A Fault Prediction Method, Device, Equipment and Medium for a Flexible DC Distribution Network

Abstract

The flexible DC distribution network fault prediction method, device, equipment and medium disclosed by the present invention collect the grid operation data of the flexible DC distribution network; extract features from the grid operation data to obtain feature vectors; optimize the preset initial weight vector by using a preset optimization target model to adjust the feature weight allocation; and calculate the fault probability according to the feature weight allocation, the feature vectors and a pre-constructed fault probability prediction model. This solution can improve the adaptability and accuracy of fault prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular, to a method, device, equipment and medium for fault prediction of a flexible DC distribution network. Background Art

[0002] With the transformation of the global energy structure and the proposal of the "dual carbon" goal, flexible DC distribution networks have gradually become an important direction for the future development of power systems. Flexible DC distribution networks have the advantages of high efficiency, flexibility, and controllability, and show significant application value in large-scale renewable energy access, long-distance power transmission, and power systems with large load fluctuations.

[0003] However, an important challenge faced by flexible DC distribution networks in actual operation is the rapidity and complexity of their fault processes. Especially in the case of frequent fluctuations in grid data, traditional fault outage prediction methods are often difficult to respond in a timely manner, and prediction deviations are likely to occur when the grid operation state changes rapidly, resulting in low fault prediction accuracy. Summary of the Invention

[0004] In view of the above defects, the present invention provides a method, device, equipment and medium for fault prediction of a flexible DC distribution network, which can improve the adaptability and accuracy of fault prediction.

[0005] An embodiment of the present invention provides a method for fault prediction of a flexible DC distribution network, the method comprising:

[0006] Collecting grid operation data of the flexible DC distribution network;

[0007] Performing feature extraction on the grid operation data to obtain a feature vector;

[0008] Optimizing a preset initial weight vector by using a preset optimization target model to adjust the feature weight distribution;

[0009] Calculating a fault probability according to the feature weight distribution, the feature vector, and a pre-constructed fault probability prediction model.

[0010] Preferably, performing feature extraction on the grid operation data to obtain a feature vector includes:

[0011] Performing missing value processing and outlier processing on the grid operation data;

[0012] According to the processed grid operation data, extracting key features affecting the fault detection of the flexible DC distribution network to obtain the feature vector.

[0013] Further, the key features include voltage volatility, power mutation rate, and load change rate;

[0014] Among them, the voltage volatility and the power mutation rate and the load change rate , is the voltage at the i-th time point, is the average value of the voltage, N is the number of time points, and are the power values at the current time point and the previous time point respectively, and are the load values at the current time point and the previous time point respectively.

[0015] As a preferred solution, a preset optimization target model is used to optimize the preset initial weight vector and adjust the feature weight allocation, including:

[0016] Determine the initial weight vector of the feature vector according to the obtained historical power grid operation data;

[0017] Use the gradient descent algorithm to optimize the initial weight vector in real time with the goal of minimizing the preset optimization target model and adjust the feature weight allocation.

[0018] Furthermore, the optimization target model includes: ;

[0019] Among them, is the loss function, that is, the optimization target, represents the feature weight vector, represents the input vector after weighting the feature vector, w represents the weight vector of the optimization target model, represents the input feature vector of the i-th sample, represents the true label value of the i-th sample, represents the total number of samples of the input data.

[0020] Preferably, determining the initial weight vector of the feature vector according to the obtained historical power grid operation data includes:

[0021] When the load change rate calculated according to the historical power grid operation data exceeds a preset first threshold, it is recognized that the high-load working condition is in place, and the preset first vector is set as the initial weight vector;

[0022] When the proportion of the voltage volatility in the preset abnormal index range monitored according to the historical power grid operation data exceeds a preset second threshold, it is recognized that the low-load working condition is in place, and the preset second vector is set as the initial weight vector.

[0023] Preferably, the construction process of the fault probability prediction model specifically includes:

[0024] Divide the pre-collected feature data into a training set and a test set;

[0025] Initialize the population of the genetic algorithm and randomly generate several initial solutions as initial individuals;

[0026] Performance parameters for fault prediction through a pre-constructed fitness function;

[0027] According to the performance metrics calculated by the fitness function, use a preset selection method to select individuals with better fitness as the next generation;

[0028] Adopt a preset crossover probability adjustment model to dynamically adjust the crossover probability of the population;

[0029] Adopt a preset mutation probability adjustment model to dynamically adjust the mutation probability of the population;

[0030] Calculate the parameters of the current generation of individuals. During the optimization process of each generation, calculate the standard deviation change rate and fitness gain rate of the power grid data in real time as the input of the fitness function, and adjust the crossover probability and mutation probability in real time for iteration until the calculation result of the fitness function reaches a preset threshold or the iteration reaches a preset maximum number of iterations, and output the latest individual as the parameters of the Logistic regression model to obtain the fault probability prediction model.

[0031] Furthermore, the crossover probability adjustment model is ;

[0032] The mutation probability adjustment model is ;

[0033] The performance parameters include accuracy and recall rate;

[0034] The accuracy ;

[0035] The recall rate is ;

[0036] Wherein, is the crossover probability, the change rate of the standard deviation of the power grid operation data ,

[0037] and are the standard deviations of the current generation population data and the previous generation population data respectively, is a preset value, the fitness gain rate , and are the average fitness of the current generation population data and the previous generation population data; is the mutation probability, It represents the non - linear dynamics of the enhanced mutation probability. TP, TN, FP, and FN respectively represent the numbers of true positive examples, true negative examples, false positive examples, and false negative examples. and are respectively the minimum and maximum values of the crossover probability. and are respectively the minimum and maximum values of the mutation probability.

[0038] Preferably, the fault probability prediction model is ;

[0039] Among them, represents the initial weight vector, represents the input vector after weighting the feature vector, represents the feature vector, represents the feature weight vector optimized by the genetic algorithm, is the result output by the flexible DC distribution network fault prediction model, indicating the probability of a fault occurring under the given feature vector x. The probability of occurrence.

[0040] An embodiment of the present invention also provides a flexible DC distribution network fault prediction device, and the device includes:

[0041] A collection module, configured to collect the grid operation data of the flexible DC distribution network;

[0042] An extraction module, configured to perform feature extraction on the grid operation data to obtain a feature vector;

[0043] An optimization module, configured to optimize the preset initial weight vector by using a preset optimization target model and adjust the feature weight distribution;

[0044] A calculation module, configured to calculate the fault probability according to the feature weight distribution, the feature vector, and a pre - constructed fault probability prediction model.

[0045] An embodiment of the present invention also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the flexible DC distribution network fault prediction method as described in any one of the above - mentioned embodiments.

[0046] An embodiment of the present invention also provides a computer - readable storage medium. The computer - readable storage medium includes a stored computer program. Among them, when the computer program runs, it controls the device where the computer - readable storage medium is located to execute the flexible DC distribution network fault prediction method as described in any one of the above - mentioned embodiments.

[0047] The flexible DC distribution network fault prediction method, device, equipment and medium provided by the present invention collect the grid operation data of the flexible DC distribution network; extract features from the grid operation data to obtain feature vectors; optimize the preset initial weight vector by using a preset optimization target model to adjust the feature weight distribution; calculate the fault probability according to the feature weight distribution, the feature vector and a pre-constructed fault probability prediction model. This solution can improve the adaptability and accuracy of fault prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a schematic flowchart of a flexible DC distribution network fault prediction method provided by an embodiment of the present invention;

[0049] Figure 2 is a schematic structural diagram of a flexible DC distribution network fault prediction device provided by an embodiment of the present invention;

[0050] Figure 3 is a schematic structural diagram of a terminal device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0052] See Figure 1 , which is a schematic flowchart of a flexible DC distribution network fault prediction method provided by an embodiment of the present invention. The method includes steps S1 to S3:

[0053] Step S1, collect the grid operation data of the flexible DC distribution network;

[0054] Step S2, extract features from the grid operation data to obtain feature vectors;

[0055] Step S3, optimize the preset initial weight vector by using a preset optimization target model to adjust the feature weight distribution;

[0056] Step S4, calculate the fault probability according to the feature weight distribution, the feature vector and a pre-constructed fault probability prediction model.

[0057] In the specific implementation of this embodiment, the grid operation data is collected in real time through the monitoring equipment of the distribution network, including voltage, current, power, load change rate, etc.

[0058] By analyzing the power grid operation data, the key feature vectors that affect the fault detection of the flexible DC distribution network are used as input features, with the goal of enabling the model to more effectively identify faults.

[0059] These features are incorporated into the input vector of the Logistic regression model, enabling the model to more accurately capture the possibility of power grid faults.

[0060] To further improve the accuracy of power grid fault prediction, based on dynamic feature weighting, the present invention strengthens the robustness and accuracy of the model by exemplifying different power grid operation conditions, clarifying the setting method of the optimal parameter vector, and its precise association with the prediction probability. That is, the dynamic weight is set and optimized. First, the weight vector is initialized, and then the gradient descent algorithm is used to optimize the weight vector in real time, enabling the model to adjust the feature weight allocation according to the current power grid operation conditions.

[0061] Based on the feature weighting optimization, a fault probability prediction model based on the device state is generated through an improved weighted optimization Logistic regression model. Calculate the fault probability according to the feature weight allocation, the feature vector, and the pre-constructed fault probability prediction model;

[0062] Judge whether the flexible DC distribution network has a fault according to the fault probability.

[0063] That is, when the predicted fault probability exceeds the preset fault threshold, it is judged that the flexible DC distribution network has a fault.

[0064] This application provides a fault prediction algorithm with adaptive adjustment ability, and the adaptability and accuracy of fault outage prediction are relatively high.

[0065] In another embodiment provided by the present invention, the step S1 specifically includes the following steps:

[0066] After obtaining the power grid operation data, data cleaning is also required.

[0067] Data cleaning includes missing value processing and outlier processing. Since the missing data accounts for a small part, the deletion method is used to directly discard the missing data; for abnormal data, the average value of the same nature is used for interpolation to replace the abnormal data. Standardize the numerical data to unify the data scale.

[0068] After data cleaning, feature selection and definition will be carried out.

[0069] By analyzing the power grid operation data, the key feature vectors that affect the fault detection of the flexible DC distribution network are used as input features, enabling the model to more effectively identify faults.

[0070] In another embodiment provided by the present invention, the extracted key features include voltage volatility rate, power mutation rate, and load change rate;

[0071] Voltage volatility rate : Defined as the standard deviation change rate of voltage per unit time, it characterizes the voltage fluctuation under grid fault conditions, especially obvious in scenarios such as short circuits and overloads.

[0072] Voltage volatility rate ;

[0073] Wherein, represents the voltage at the i-th time point, is the average value of the voltage, and N is the number of time points.

[0074] Power mutation rate Used to describe the degree of power change per unit time, it can capture the characteristics of sharp power changes caused by abnormal grid equipment or sudden load changes, which is a highly sensitive manifestation of the change of equipment state.

[0075] The power mutation rate .

[0076] Wherein, and are the power values at the current and previous time points respectively.

[0077] Load change rate Describes the sudden fluctuation of the load, reveals the load transfer and abnormal load disturbance between different nodes of the power grid. Compared with other variables such as voltage and current, it can more accurately reflect the balance and dynamic changes of the power grid load.

[0078] The load change rate .

[0079] Wherein, and are the load values at the current and previous time points respectively.

[0080] The characteristics of the power grid are described from different angles through three variables, but their fluctuations show a synergistic effect in time. For example, when the voltage volatility rate is abnormal, the power mutation rate usually fluctuates violently in synchronization, and the load change rate also has corresponding sudden increases or decreases in the load distribution. This dynamic consistency determines their core role as "key feature vectors". The surface characteristics of power grid operation faults (such as voltage dips and current abnormalities) can only be used as the basis for the first layer of identification and cannot fully reflect the essence of the faults. While the voltage volatility rate, power mutation rate, and load change rate as the second-layer key features are further in-depth identifications of fault patterns and can significantly improve the identification accuracy.

[0081] In another embodiment provided by the present invention, step S3 in the above embodiment specifically includes the following steps:

[0082] First, initialize the weight vector. Based on historical power grid operation data and combined with a feature selection algorithm, determine the initial weight vectors of voltage volatility, power mutation rate, and load change rate.

[0083] Subsequently, use the gradient descent algorithm to optimize the weight vector in real time, with the preset optimization target model minimized as the goal, so that the model can adjust the feature weight allocation according to the current power grid operation conditions.

[0084] It should be noted that the optimization goal of the weight is to minimize the loss function. To avoid unbalanced feature weight allocation, a regularization term is introduced to limit the amplitude of weight change. The formula is: .

[0085] Among them, λ is the regularization coefficient, which is used to prevent the model from overfitting.

[0086] In another embodiment provided by the present invention, the optimization target model includes: ;

[0087] Among them, is the loss function, that is, the optimization goal, represents the initial weight vector, represents the input vector after feature vector weighting, w represents the weight vector of the optimization target model, represents the input feature vector of the i-th sample, represents the true label value of the i-th sample, represents the total number of samples of the input data.

[0088] w represents the weight vector of the optimization model, which is a key parameter in the logistic regression model and is used to quantify the influence degree of each input feature on the prediction target. In the present invention, the value of w is optimized by a genetic algorithm and dynamically adjusted to adapt to different power grid operation conditions. The high or low value of the weight w of different features (such as voltage volatility, power mutation rate, load change rate) represents their importance in predicting fault events.

[0089] represents the input feature vector of the i-th sample, which contains multiple key feature vectors, such as voltage volatility, power mutation rate, load change rate, etc. Each corresponds to a moment or state point in the actual power grid operation, and the feature value is derived from real-time collected power grid monitoring data (such as voltage, current, load change). is the core input of the model, and through feature weighting (combined with w and ), it is used to predict the probability of a fault occurring in the i-th sample.

[0090] is the true label value of the i-th sample, used to represent the power grid state corresponding to this sample: = 1: indicates that a fault has occurred. = 0 indicates that the power grid is operating normally. It comes from the historical operation data of the power grid and is generated by annotating fault records, representing the supervision signal for model training.

[0091] represents the total number of samples of the input data, that is, the number of samples included in the historical operation data of the power grid. The size of determines the coverage of the training data and the robustness of the model. The power grid operation samples cover various scenarios such as high load, low load, and multiple-point faults. determines the loss function The summation range of, which is a basic index for model training. A larger provides rich training samples and improves the model's adaptability to complex scenarios. This embodiment provides an optimization mechanism with strong adaptability, significantly improving the ability of power grid fault identification.

[0092] In another embodiment provided by the present invention, when determining the initial weight vector, it specifically includes the following steps:

[0093] Determine the voltage volatility rate, power mutation rate, and load change rate according to the historical power grid operation data, and perform load condition detection.

[0094] When the load change rate calculated according to the historical power grid operation data exceeds a preset first threshold, that is, the load change rate fluctuates significantly, it is recognized that the high load condition is in the initial weight vector of the high load condition .

[0095] When the proportion of the voltage volatility rate within the preset abnormal index range monitored according to the historical power grid operation data exceeds a preset second threshold, that is, the voltage volatility rate is the main abnormal index, it is recognized that the low load condition is in the initial weight vector of the low load condition .

[0096] It should be noted that in this embodiment, the given initial weight vector is only a preferred embodiment. In other embodiments, other initial weight vectors can be set under different working conditions.

[0097] In another embodiment provided by the present invention, when constructing a fault probability prediction model, it specifically includes the following steps:

[0098] Conduct a comparison test on the power grid state characteristics. To verify the performance of the model in power grid fault detection, divide the training set and the test set according to 8:2.

[0099] It has been verified that the prediction accuracy (Accuracy) of the model under different operating conditions has been improved from 85% of the traditional method to 94%. In complex working conditions with severe load fluctuations, the recall rate has been improved from 82% to 91%.

[0100] A genetic algorithm-based optimization model is established. First, the operation of initializing the population is carried out. A certain number of initial solutions (chromosomes) are randomly generated, and each solution represents a set of parameters (such as weights and biases) of the Logistics regression model. The population is . Among them represents the i-th chromosome, that is, the parameter solution.

[0101] Construct a fitness function. It is used to evaluate the quality of each solution. The fitness function is based on the prediction performance of the model.

[0102] Perform the selection operation. According to the results of the fitness function, the roulette wheel selection method is used to select individuals with higher fitness to enter the next generation.

[0103] Perform the mutation operation. On the basis of the traditional genetic algorithm, the present invention proposes a dynamic crossover probability adjustment method deeply coupled with the grid operation characteristics for distribution network fault detection. The crossover probability of the population is dynamically adjusted by using a preset crossover probability adjustment model; the mutation probability of the population is dynamically adjusted by using a preset mutation probability adjustment model;

[0104] Execute the adaptive adjustment mechanism. First, calculate the parameters of the current generation. During the optimization process of each generation, the standard deviation change rate of the grid data is calculated in real time and the fitness gain rate , which are used as the inputs of the adaptive adjustment formula, and the crossover rate and mutation rate are adjusted in real time, so as to dynamically adjust the optimization strategy of the genetic algorithm.

[0105] Iterative optimization process. According to the adjusted parameters, perform selection, crossover, and mutation operations to generate a new population, and evaluate the fitness of the population.

[0106] Repeat the iterative process until the fitness function reaches the preset threshold or the maximum number of iterations is reached. And evaluate the performance indicators of the optimized model on the test set to ensure the adaptability of the model under different grid operating states.

[0107] Until the calculation result of the fitness function reaches the preset threshold or the iteration reaches the preset maximum number of iterations, output the latest individual as the parameters of the Logistic regression model to obtain the fault probability prediction model.

[0108] To adapt to different grid working conditions, the present invention designs a multi-condition adaptive convergence mechanism to ensure the efficiency and accuracy of the optimization process.

[0109] High-load condition: Increase the crossover probability to be greater than 0.7 and the mutation probability to be greater than 0.5 to enhance the global search ability.

[0110] Low-fluctuation condition: Decrease the crossover probability to be less than 0.5 and the mutation probability to be less than 0.3 to enhance the local optimization ability.

[0111] After inspection, under different grid conditions, the fault location time is shortened by 25% and the detection accuracy reaches 95%.

[0112] The adaptive parameter adjustment mechanism proposed in this application breaks through the application limitations of the existing technology in the field of distribution network fault detection. It innovatively combines grid feature feedback and dynamic non-linear adjustment methods, significantly improving the efficiency and accuracy of fault detection. Through experimental verification, the performance of this solution in the multi-condition optimization scenario is superior to traditional methods, providing a strong guarantee for the operation safety of complex power grids.

[0113] In another embodiment provided by the present invention, the dynamic crossover probability adopted by the crossover probability adjustment model is ;

[0114] The mutation probability adjustment model is ;

[0115] The performance parameters include accuracy and recall rate;

[0116] The accuracy ;

[0117] The recall rate is ; where, is the crossover probability, and the change rate of the standard deviation of grid operation data , and are the standard deviations of the current generation population data and the previous generation population data respectively, is a preset value, and the fitness gain rate , and are the average fitness of the current generation population data and the previous generation population data; is the mutation probability. When the grid data fluctuates greatly (i.e., is large), the mutation rate increases to enhance the global search ability and prevent getting stuck in local optimal solutions. represents enhancing the non-linear dynamics of the mutation probability, and TP, TN, FP, and FN represent the numbers of true positive cases, true negative cases, false positive cases, and false negative cases respectively and are the minimum and maximum values of the crossover probability respectively, and They are the minimum and maximum values of the mutation probability respectively.

[0118] C r represents the crossover probability used in the current genetic algorithm, which determines the probability of gene crossover operation in the genetic algorithm. and are the minimum and maximum values of the crossover probability respectively, ensuring the adjustment range of the crossover probability.

[0119] M r represents the probability of gene mutation operation in the current genetic algorithm. and are the minimum and maximum values of the mutation probability respectively, restricting the range of the mutation probability. uses the sine function to adjust the mutation probability, making its fluctuation more flexible.

[0120] The proposed solution of this application overcomes the limitation of the traditional method that simply depends on the change of fitness by introducing the core features of power grid faults (such as voltage volatility and power change rate) into the adjustment of crossover probability, dynamically optimizes the feature weights through historical data and real-time feedback, enables the adjustment mechanism to accurately respond to different power grid conditions, solves the problem that the existing technology cannot quickly adapt to the high-volatility power grid conditions, and the convergence speed is increased by 25% in the experiment.

[0121] Realize the dynamic fluctuation of the mutation probability through the sine function, enabling it to be flexibly adjusted in different optimization stages and avoiding falling into local optima; dynamically adjust the mutation probability according to the characteristic fluctuations of different fault scenarios. For example, increase the mutation probability in high-fluctuation scenarios and decrease the mutation probability in stable scenarios; significantly enhance the global search ability of the algorithm, and the detection accuracy of complex power grid faults (such as multi-point faults) is increased by 12%.

[0122] In another embodiment provided by the present invention, the fault probability prediction model is ;

[0123] wherein, represents the initial weight vector, represents the input vector after weighting the feature vector, represents the feature vector, represents the feature weight vector optimized by the genetic algorithm, is the result output by the flexible DC distribution network fault prediction model, indicating the probability of fault occurring under the given feature vector x.

[0124] represents the optimized feature weight vector, which is the final weight parameter obtained through iterative optimization by the genetic algorithm. It is used to weight each input feature ( The contribution value is dynamically weighted to maximize the accuracy and robustness of fault prediction. The optimization combines the dynamic crossover and mutation strategies of the genetic algorithm, enabling it to adapt to changes in feature importance under multiple working conditions (such as high-load conditions, low-load conditions, etc.).

[0125] is the result output by the flexible DC distribution network fault prediction model, indicating that under the given feature vector the probability of a fault occurring. Its value range is [0, 1]. The closer the value is to 1, the higher the likelihood of a fault occurring, and the closer it is to 0, the more normal the grid operation is. The calculation combines dynamically optimized and multi-dimensional feature weights , significantly enhancing the model's prediction ability in complex fault scenarios.

[0126] Through an improved weighted optimization Logistic regression model, a fault probability prediction model based on equipment status is generated. The optimized model further combines the dynamic weights of different feature vectors, enabling more accurate prediction of the probability of fault events under different grid operating conditions.

[0127] The embodiment of the present invention also provides a flexible DC distribution network fault prediction device. Refer to Figure 2 , which is the structural schematic diagram of the flexible DC distribution network fault prediction device provided by the embodiment of the present invention. The device includes:

[0128] A collection module for collecting grid operation data of the flexible DC distribution network;

[0129] An extraction module for extracting features from the grid operation data to obtain a feature vector;

[0130] An optimization module for optimizing a preset initial weight vector using a preset optimization target model to adjust the feature weight distribution;

[0131] A calculation module for calculating the fault probability according to the feature weight distribution, the feature vector, and a pre-constructed fault probability prediction model.

[0132] It should be noted that the flexible DC distribution network fault prediction device provided by the embodiment of the present invention can execute the flexible DC distribution network fault prediction method described in any of the above embodiments. The specific functions of the flexible DC distribution network fault prediction device are not elaborated here.

[0133] Refer to Figure 3, which is a schematic structural diagram of a terminal device provided by an embodiment of the present invention. The terminal device in this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a flexible DC distribution network fault prediction program. When the processor executes the computer program, it implements the steps in each of the above embodiments of the flexible DC distribution network fault prediction method, such as Figure 1 the steps S1 to S4 shown. Or when the processor executes the computer program, it implements the functions of each module in each of the above device embodiments.

[0134] Exemplarily, the computer program can be divided into one or more modules / units. The one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of completing functions, and these instruction segments are used to describe the execution process of the computer program in the terminal device. For example, the computer program can be divided into various modules, and the specific functions of each module will not be elaborated again.

[0135] The terminal device can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the schematic diagram is only an example of the terminal device and does not constitute a limitation on the terminal device. It may include more or fewer components than shown, or combine some components, or different components. For example, the terminal device may further include input / output devices, network access devices, a bus, etc.

[0136] The so-called processor may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The processor is the control center of the terminal device and connects various parts of the entire terminal device through various interfaces and lines.

[0137] The memory can be used to store the computer program and / or modules. By running or executing the computer program and / or modules stored in the memory, and invoking the data stored in the memory, the processor realizes various functions of the terminal device. The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0138] Among them, if the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-mentioned embodiment methods of the present invention, it can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned various method embodiments can be realized. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, an executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0139] The above is the preferred implementation manner of the present invention. It should be noted that for those of ordinary skill in the art of this technology, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements are also regarded as the protection scope of the present invention.

Claims

1. A fault prediction method for a flexible DC distribution network, characterized in that, The method includes: Collecting the grid operation data of the flexible DC distribution network; Performing feature extraction on the grid operation data to obtain a feature vector; Optimizing a preset initial weight vector using a preset optimization objective model to adjust the feature weight allocation; Calculating the fault probability according to the feature weight allocation, the feature vector, and a pre-constructed fault probability prediction model; Determining the initial weight vector of the feature vector according to the obtained historical grid operation data; Using the gradient descent algorithm to perform real-time optimization of the initial weight vector with the goal of minimizing a preset optimization objective model, and adjusting the feature weight allocation; Among them, the optimization target model includes: ; Among them, is the loss function, that is, the optimization objective, represents the initial weight vector of the feature vector, represents the input vector after weighting the feature vector, and w represents the weight vector of the optimization objective model, represents the input feature vector of the i-th sample, represents the true label value of the i-th sample, represents the total number of samples of the input data; The fault probability prediction model is ; Among them, represents the input vector after weighting the feature vectors, represents the feature vector, represents the weight vector optimized by the genetic algorithm, is the result output by the flexible DC distribution network fault prediction model, indicating the probability of fault occurrence under the given feature vector x.

2. The fault prediction method for a flexible DC distribution network according to claim 1, wherein Performing feature extraction on the grid operation data to obtain a feature vector, including: Performing missing value processing and outlier processing on the grid operation data; Extracting the key features affecting the fault detection of the flexible DC distribution network according to the processed grid operation data to obtain the feature vector.

3. The flexible DC distribution network fault prediction method according to claim 2, wherein The key features include voltage volatility rate, power mutation rate, and load change rate; Among them, the voltage volatility , the power mutation rate , the load change rate , is the voltage at the i-th time point, is the average value of the voltage, N is the number of time points, and are the power values at the current time point and the previous time point respectively, and are the load values at the current time point and the previous time point respectively.

4. The fault prediction method for a flexible DC distribution network according to claim 1, characterized in that Determining the initial weight vector of the feature vector according to the obtained historical grid operation data, including: When the load change rate calculated according to the historical grid operation data exceeds a preset first threshold, it is recognized that the high-load condition is in place, and a preset first vector is set as the initial weight vector; When the proportion of the voltage volatility rate in the preset abnormal index range monitored according to the historical grid operation data exceeds a preset second threshold, it is recognized that the low-load condition is in place, and a preset second vector is set as the initial weight vector.

5. The fault prediction method for a flexible DC distribution network according to claim 3, wherein, The construction process of the fault probability prediction model specifically includes: Dividing the pre-collected feature data into a training set and a test set; Initializing the population of the genetic algorithm and randomly generating several initial solutions as initial individuals; Performing performance parameters of fault prediction through a pre-constructed fitness function; Selecting individuals with better fitness as the next generation using a preset selection method according to the performance index calculated by the fitness function; Using a preset crossover probability adjustment model to dynamically adjust the crossover probability of the population; Using a preset mutation probability adjustment model to dynamically adjust the mutation probability of the population; Calculating the parameters of the current generation of individuals. In the optimization process of each generation, the standard deviation change rate and fitness gain rate of the grid data are calculated in real time as the input of the fitness function, and the crossover probability and mutation probability are adjusted in real time for iteration until the calculation result of the fitness function reaches a preset threshold or the iteration reaches a preset maximum number of iterations; Outputting the latest individual as the parameters of the Logistic regression model to obtain the fault probability prediction model; Among them, the crossover probability adjustment model is ; The mutation probability adjustment model is ; The performance parameters include accuracy and recall rate; The accuracy rate ; The recall rate is ; Among them, is the crossover probability, the change rate of the standard deviation of the power grid operation data , and are the standard deviations of the current generation population data and the previous generation population data respectively, is a preset value, the fitness gain rate , and are the average fitnesses of the current generation population data and the previous generation population data respectively; is the mutation probability, represents the non - linear dynamics of enhancing the mutation probability, TP, TN, FP, and FN represent the numbers of true positives, true negatives, false positives, and false negatives respectively, and are the minimum and maximum values of the crossover probability respectively, and are the minimum and maximum values of the mutation probability respectively.

6. A flexible DC distribution network fault prediction device, characterized in that The device includes: A collection module for collecting the grid operation data of the flexible DC distribution network; An extraction module for performing feature extraction on the grid operation data to obtain a feature vector; An optimization module for optimizing a preset initial weight vector using a preset optimization objective model to adjust the feature weight allocation; A calculation module for calculating the fault probability according to the feature weight allocation, the feature vector, and a pre-constructed fault probability prediction model; Determine the initial weight vector of the feature vector according to the obtained historical power grid operation data; Use the gradient descent algorithm to minimize the preset optimization objective model, and perform real-time optimization on the initial weight vector to adjust the feature weight allocation; Among them, the optimized target model includes: ; Among them, is the loss function, that is, the optimization objective, represents the initial weight vector of the feature vector, represents the input vector after weighting the feature vector, and w represents the weight vector of the optimization objective model, represents the input feature vector of the i-th sample, represents the true label value of the i-th sample, represents the total number of samples of the input data; The failure probability prediction model is ; Among them, represents the input vector after weighting the feature vectors, represents the feature vectors, represents the weight vector optimized by the genetic algorithm, is the result output by the flexible DC distribution network fault prediction model, indicating the probability of fault occurrence under the given feature vector x.

7. A terminal device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the flexible DC distribution network fault prediction method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the flexible DC distribution network fault prediction method according to any one of claims 1 to 5.

Images