter changes as described in claim 1, characterized in that: The method of analyzing the linear autocorrelation features of the multidimensional feature time series using the linear statistical evolution model includes: Construct an autoregressive multinomial logic to quantify the linear dependence strength between the current time value and the historical time value of the multidimensional feature time series; Construct a moving average polynomial logic to describe the lag effect of the prediction error and the linear correction effect of the lag effect of the prediction error on the current value. Solve for the coefficients of the autoregressive polynomial logic and the moving average polynomial logic to establish a mathematical expression describing the linear autocorrelation characteristics.

4. The method for predicting the trend of short-circuit parameter changes as described in claim 1, characterized in that: The process of capturing the long- and short-term dependencies and high-dimensional nonlinear features of the multidimensional feature time series through the nonlinear intelligent mapping model includes: An information filtering mechanism is constructed, which processes historical state information and current input information through nonlinear transformation, and calculates the memory retention strength to regulate the transmission ratio of historical information. A state update mechanism is constructed to update the temporal memory state inside the nonlinear intelligent mapping model by combining the filtered current input features in order to accumulate long-term dependency information. An output mapping mechanism is constructed to generate high-dimensional feature outputs to characterize the high-dimensional nonlinear features based on the updated temporal memory state and mapping weights.

5. The method for predicting the trend of short-circuit parameter changes as described in claim 1, characterized in that: The step of dynamically updating the weight coefficients corresponding to the linear statistical evolution model and the nonlinear intelligent mapping model based on the real-time predicted residuals includes: An error feedback mechanism is constructed, and the step size is adjusted by calculating the weight based on the magnitude of the real-time predicted residual. Perform a dynamic correction operation, and update the weight coefficients of the previous time step by applying a penalty or reward based on the weight adjustment step size; Normalization constraint processing is implemented to map the updated linear statistical evolution model and the nonlinear intelligent mapping model weight coefficients to a unit interval, ensuring that the probability sum of the fused weights is one.

6. The method for predicting the trend of short-circuit parameter changes as described in claim 1, characterized in that: The calculation of the anomaly score based on the deviation between the high-precision trend prediction value and the actual observed value, combined with statistical process control theory, includes: Calculate the residual sequence between the actual observed values ​​and the high-precision trend prediction values, and statistically analyze the distribution characteristics of the residual sequence; Construct a dynamic control interval and set a fluctuation range including a statistical upper limit and a statistical lower limit based on the distribution characteristic value; Perform deviation quantification calculation to measure the degree of deviation of the current residual value relative to the dynamic control interval, and map it to the normalized anomaly score.

7. The method for predicting the trend of short-circuit parameter changes as described in claim 1, characterized in that: The predicted future trends of the short-circuit parameters include: Construct a nonparametric trend test statistic and analyze the rank relationship of the high-precision trend prediction values ​​in the time dimension; Calculate the standardized trend index, and then standardize the data distribution by combining the variance of the nonparametric trend test statistic. To determine the significance of the trend, the monotonic direction and confidence level of the short-circuit parameter change are determined based on the positive and negative signs and numerical values ​​of the standardized trend index.

8. A short-circuit parameter variation trend prediction system, using the method described in any one of claims 1 to 7, characterized in that, include: The data acquisition and processing module is used to acquire short-circuit parameter data of power grid equipment, construct multi-dimensional feature time series, and process them to obtain stable input samples; The model building module is used to construct a heterogeneous feature prediction model system that includes linear statistical evolution models and nonlinear intelligent mapping models; The feature parsing module is used to parse linear autocorrelation features through the linear statistical evolution model and to capture long- and short-term dependencies and high-dimensional nonlinear features through the nonlinear intelligent mapping model. The fusion prediction module is used to generate a basic prediction set, calculate the real-time prediction residuals and dynamically update the weight coefficients, and use the updated weight coefficients to obtain high-precision trend prediction values. The monitoring and simulation module is used to calculate anomaly scores to identify anomaly patterns, and combines the panoramic view of parameter evolution to infer the future trend of short-circuit parameters and output prediction information.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the short-circuit parameter change trend prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the short-circuit parameter change trend prediction method according to any one of claims 1 to 7.

Images