Power grid fault probability prediction method based on semantic vectorization normalization and adaptive regression

Abstract

The invention relates to the technical field of power grid fault prediction, and particularly provides a power grid fault probability prediction method based on semantic vectorization normalization and adaptive regression. On one hand, structured data and unstructured data in a power grid operation process can be simultaneously utilized to carry out model training; therefore, the fault prediction model can be combined with various information such as the power grid operation data, the meteorological data, the fault description information and the manual operation and maintenance information to perform feature learning, so that the model training effect can be improved, and the fault prediction accuracy can be improved. And on the other hand, at least two fault prediction models can be regularly trained again and back-tested, and the optimal model is automatically selected as the real-time prediction model according to the back-testing result, so that the fault prediction accuracy can be further improved.

Description

Technical Field

[0001] The present application relates to the technical field of power grid fault prediction, and in particular to a power grid fault probability prediction method based on semantic vectorization normalization and adaptive regression. Background Art

[0002] With the development of automated data collection technology and drone technology, real-time operation data of power grids and real-time environmental data of the environment in which power grids are located can be obtained more and more conveniently. At the same time, with the application and popularization of digital information, information manually determined by operation and maintenance inspection personnel, such as inspection records and equipment defect descriptions, has been uniformly stored and managed through local databases or cloud services.

[0003] After a long period of monitoring and operation and maintenance, the power grid has accumulated a large amount of historical fault event data, which can be used to predict the fault probability of the power grid. How to use these historical fault event data is one of the key factors that determine the accuracy of the prediction. Therefore, it is urgent to provide an effective power grid fault probability prediction solution to achieve accurate prediction. Summary of the invention

[0004] The purpose of this application is to solve at least one of the above-mentioned technical defects, especially the technical defect of low prediction accuracy in the prior art.

[0005] In a first aspect, an embodiment of the present application provides a method for predicting power grid fault probability based on semantic vectorization normalization and adaptive regression, the method comprising:

[0006] In response to the arrival of the current update cycle, historical fault event data corresponding to the previous update cycle is obtained; wherein the historical fault event data includes structured data and unstructured data, the structured data includes power grid operation data and meteorological data, and the unstructured data includes fault description information and manual operation and maintenance information;

[0007] Encoding the unstructured data using a one-hot encoding algorithm to obtain a semantic feature vector;

[0008] Classifying the unstructured data according to the semantic feature vector and obtaining a semantic description type; wherein the semantic description type is used to reflect the power grid state described by the content of the unstructured data;

[0009] Using the semantic description types corresponding to the structured data and the unstructured data as training sets, model training is performed on at least two first fault prediction models, and at least two second fault prediction models are obtained; wherein the first fault prediction model is a fault prediction model obtained by training in the previous update cycle, and the second fault prediction model is a fault prediction model obtained by training in the current update cycle;

[0010] Backtesting each of the second fault prediction models respectively, and determining an optimal model among the second fault prediction models according to the backtesting results of each of the second fault prediction models;

[0011] The optimal model is deployed as a real-time prediction model for predicting the probability of power grid failure.

[0012] In some embodiments, the classifying and processing the unstructured data according to the semantic feature vector to obtain a semantic description type includes:

[0013] Calculating a vector distance between the semantic feature vector and a preset reference feature vector;

[0014] The unstructured data is classified according to the vector distance to obtain the semantic description type.

[0015] In some embodiments, at least two of the first fault prediction models include a first linear regression model and a first XGBoost regression model, and at least two of the second fault prediction models include a second linear regression model and a second XGBoost regression model.

[0016] In some embodiments, backtesting each of the second fault prediction models includes:

[0017] Backtesting the second linear regression model using a backtest data set, and obtaining a linear regression prediction result output by the second linear regression model, and transforming the linear regression prediction result to obtain a first fault prediction probability;

[0018] Backtesting the second XGBoost regression model using the backtest data set, and obtaining a second fault prediction probability output by the second XGBoost regression model;

[0019] Calculating a third fault prediction probability according to a preset weight coefficient, the first fault prediction probability and the second fault prediction probability;

[0020] Calculate a first backtest result according to a preset backtest algorithm and the first fault prediction probability;

[0021] Calculating a second backtest result according to the preset backtest algorithm and the second fault prediction probability;

[0022] A third backtest result is calculated according to the preset backtest algorithm and the third fault prediction probability.

[0023] In some embodiments, the preset backtesting algorithms include a mean absolute error (MAE) algorithm, a mean square error (MSE) algorithm, and an R-squared algorithm.

[0024] In some embodiments, calculating the first backtest result according to a preset backtest algorithm and the first fault prediction probability includes:

[0025] Based on the first fault prediction probability, using the mean absolute error (MAE) algorithm to calculate a first MAE value;

[0026] Based on the first fault prediction probability, using the mean square error (MSE) algorithm to calculate a first MSE value;

[0027] Based on the first fault prediction probability, using the R-squared algorithm to calculate a first result, and taking the difference between 1 and the first result as a first R-squared value;

[0028] Performing a weighted summation on the first MAE value, the first MSE value, and the first R-squared value, and obtaining the first backtest result;

[0029] The calculating a second backtest result according to the preset backtest algorithm and the second fault prediction probability includes:

[0030] Based on the second fault prediction probability, calculating a second MAE value using the mean absolute error (MAE) algorithm;

[0031] Based on the second fault prediction probability, using the mean square error (MSE) algorithm to calculate a second MSE value;

[0032] Based on the second fault prediction probability, using the R-squared algorithm to calculate a second result, and subtracting the difference between 1 and the second result as a second R-squared value;

[0033] Performing a weighted summation on the second MAE value, the second MSE value, and the second R-squared value, and obtaining the second backtest result;

[0034] The calculating a third backtest result according to the preset backtest algorithm and the third fault prediction probability includes:

[0035] Based on the third fault prediction probability, calculating a third MAE value using the mean absolute error (MAE) algorithm;

[0036] Based on the third fault prediction probability, using the mean square error (MSE) algorithm to calculate a third MSE value;

[0037] Based on the third fault prediction probability, the R-squared algorithm is used to calculate a third result, and the difference between 1 and the third result is used as a third R-squared value;

[0038] A weighted sum is performed on the third MAE value, the third MSE value, and the third R-squared value to obtain the third backtest result.

[0039] In a second aspect, an embodiment of the present application provides a power grid fault probability prediction device based on semantic vectorization normalization and adaptive regression, the device comprising:

[0040] A data acquisition module, configured to acquire historical fault event data corresponding to a previous update cycle in response to the arrival of a current update cycle; wherein the historical fault event data includes structured data and unstructured data, the structured data includes power grid operation data and meteorological data, and the unstructured data includes fault description information and manual operation and maintenance information;

[0041] An encoding module, used for encoding the unstructured data using a one-hot encoding algorithm to obtain a semantic feature vector;

[0042] A classification module, used to classify the unstructured data according to the semantic feature vector and obtain a semantic description type; wherein the semantic description type is used to reflect the power grid state described by the content of the unstructured data;

[0043] A model training module, used to perform model training on at least two first fault prediction models using the semantic description types corresponding to the structured data and the unstructured data as training sets, and obtain at least two second fault prediction models; wherein the first fault prediction model is a fault prediction model obtained by training in the previous update cycle, and the second fault prediction model is a fault prediction model obtained by training in the current update cycle;

[0044] A backtesting module, used to backtest each of the second fault prediction models respectively, and determine the optimal model among the second fault prediction models according to the backtesting result of each of the second fault prediction models;

[0045] A model deployment module is used to deploy the optimal model as a real-time prediction model for predicting the probability of power grid failure.

[0046] In a third aspect, the present application further provides a computer device. The computer device includes: one or more processors, and a memory;

[0047] The memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the power grid fault probability prediction method based on semantic vectorization normalization and adaptive regression described in any of the above embodiments are performed.

[0048] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for predicting power grid fault probability based on semantic vectorization normalization and adaptive regression described in any of the above embodiments.

[0049] In a fifth aspect, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the power grid fault probability prediction method based on semantic vectorization normalization and adaptive regression described in any of the above embodiments are implemented.

[0050] In a power grid fault probability prediction method based on semantic vectorization normalization and adaptive regression provided in some embodiments of the present application, on the one hand, the structured data and unstructured data in the power grid operation process can be used simultaneously for model training, so that the fault prediction model can combine multiple information such as power grid operation data, meteorological data, fault description information and manual operation and maintenance information for feature learning, thereby improving the model training effect, and further improving the accuracy of fault prediction. On the other hand, the present application can regularly retrain and backtest at least two fault prediction models, and automatically select the optimal model as the real-time prediction model based on the backtest results, thereby further improving the accuracy of fault prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0052] Figure 1 A schematic diagram of a flow chart of a method for predicting power grid fault probability based on semantic vectorization normalization and adaptive regression in some embodiments;

[0053] Figure 2 A schematic diagram of the structure of a power grid fault probability prediction device based on semantic vectorization normalization and adaptive regression in some embodiments;

[0054] Figure 3 This is a diagram of the internal structure of a computer device in some embodiments. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0056] In some embodiments, the present application provides a method for predicting power grid fault probability based on semantic vectorization normalization and adaptive regression. The following embodiments are described by taking the method applied to a computer device as an example. It can be understood that the computer device described in the present application can be any device with data acquisition and data processing functions, and can be but not limited to desktop computers, laptops, notebook computers, tablet computers, servers, smart phones, Internet of Things devices, etc.

[0057] like Figure 1 As shown, the power grid fault probability prediction method based on semantic vectorization normalization and adaptive regression of the present application may include the following steps:

[0058] S102: In response to the arrival of the current update cycle, historical fault event data corresponding to the previous update cycle is obtained.

[0059] Among them, historical fault event data refers to historical data related to power grid fault events, which may include structured data and unstructured data. Furthermore, historical fault event data may also include a predetermined true fault probability, which may be manually determined or automatically generated using other algorithms, and this application does not impose specific restrictions on this. Exemplarily, the process of determining the true fault probability may be: first, obtain the historical operation data of the power grid, and classify the historical operation data, so that the operation data corresponding to the same working condition are classified into the same category. Then, according to the historical operating status of the power grid, the fault probability corresponding to each working condition is counted separately to obtain the true fault probability. For example, the power grid operates 10 times under a certain working condition, 9 of which are without fault and 1 is faulty, then the true fault probability corresponding to the working condition may be 10%.

[0060] Among them, structured data may be data with precise formats and rules, for example, data collected by sensor acquisition equipment. In the present application, structured data may include power grid operation data and meteorological data. Power grid operation data refers to data that can be used to reflect the operation of the power grid, including but not limited to at least one of the voltage data, current data, active power data, reactive power data, frequency data and other data of the power grid before and after the fault event occurs. Meteorological data refers to environmental data of the environment in which the power grid facilities are located, including but not limited to at least one of the temperature data, humidity data, wind direction data, irradiance data, lightning probability data and other data.

[0061] Unstructured data may be data with irregular data structure, for example, text information input by operation and maintenance personnel. In the present application, unstructured data may include fault description information and manual operation and maintenance information. Fault description information refers to text used to describe the fault situation of the power grid, including but not limited to at least one of the data such as fault location text, fault type text, and environmental description text of the fault location. Manual operation and maintenance information refers to text data generated during the operation and maintenance process of the power grid and used to describe the operation and maintenance situation. Exemplarily, manual operation and maintenance information may include data such as inspection records and equipment defect history records.

[0062] In this application, the computer device may periodically execute steps S102 to S112 according to the cycle duration of the update cycle, so as to periodically train and backtest the fault prediction model according to the historical fault event data. Specifically, when the current update cycle arrives, the computer device may obtain the historical fault event data generated or recorded in the previous update cycle, so as to perform model training through the latest generated or recorded historical fault event data. It can be understood that the cycle duration of the update cycle can be determined according to the actual situation, and this application does not impose specific restrictions on this.

[0063] S104: Encode the unstructured data using a one-hot encoding algorithm and obtain a semantic feature vector.

[0064] In this step, considering that the data content of unstructured data is a description of the fault event, its data content is one of the characteristics of the fault event, and it is necessary to use unstructured data for model training. However, since unstructured data is not a specific value, it cannot be directly involved in model training. Therefore, it is necessary to vectorize the unstructured data so that the vectorized structure can reflect the data content of the unstructured data.

[0065] Specifically, the computer equipment can use the One-Hot Encoding algorithm to encode the fault description information and manual operation and maintenance information, so as to vectorize and normalize the unstructured data and obtain the semantic feature vector. It can be understood that the semantic feature vector can represent the data content in the unstructured data in the form of a vector.

[0066] Among them, the one-hot encoding algorithm refers to the use of an N-bit state register to encode N states, which is a representation of a categorical variable as a binary vector. In the process of encoding using the one-hot encoding algorithm, a dictionary can be first constructed based on the provided text. Then, based on the one-hot encoding expression, an N-dimensional vector corresponding to the unstructured data (for example, an N-dimensional vector corresponding to the fault description information) is constructed, and the dimension of the vector is consistent with the length of the dictionary. When a given word is expressed in a vector, the register at the corresponding position where it appears in the dictionary is assigned a value of 1, and the rest are 0. The assignment can be regarded as the label information of the corresponding word or the classification information of the object. In some examples, the text construction dictionary can be constructed based on power keywords, for example, it may include keywords such as grounding, overcurrent, and external force damage.

[0067] S106: Classify the unstructured data according to the semantic feature vector and obtain a semantic description type.

[0068] In this step, since the semantic feature vector can reflect the data content of the unstructured data, the power keyword state presented by the semantic feature vector can be determined, and classification can be implemented accordingly, thereby obtaining a semantic description type that can reflect the power grid state described by the content of the unstructured data. For example, the semantic description type may include: grounding, overcurrent, external force damage, and the simultaneous existence of grounding and overcurrent.

[0069] It is understandable that the computer device may implement classification processing in any manner and obtain the semantic description type, for example, classification may be implemented through a neural network model. In some examples, S106 may include the following sub-steps:

[0070] Step A1: Calculate the vector distance between the semantic feature vector and the preset reference feature vector;

[0071] Step A2: Classify the unstructured data according to the vector distance and obtain the semantic description type.

[0072] In this example, the reference feature vector can be pre-set according to the actual situation such as the vector format of the unique hot encoding algorithm, for example, a full 0 vector can be used as the reference feature vector. The computer device can calculate the vector distance between the semantic feature vector and the reference feature vector to reflect the gap between the semantic feature vector and the reference feature vector through the vector distance. It can be understood that the vector distance can be any type of vector distance, for example, it can be the Euclidean distance, and this application does not impose any specific restrictions on this.

[0073] In this example, by calculating the vector distance between the semantic feature vector and the preset reference feature vector and performing classification based on the distance, both classification accuracy and classification efficiency can be considered.

[0074] S108: Using the semantic description types corresponding to the structured data and the unstructured data as training sets, perform model training on at least two first fault prediction models, and obtain at least two second fault prediction models.

[0075] The first fault prediction model is a fault prediction model obtained by training in the previous update cycle, and the second fault prediction model is a fault prediction model obtained by training in the current update cycle.

[0076] In this step, in each update cycle, the computer device can perform model training on multiple first fault prediction models trained in the previous update cycle based on the historical fault event data corresponding to the previous update cycle, and obtain the second fault prediction model trained in the current update cycle.

[0077] Specifically, the computer device can use structured data such as power grid operation data and meteorological data, as well as the semantic description type of unstructured data as training sets, and perform model training on multiple first fault prediction models obtained by training in the previous update cycle until the preset training end condition is met, and obtain the second fault prediction model corresponding to the current update cycle. The multiple second fault prediction models corresponding to the current update cycle are used as the multiple first fault prediction models corresponding to the next update cycle.

[0078] It can be understood that the model types of multiple fault prediction models can be selected according to actual conditions, and this application does not impose specific restrictions on this. In some examples, linear regression models and XGBoost regression models can be used as fault prediction models. That is, in the current update cycle, the training set corresponding to the current update cycle can be used to perform synchronous machine learning training on the first linear regression model and the first XGBoost regression model trained in the previous update cycle, and obtain the trained second linear regression model and the second XGBoost regression model. In this way, both prediction efficiency and prediction accuracy can be taken into account.

[0079] S110: backtesting each second fault prediction model respectively, and determining an optimal model among the second fault prediction models according to the backtesting result of each second fault prediction model.

[0080] In this step, after completing the model training of the current update cycle, the computer device can backtest each second fault prediction model and obtain the backtest results corresponding to each second fault prediction model. Since the backtest results can reflect the prediction accuracy of the second fault prediction model, after obtaining the backtest results of each second fault prediction model, the present application can determine the optimal model with the best prediction effect among multiple second fault prediction models based on each backtest result.

[0081] It is understood that the present application may implement model backtesting in any manner. In some examples, backtesting each second fault prediction model may include the following sub-steps:

[0082] Step B1: backtesting the second linear regression model using the backtest data set, and obtaining a linear regression prediction result output by the second linear regression model, and transforming the linear regression prediction result to obtain a first fault prediction probability;

[0083] Step B2: backtesting the second XGBoost regression model using the backtest data set, and obtaining a second fault prediction probability output by the second XGBoost regression model;

[0084] Step B3: Calculating a third fault prediction probability according to a preset weight coefficient, the first fault prediction probability and the second fault prediction probability;

[0085] Step B4: Calculate a first backtest result according to a preset backtest algorithm and a first fault prediction probability;

[0086] Step B5: Calculate the second backtest result according to the preset backtest algorithm and the second fault prediction probability;

[0087] Step B6: Calculate the third backtest result according to the preset backtest algorithm and the third fault prediction probability.

[0088] In this example, the backtest data set can be a subset of the training set, that is, the computer device can select part of the data from the training set as the backtest data set, and use the backtest data set to backtest the second linear regression model, the second XGBoost regression model and the combined model respectively. Among them, the combined model refers to a model obtained by combining the second linear regression model and the second XGBoost regression model according to a preset weight coefficient.

[0089] Specifically, the computer device may input the structured data and unstructured data in the backtest data set into the second linear regression model, and obtain the linear regression prediction result output by the second linear regression model. The linear regression prediction result is transformed to obtain the first fault prediction probability. For the second XGBoost regression model, the computer device may input the structured data and unstructured data in the backtest data set into the second XGBoost regression model, and obtain the second fault prediction probability output by the second XGBoost regression model.

[0090] When the first fault prediction probability and the second fault prediction probability are obtained, the computer device may perform weighted summation of the first fault prediction probability and the second fault prediction probability according to a preset weight coefficient to obtain a third fault prediction probability corresponding to the combined model. It is understood that the specific value of the weight coefficient may be determined according to actual conditions, for example, it may be 0.5 / 0.5 (i.e., the third fault prediction probability is the average value of the first fault prediction probability and the second fault prediction probability), and this application does not impose specific restrictions on this.

[0091] When the fault prediction probability of the model output is obtained, the computer device can compare the model prediction result with the entire training set, and use the preset backtesting algorithm to obtain the backtesting results corresponding to the second linear regression model (i.e., the first backtesting result), the backtesting results corresponding to the second XGBoost regression model (i.e., the second backtesting result), and the backtesting results corresponding to the combined model (i.e., the third backtesting result).

[0092] It can be understood that the preset backtesting algorithm can be one or more, and the number and type of algorithms can be determined according to the actual situation. In some examples, the preset backtesting algorithm may include the mean absolute error (MAE) algorithm, the mean square error (MSE) algorithm, and the R-squared algorithm to comprehensively and accurately determine the backtesting effect of the fault prediction model. Among them, the calculation formula of the mean absolute error (MAE) algorithm can be as follows:

[0093]

[0094] Where m is the number of data samples in the backtest dataset. is the true failure probability, is the predicted failure probability output by the model.

[0095] The calculation formula of the mean square error MSE algorithm can be shown as follows:

[0096]

[0097] The calculation formula of the R-squared algorithm can be shown as follows:

[0098]

[0099] Where n is the number of data samples in the backtest dataset. is the average value of all true failure probabilities corresponding to the backtest data set.

[0100] In some examples, when the preset backtesting algorithm may include a mean absolute error (MAE) algorithm, a mean square error (MSE) algorithm, and an R-squared algorithm, the step of calculating the first backtesting result may include:

[0101] Step B41: Based on the first fault prediction probability, a first MAE value is calculated using a mean absolute error (MAE) algorithm;

[0102] Step B42: Based on the first fault prediction probability, a first MSE value is calculated using a mean square error (MSE) algorithm;

[0103] Step B43: Based on the first fault prediction probability, an R-squared algorithm is used to calculate a first result, and the difference between 1 and the first result is used as a first R-squared value.

[0104] Step B44: perform weighted summation on the first MAE value, the first MSE value, and the first R-squared value to obtain a first backtest result.

[0105] In this example, the computer device may use the mean absolute error MAE algorithm, the mean square error MSE algorithm, and the R-squared algorithm to calculate the gap between the first fault prediction probability and the actual fault probability, and obtain the first MAE value, the first MSE value, and the first R-squared value, respectively. The computer device may perform a weighted summation of the first MAE value, the first MSE value, and the first R-squared value to obtain the backtest effect corresponding to the second linear regression model. In some examples, the weights corresponding to the first MAE value, the first MSE value, and the first R-squared value are all 1 / 3.

[0106] In some examples, when the preset backtesting algorithm may include a mean absolute error (MAE) algorithm, a mean square error (MSE) algorithm, and an R-squared algorithm, the step of calculating the second backtesting result may include:

[0107] Step B51: Based on the second fault prediction probability, a second MAE value is calculated using a mean absolute error (MAE) algorithm;

[0108] Step B52: Based on the second fault prediction probability, a second MSE value is calculated using a mean square error (MSE) algorithm;

[0109] Step B53: Based on the second fault prediction probability, the R-squared algorithm is used to calculate a second result, and the difference between 1 and the second result is taken as a second R-squared value;

[0110] Step B54: perform weighted summation on the second MAE value, the second MSE value, and the second R-squared value to obtain a second backtest result.

[0111] In this example, the computer device may use the mean absolute error MAE algorithm, the mean square error MSE algorithm, and the R-squared algorithm to calculate the gap between the second fault prediction probability and the actual fault probability, and obtain the second MAE value, the second MSE value, and the second R-squared value, respectively. The computer device may perform weighted summation on the second MAE value, the second MSE value, and the second R-squared value to obtain the backtest effect corresponding to the second XGBoost regression model. In some examples, the weights corresponding to the second MAE value, the second MSE value, and the second R-squared value are all 1 / 3.

[0112] In some examples, when the preset backtesting algorithm may include a mean absolute error (MAE) algorithm, a mean square error (MSE) algorithm, and an R-squared algorithm, the calculation step of the third backtesting result may include:

[0113] Step B61: Based on the third fault prediction probability, a third MAE value is calculated using a mean absolute error (MAE) algorithm;

[0114] Step B62: Based on the third fault prediction probability, a third MSE value is calculated using a mean square error (MSE) algorithm;

[0115] Step B63: Based on the third fault prediction probability, the R-squared algorithm is used to calculate a third result, and the difference between 1 and the third result is taken as a third R-squared value;

[0116] Step B64: perform weighted summation on the third MAE value, the third MSE value and the third R-squared value to obtain a third backtest result.

[0117] In this example, the computer device may respectively use the mean absolute error MAE algorithm, the mean square error MSE algorithm, and the R-squared algorithm to calculate the gap between the third fault prediction probability and the actual fault probability, and obtain the third MAE value, the third MSE value, and the third R-squared value, respectively. The computer device may perform weighted summation on the third MAE value, the third MSE value, and the third R-squared value to obtain the backtest effect corresponding to the combined model. In some examples, the weights corresponding to the third MAE value, the third MSE value, and the third R-squared value are all 1 / 3.

[0118] S112: deploying the optimal model as a real-time prediction model for predicting the probability of power grid failure.

[0119] In this step, the computer device can automatically select the optimal model with the highest prediction accuracy as the real-time prediction model based on the backtest results corresponding to the current update cycle, and deploy the real-time prediction model online to use the real-time prediction model to predict the failure rate of lines or equipment at various points in the power grid in the future.

[0120] In the above embodiments, on the one hand, the structured data and unstructured data in the operation process of the power grid can be used simultaneously for model training, so that the fault prediction model can combine multiple information such as power grid operation data, meteorological data, fault description information and manual operation and maintenance information for feature learning, thereby improving the model training effect, and further improving the accuracy of fault prediction. On the other hand, the present application can regularly retrain and backtest at least two fault prediction models, and automatically select the optimal model as the real-time prediction model according to the backtest results, thereby further improving the accuracy of fault prediction.

[0121] The following is a description of a power grid fault probability prediction device based on semantic vectorization normalization and adaptive regression provided in an embodiment of the present application. The power grid fault probability prediction device based on semantic vectorization normalization and adaptive regression described below and the power grid fault probability prediction method based on semantic vectorization normalization and adaptive regression described above can be referenced to each other.

[0122] In some embodiments, Figure 2 As shown, the present application provides a power grid fault probability prediction device 200 based on semantic vectorization normalization and adaptive regression, comprising:

[0123] The data acquisition module 202 is used to acquire historical fault event data corresponding to the previous update cycle in response to the arrival of the current update cycle; wherein the historical fault event data includes structured data and unstructured data, the structured data includes power grid operation data and meteorological data, and the unstructured data includes fault description information and manual operation and maintenance information;

[0124] An encoding module 204, configured to encode the unstructured data using a one-hot encoding algorithm and obtain a semantic feature vector;

[0125] A classification module 206 is used to classify the unstructured data according to the semantic feature vector and obtain a semantic description type; wherein the semantic description type is used to reflect the power grid state described by the content of the unstructured data;

[0126] A model training module 208, configured to use the semantic description types corresponding to the structured data and the unstructured data as training sets to perform model training on at least two first fault prediction models, and obtain at least two second fault prediction models; wherein the first fault prediction model is a fault prediction model obtained by training in the previous update cycle, and the second fault prediction model is a fault prediction model obtained by training in the current update cycle;

[0127] A backtesting module 210, configured to backtest each of the second fault prediction models respectively, and determine an optimal model among the second fault prediction models according to the backtesting result of each of the second fault prediction models;

[0128] The model deployment module 212 is used to deploy the optimal model as a real-time prediction model for predicting power grid failure probability.

[0129] In some embodiments, the classification module 206 of the present application includes:

[0130] A distance calculation unit, used for calculating the vector distance between the semantic feature vector and a preset reference feature vector;

[0131] A classification unit is used to classify the unstructured data according to the vector distance and obtain the semantic description type.

[0132] In some embodiments, at least two of the first fault prediction models include a first linear regression model and a first XGBoost regression model, and at least two of the second fault prediction models include a second linear regression model and a second XGBoost regression model.

[0133] In some embodiments, the backtest module 210 of the present application includes:

[0134] A first probability acquisition unit is used to backtest the second linear regression model using a backtest data set, obtain a linear regression prediction result output by the second linear regression model, and transform the linear regression prediction result to obtain a first fault prediction probability;

[0135] A second probability acquisition unit, configured to backtest the second XGBoost regression model using the backtest data set, and obtain a second fault prediction probability output by the second XGBoost regression model;

[0136] A third probability acquisition unit, configured to calculate a third fault prediction probability according to a preset weight coefficient, the first fault prediction probability and the second fault prediction probability;

[0137] A first result acquisition unit, used to calculate a first backtest result according to a preset backtest algorithm and the first fault prediction probability;

[0138] A second result acquisition unit, used to calculate a second backtest result according to the preset backtest algorithm and the second fault prediction probability;

[0139] The third result acquisition unit is used to calculate a third backtest result according to the preset backtest algorithm and the third fault prediction probability.

[0140] In some embodiments, the preset backtesting algorithms include a mean absolute error (MAE) algorithm, a mean square error (MSE) algorithm, and an R-squared algorithm.

[0141] In some embodiments, the first result obtaining unit of the present application includes:

[0142] A first score calculation unit, configured to calculate a first MAE value based on the first fault prediction probability by using the mean absolute error (MAE) algorithm;

[0143] A second score calculation unit, configured to calculate a first MSE value by using the mean square error (MSE) algorithm based on the first fault prediction probability;

[0144] a third score calculation unit, configured to calculate a first result by using the R-squared algorithm based on the first fault prediction probability, and to use a difference between 1 and the first result as a first R-squared value;

[0145] The first weighted summation unit is used to perform weighted summation on the first MAE value, the first MSE value and the first R-squared value to obtain the first backtest result.

[0146] The second result acquisition unit of the present application includes:

[0147] a fourth score calculation unit, configured to calculate a second MAE value based on the second fault prediction probability by using the mean absolute error (MAE) algorithm;

[0148] a fifth score calculation unit, configured to calculate a second MSE value by using the mean square error (MSE) algorithm based on the second fault prediction probability;

[0149] a sixth score calculation unit, configured to calculate a second result by using the R-squared algorithm based on the second fault prediction probability, and to use a difference between 1 and the second result as a second R-squared value;

[0150] The second weighted summation unit is used to perform weighted summation on the second MAE value, the second MSE value and the second R-squared value to obtain the second backtest result.

[0151] The third result acquisition unit of the present application includes:

[0152] a seventh score calculation unit, configured to calculate a third MAE value based on the third fault prediction probability by using the mean absolute error (MAE) algorithm;

[0153] an eighth score calculation unit, configured to calculate a third MSE value by using the mean square error (MSE) algorithm based on the third fault prediction probability;

[0154] a ninth score calculation unit, configured to calculate a third result by using the R-squared algorithm based on the third fault prediction probability, and to use a difference between 1 and the third result as a third R-squared value;

[0155] The third weighted summation unit is used to perform weighted summation on the third MAE value, the third MSE value and the third R-squared value to obtain the third backtest result.

[0156] In one embodiment, the present application also provides a computer device having computer-readable instructions stored therein. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the power grid fault probability prediction method based on semantic vectorization normalization and adaptive regression as in any embodiment.

[0157] Indicatively, Figure 3 The internal structure diagram of a computer device provided in an embodiment of the present application is shown in FIG. 1 . In one example, the computer device may be a server. Figure 3 The computer device 900 includes a processing component 902, which further includes one or more processors, and a memory resource represented by a memory 901, for storing instructions executable by the processing component 902, such as an application. The application stored in the memory 901 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 902 is configured to execute instructions to perform the steps of the power grid fault probability prediction method based on semantic vectorization normalization and adaptive regression described in any of the above embodiments.

[0158] The computer device 900 may further include a power supply component 903 configured to perform power management of the computer device 900, a wired or wireless network interface 904 configured to connect the computer device 900 to a network, and an input / output (I / O) interface 905. The computer device 900 may operate based on an operating system stored in the memory 901, such as Windows Server TM, Mac OS X TM, Unix TM, Linux TM, Free BSD TM, or the like.

[0159] Those skilled in the art will understand that the internal structure of the computer device shown in the present application is merely a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0160] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the power grid fault probability prediction method based on semantic vectorization normalization and adaptive regression described in any of the above embodiments are implemented.

[0161] In one embodiment, a computer program product is provided, including a computer program, which, when executed by a processor, implements the steps of the method for predicting power grid fault probability based on semantic vectorization normalization and adaptive regression described in any of the above embodiments.

[0162] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the term "include", "comprise" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not clearly listed, or also includes elements inherent to such process, method, article or equipment. In the absence of more restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, article or equipment including the elements. Herein, "one", "one", "said", "the" and "it" may also include plural forms, unless the context clearly indicates another way. A plurality refers to at least two cases, such as 2, 3, 5 or 8, etc. "And / or" includes any and all combinations of the relevant listed items.

[0163] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can refer to each other.

[0164] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting power grid fault probability based on semantic vectorization normalization and adaptive regression, characterized in that: The method comprises: In response to the arrival of the current update cycle, historical fault event data corresponding to the previous update cycle is obtained; wherein the historical fault event data includes structured data and unstructured data, the structured data includes power grid operation data and meteorological data, and the unstructured data includes fault description information and manual operation and maintenance information; Encoding the unstructured data using a one-hot encoding algorithm to obtain a semantic feature vector; Classifying the unstructured data according to the semantic feature vector and obtaining a semantic description type; wherein the semantic description type is used to reflect the power grid state described by the content of the unstructured data; Using the semantic description types corresponding to the structured data and the unstructured data as training sets, model training is performed on at least two first fault prediction models, and at least two second fault prediction models are obtained; wherein the first fault prediction model is a fault prediction model obtained by training in the previous update cycle, and the second fault prediction model is a fault prediction model obtained by training in the current update cycle; Backtesting each of the second fault prediction models respectively, and determining an optimal model among the second fault prediction models according to the backtesting results of each of the second fault prediction models; The optimal model is deployed as a real-time prediction model for predicting the probability of power grid failure.

2. The method according to claim 1, characterized in that: The classifying and processing the unstructured data according to the semantic feature vector and obtaining a semantic description type includes: Calculating a vector distance between the semantic feature vector and a preset reference feature vector; The unstructured data is classified according to the vector distance to obtain the semantic description type.

3. The method according to claim 1 or 2, characterized in that: At least two of the first fault prediction models include a first linear regression model and a first XGBoost regression model, and at least two of the second fault prediction models include a second linear regression model and a second XGBoost regression model.

4. The method according to claim 3, characterized in that The backtesting each of the second fault prediction models respectively includes: Backtesting the second linear regression model using a backtest data set, and obtaining a linear regression prediction result output by the second linear regression model, and transforming the linear regression prediction result to obtain a first fault prediction probability; Backtesting the second XGBoost regression model using the backtest data set, and obtaining a second fault prediction probability output by the second XGBoost regression model; Calculating a third fault prediction probability according to a preset weight coefficient, the first fault prediction probability and the second fault prediction probability; Calculate a first backtest result according to a preset backtest algorithm and the first fault prediction probability; Calculating a second backtest result according to the preset backtest algorithm and the second fault prediction probability; A third backtest result is calculated according to the preset backtest algorithm and the third fault prediction probability.

5. The method according to claim 4, characterized in that The preset backtesting algorithms include mean absolute error (MAE) algorithm, mean square error (MSE) algorithm and R-squared algorithm.

6. The method according to claim 5, characterized in that The calculating a first backtest result according to a preset backtest algorithm and the first fault prediction probability includes: Based on the first fault prediction probability, using the mean absolute error (MAE) algorithm to calculate a first MAE value; Based on the first fault prediction probability, using the mean square error (MSE) algorithm to calculate a first MSE value; Based on the first fault prediction probability, using the R-squared algorithm to calculate a first result, and taking the difference between 1 and the first result as a first R-squared value; Performing a weighted summation on the first MAE value, the first MSE value, and the first R-squared value, and obtaining the first backtest result; The calculating a second backtest result according to the preset backtest algorithm and the second fault prediction probability includes: Based on the second fault prediction probability, calculating a second MAE value using the mean absolute error (MAE) algorithm; Based on the second fault prediction probability, using the mean square error (MSE) algorithm to calculate a second MSE value; Based on the second fault prediction probability, using the R-squared algorithm to calculate a second result, and subtracting the difference between 1 and the second result as a second R-squared value; Performing a weighted summation on the second MAE value, the second MSE value, and the second R-squared value, and obtaining the second backtest result; The calculating a third backtest result according to the preset backtest algorithm and the third fault prediction probability includes: Based on the third fault prediction probability, calculating a third MAE value using the mean absolute error (MAE) algorithm; Based on the third fault prediction probability, using the mean square error (MSE) algorithm to calculate a third MSE value; Based on the third fault prediction probability, the R-squared algorithm is used to calculate a third result, and the difference between 1 and the third result is used as a third R-squared value; A weighted sum is performed on the third MAE value, the third MSE value, and the third R-squared value to obtain the third backtest result.

7. A power grid fault probability prediction device based on semantic vectorization normalization and adaptive regression, characterized in that: The device comprises: A data acquisition module, configured to acquire historical fault event data corresponding to a previous update cycle in response to the arrival of a current update cycle; wherein the historical fault event data includes structured data and unstructured data, the structured data includes power grid operation data and meteorological data, and the unstructured data includes fault description information and manual operation and maintenance information; An encoding module, used for encoding the unstructured data using a one-hot encoding algorithm to obtain a semantic feature vector; A classification module, used to classify the unstructured data according to the semantic feature vector and obtain a semantic description type; wherein the semantic description type is used to reflect the power grid state described by the content of the unstructured data; A model training module, used to perform model training on at least two first fault prediction models using the semantic description types corresponding to the structured data and the unstructured data as training sets, and obtain at least two second fault prediction models; wherein the first fault prediction model is a fault prediction model obtained by training in the previous update cycle, and the second fault prediction model is a fault prediction model obtained by training in the current update cycle; A backtesting module, used to backtest each of the second fault prediction models respectively, and determine the optimal model among the second fault prediction models according to the backtesting result of each of the second fault prediction models; A model deployment module is used to deploy the optimal model as a real-time prediction model for predicting the probability of power grid failure.

8. A computer device, characterized in that: include: one or more processors, and memory; The memory stores computer-readable instructions, which, when executed by the one or more processors, perform the steps of the power grid fault probability prediction method based on semantic vectorization normalization and adaptive regression as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting power grid fault probability based on semantic vectorization normalization and adaptive regression as described in any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method for predicting power grid fault probability based on semantic vectorization normalization and adaptive regression as described in any one of claims 1 to 6 are implemented.

Images