Power transaction center risk assessment system and method based on multi-risk superposition

Abstract

The invention provides a power transaction center risk assessment system and method based on multi-risk superposition. The system comprises an input terminal, a processing terminal and a display device. The input terminal obtains risk basic data and sends the risk basic data to the processing terminal; the processing terminal performs abnormal and missing value filling on the risk basic data by applying an adversarial network method to obtain first risk data; determining an association relationship between influence factors and risks based on the first risk data through a grey relational analysis method, and screening out first influence factors of multiple target risks; analyzing a probability relationship between the first influence factors of the target risk in combination with a plurality of preset risk models; performing multi-objective optimization on target risks based on a genetic algorithm to obtain a plurality of weight combinations of the target risks; performing risk grade division based on an optimal weight combination in the weight combinations, and outputting risk grade data; and sending the risk level data to a display device. According to the invention, accurate prediction and correlation analysis of risks can be realized.

Description

Technical Field

[0001] The present application relates to the field of electric power technology, and in particular to a risk assessment system and method for an electric power trading center based on the superposition of multiple risks. Background Art

[0002] With the deepening of the market-oriented reform of the power industry, the power trading center assumes the core functions in power dispatching, market bidding and resource allocation. However, the operation of the trading center faces multiple risks, including market risk, credit risk and operational risk, which are highly complex and overlap with each other. If the risks are not identified and handled in time, it may lead to abnormal fluctuations in market prices, increase in credit default events, and even affect market stability and economic operation. Therefore, the development of scientific and efficient multi-risk superposition assessment technology is crucial for the operation and management of the power trading center.

[0003] Most existing risk management methods analyze and predict a single type of risk, and these methods are limited in their effectiveness when dealing with the interactions between different risks. At the same time, in the process of data collection and use, problems such as missing data and anomalies are often encountered, which affect the reliability of the analysis model and the accuracy of the results. Therefore, there is an urgent need for a technical framework that can efficiently process multidimensional data, achieve accurate risk prediction and correlation analysis, and help power trading centers conduct scientific risk assessment and decision-making. Summary of the invention

[0004] The present application provides a risk assessment system and method for a power trading center based on the superposition of multiple risks, so as to achieve accurate prediction and correlation analysis of risks and improve user experience.

[0005] In a first aspect, the present application provides a risk assessment system for a power trading center based on multi-risk superposition, comprising: an input terminal, a processing terminal and a display device; the input terminal and the display device are respectively connected to a communication interface of the processing terminal;

[0006] The input terminal is configured to: obtain risk basic data and send it to the processing terminal;

[0007] The processing terminal is configured to:

[0008] Receiving the risk basic data, and using an adversarial network method to fill abnormal values ​​and missing values ​​in the risk basic data to obtain first risk data;

[0009] Determine the correlation between the influencing factors and the risks based on the first risk data by using the grey correlation analysis method, and screen out the first influencing factors of the multiple target risks;

[0010] Analyzing the probability relationship between the first influencing factors of the target risk in combination with a plurality of preset risk models, and determining the weight of the target risk based on the probability relationship;

[0011] Based on the genetic algorithm, the multi-objective optimization of the target risk is performed to obtain a weighted combination of multiple target risks;

[0012] Performing risk level classification based on the optimal weight combination in the weight combinations, and outputting risk level data;

[0013] sending the risk level data to the display device;

[0014] The display device is configured to generate a target image according to the risk level data, and display the target image on a display of the display device; wherein the target image is used to characterize the distribution of the risk level of each target risk.

[0015] In some possible implementations, the processing terminal includes a storage unit, a computing unit, and a control unit, wherein the storage unit and the computing unit are electrically connected to the control unit respectively; the storage unit is used to store the calculated intermediate values ​​of the evaluation system; the computing unit is used to perform solution calculations; and the control unit is used to send control instructions.

[0016] In some possible implementations, adversarial network methods are used to fill abnormal values ​​and missing values ​​in the risk basic data, including:

[0017] Receiving the risk basic data; the risk basic data includes market risk data, credit risk data and operational risk data;

[0018] Performing outlier detection on the risk basic data based on a preset neural network model, and replacing or correcting the outliers;

[0019] The locations of missing values ​​in the risk basic data are marked based on a preset neural network model, and data is filled in at the locations.

[0020] In some possible implementations, the grey correlation analysis method is used to determine the correlation between the influencing factors and the risks based on the first risk data, and the first influencing factors of the plurality of target risks are screened out, including:

[0021] performing standardization processing on the first risk data;

[0022] Taking the target risk as the reference sequence and the data of each influencing factor as the comparison sequence, the difference between each comparison sequence and the reference sequence is calculated;

[0023] Based on the maximum and minimum values ​​of all differences, the grey relational coefficient is calculated, and the resolution coefficient is adjusted to control the sensitivity;

[0024] By averaging the correlation coefficients of each factor, the grey correlation degree is obtained;

[0025] Sort each factor from high to low according to the degree of correlation, and select the first influencing factor with the highest correlation with each type of risk.

[0026] In some possible implementations, analyzing the probability relationship between the first influencing factors of the target risk in combination with multiple preset risk models includes:

[0027] Market risk uses LSTM neural network to predict the probability of price fluctuations and capture time series characteristics;

[0028] Credit risk is optimized through support vector machine combined with logistic regression to calculate the probability of default;

[0029] Operational risk is modeled with the help of Bayesian networks, and the conditional probability relationship between multiple factors is analyzed.

[0030] Among some possible implementations, market risk uses LSTM neural networks to predict the probability of price fluctuations, including:

[0031] Build an LSTM model based on the characteristics of the data, define the input layer, LSTM hidden layer and output layer, and set the key parameters of the model;

[0032] The data is divided into training set and validation set according to the time window, and the model is trained through the back propagation algorithm, and the hyperparameters are adjusted using the validation set;

[0033] After the model training is completed, the test data is used for prediction, and the mean square error and mean absolute error are used to evaluate the prediction results.

[0034] In some possible implementations, credit risk is optimized through support vector machines combined with logistic regression to calculate the probability of default including:

[0035] Collect credit data of transaction partners, including financial status and historical default records;

[0036] Standardizing the credit data;

[0037] Select an appropriate kernel function and penalty parameter C, and build an SVM model;

[0038] Train the SVM model, use the collected data to calculate the support vector, and determine the location of the optimal hyperplane;

[0039] Combined with the parameters of the logistic regression optimization model, the test data is input into the model for prediction, and the performance of the model is evaluated through the confusion matrix and AUC indicators, and the default probability of each sample is output.

[0040] In some possible implementations, operational risk is modeled with the help of Bayesian networks, and the conditional probability relationship between multiple factors is analyzed, including:

[0041] Based on the operation process of the trading center, define the nodes and directed edges of the Bayesian network, where each node represents a variable that affects the operational risk;

[0042] Collect historical data, estimate the parameters of the probability distribution of each node, and use maximum likelihood estimation or expectation maximization algorithm to optimize the conditional probability;

[0043] Use Bayesian inference algorithms to calculate the probability of operational risk in specific scenarios and analyze the contribution of different variables to risk;

[0044] By continuously updating data and models, and conducting targeted risk control and prevention based on model results.

[0045] In some possible implementations, performing risk level classification based on the optimal weight combination in the weight combinations, and outputting risk level data includes:

[0046] Genetic algorithms are used to generate a new generation of individuals through selection, crossover and mutation operations, and the performance of each group of weights is evaluated based on the fitness function, with the goal of minimizing the risk error;

[0047] After multiple generations of evolution, the algorithm converges to the optimal weight combination, thereby deriving reasonable weights for each risk factor and outputting risk level data.

[0048] In a second aspect, the present application provides a risk assessment method for a power trading center based on multiple risk superposition, which is configured in the risk assessment system for a power trading center based on multiple risk superposition described in the first aspect, and the method includes:

[0049] The input terminal obtains the risk basic data and sends it to the processing terminal;

[0050] The processing terminal receives the risk basic data;

[0051] Using an adversarial network method, filling abnormal values ​​and missing values ​​in the risk basic data to obtain first risk data;

[0052] Determine the correlation between the influencing factors and the risks based on the first risk data by using the grey correlation analysis method, and screen out the first influencing factors of the multiple target risks;

[0053] Analyze the probability relationship between the first influencing factors of the target risk in combination with multiple preset risk models;

[0054] Based on the genetic algorithm, the multi-objective optimization of the target risk is performed to obtain a weighted combination of multiple target risks;

[0055] Performing risk level classification based on the optimal weight combination in the weight combinations, and outputting risk level data;

[0056] sending the risk level data to a display device;

[0057] The display device generates a target image according to the risk level data, and displays the target image on a display of the display device; wherein the target image is used to characterize the distribution of the risk level of each target risk.

[0058] It can be seen from the above technical scheme that the present application provides a risk assessment system and method for a power trading center based on multiple risk superposition, the system comprising an input terminal, a processing terminal and a display device; the input terminal obtains risk basic data and sends it to the processing terminal; the processing terminal uses an adversarial network method to fill in the abnormalities and missing values ​​of the risk basic data to obtain the first risk data; through the gray correlation analysis method, the correlation between the influencing factors and the risks is determined based on the first risk data, and the first influencing factors of multiple target risks are screened out; the probability relationship between the first influencing factors of the target risks is analyzed in combination with a variety of preset risk models; the target risk is optimized based on the genetic algorithm to obtain a weight combination of multiple target risks; the risk level is divided based on the optimal weight combination in the weight combination, and the risk level data is output; the risk level data is sent to the display device. The present invention can not only realize the accurate prediction of multiple risks, but also divide the risks of the trading center into low, medium and high levels through the calculated comprehensive risk scores, providing a scientific basis for the risk warning and management decision of the trading center. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0060] Figure 1 A schematic diagram of the structure of a risk assessment system for a power trading center based on multi-risk superposition provided in an embodiment of the present application;

[0061] Figure 2 This is a flow chart of a risk assessment method for a power trading center based on multiple risk superposition according to an embodiment of the present application;

[0062] Figure 3 It is the optimal hyperplane graph of SVM in the embodiment of the present application;

[0063] Figure 4 Schematic diagram of the SVM classification function of an embodiment of the present application. DETAILED DESCRIPTION

[0064] The following embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following embodiments do not represent all implementations consistent with the present application. They are only examples of systems and methods consistent with some aspects of the present application as detailed in the claims.

[0065] With the deepening of the market-oriented reform of the power industry, the power trading center assumes the core functions in power dispatching, market bidding and resource allocation. However, the operation of the trading center faces multiple risks, including market risk, credit risk and operational risk, which are highly complex and overlap with each other. If the risks are not identified and handled in time, it may lead to abnormal fluctuations in market prices, increase in credit default events, and even affect market stability and economic operation. Therefore, the development of scientific and efficient multi-risk superposition assessment technology is crucial for the operation and management of the power trading center.

[0066] Most existing risk management methods analyze and predict a single type of risk, and these methods are limited in their effectiveness when dealing with the interactions between different risks. At the same time, in the process of data collection and use, problems such as missing data and anomalies are often encountered, which affect the reliability of the analysis model and the accuracy of the results. Therefore, there is an urgent need for a technical framework that can efficiently process multidimensional data, achieve accurate risk prediction and correlation analysis, and help power trading centers conduct scientific risk assessment and decision-making.

[0067] In recent years, data-driven intelligent models have gradually been applied to the field of risk prediction, including neural networks, Bayesian networks, support vector machines and other models, which can solve the problems of nonlinear risk analysis and multi-factor modeling to a certain extent. In addition, the adversarial generative network has shown excellent performance in data filling and outlier repair, providing technical guarantee for data integrity. At the same time, through genetic algorithms to achieve weight optimization and multi-risk superposition evaluation, a scientific division of risk levels can be achieved. The integration of these methods will build a complete and multi-dimensional risk assessment system for the power trading center to support its precise risk management and decision support. Based on this, some embodiments of the present application provide a risk assessment system and method for a power trading center based on multi-risk superposition. By continuously optimizing the iterative formula, the change in data between two iterations is calculated during the iteration process and the iterative formula is dynamically changed to reduce the number of iterations, so as to improve the utilization rate of hardware units and data parallelism, avoid overly complex calculations, and improve user experience.

[0068] In some embodiments, Figure 1 , the present application provides a risk assessment system for a power trading center based on multi-risk superposition, comprising: an input terminal, a processing terminal and a display device; the input terminal and the display device are respectively connected to a communication interface of the processing terminal;

[0069] The input terminal is configured to: obtain risk basic data and send it to the processing terminal;

[0070] The processing terminal is configured to:

[0071] Receiving the risk basic data, and using an adversarial network method to fill abnormal values ​​and missing values ​​in the risk basic data to obtain first risk data;

[0072] Among them, the responsibilities of the terminal and the power trading center are handled, the main risks of the power trading center are sorted out, and risk factors are further identified, including market risk, credit risk and operational risk.

[0073] Determine the correlation between the influencing factors and the risks based on the first risk data by using the grey correlation analysis method, and screen out the first influencing factors of the multiple target risks;

[0074] Analyzing the probability relationship between the first influencing factors of the target risk in combination with a plurality of preset risk models, and determining the weight of the target risk based on the probability relationship;

[0075] Based on the genetic algorithm, the multi-objective optimization of the target risk is performed to obtain a weighted combination of multiple target risks;

[0076] Performing risk level classification based on the optimal weight combination in the weight combinations, and outputting risk level data;

[0077] sending the risk level data to the display device;

[0078] The display device is configured to generate a target image according to the risk level data, and display the target image on a display of the display device; wherein the target image is used to characterize the distribution of the risk level of each target risk.

[0079] In some embodiments, the processing terminal includes a storage unit, a computing unit and a control unit, wherein the storage unit and the computing unit are electrically connected to the control unit respectively; the storage unit is used to store the calculated intermediate values ​​of the evaluation system; the computing unit is used to perform solution calculations; and the control unit is used to send control instructions.

[0080] In some embodiments, an adversarial network method is used to fill abnormal values ​​and missing values ​​in the risk basic data, including:

[0081] Receiving the risk basic data; the risk basic data includes market risk data, credit risk data and operational risk data;

[0082] Performing outlier detection on the risk basic data based on a preset neural network model, and replacing or correcting the outlier;

[0083] The locations of missing values ​​in the risk basic data are marked based on a preset neural network model, and data is filled in at the locations.

[0084] The preset neural network GAN consists of two neural networks: the generator and the discriminator. The generator is responsible for generating realistic data, while the discriminator is responsible for distinguishing between real data and generated data. The two compete with each other, and finally the generator can generate data close to the real data. For outlier detection and missing value filling, the core idea of ​​GAN is to make abnormal data or missing data consistent with normal data by generating data.

[0085] Step 1: Outlier Detection

[0086] Build the GAN network:

[0087] The generator takes input data and generates sample data (simulating normal data).

[0088] The discriminator distinguishes between normal and outliers in the input data.

[0089] Training the adversarial network:

[0090] The input contains data containing noise and outliers, the generator tries to generate realistic data, and the discriminator marks the outliers.

[0091] Output abnormal points:

[0092] After multiple rounds of iterations, the discriminator will accurately identify the outliers in the original data. The marked data will be replaced or corrected.

[0093] Step 2: Fill in missing values

[0094] Data preprocessing:

[0095] Mark the locations of missing values ​​in the dataset and construct an incomplete dataset.

[0096] Build the generator and discriminator models:

[0097] The generator generates padded data, and the discriminator determines whether the padded data conforms to the overall data distribution.

[0098] GAN model training:

[0099] The generator gradually optimizes the generated data until the discriminator cannot distinguish the padded data from the real data.

[0100] Fill missing values:

[0101] Finally, the trained generator is used to fill in the missing values ​​to make the data complete and consistent with statistical characteristics.

[0102] In some embodiments, the grey correlation analysis method is used to determine the correlation between the influencing factors and the risks based on the first risk data, and the first influencing factors of the plurality of target risks are screened out, including:

[0103] performing standardization processing on the first risk data;

[0104] Taking the target risk as the reference sequence and the data of each influencing factor as the comparison sequence, the difference between each comparison sequence and the reference sequence is calculated;

[0105] Based on the maximum and minimum values ​​of all differences, the grey relational coefficient is calculated, and the resolution coefficient is adjusted to control the sensitivity;

[0106] By averaging the correlation coefficients of each factor, the grey correlation degree is obtained;

[0107] Sort each factor from high to low according to the degree of correlation, and select the first influencing factor with the highest correlation with each type of risk.

[0108] Grey correlation analysis is an analytical method used to evaluate the similarity and correlation between multiple variables. It is suitable for small sample scenarios with large uncertainty and incomplete data. Its core principle is to determine the correlation between variables and the target variable by comparing the changing trends between them. The more similar the trends are, the higher the correlation is. In the risk assessment of the power trading center, this method can quantify the relationship between different influencing factors and risk categories, and identify key factors that are highly correlated with market, credit and operational risks.

[0109] The specific steps include: first, standardize the original data to ensure the comparability of data of different dimensions. Then, take the target risk as the reference sequence and the data of each influencing factor as the comparison sequence, and calculate the difference between each comparison sequence and the reference sequence. Next, calculate the grey correlation coefficient based on the maximum and minimum values ​​of all differences, and adjust the resolution coefficient to control the sensitivity. Subsequently, the grey correlation degree is obtained by averaging the correlation coefficient of each factor. Finally, sort the factors from high to low according to the correlation degree, and screen out the key influencing factors with the highest correlation degree with each risk, providing a basis for subsequent risk assessment and optimization.

[0110] The market risk finally selected four quantitative factors: market price volatility, trading volume volatility, supply and demand volatility and energy price index.

[0111] Four quantitative indicators were finally selected for credit risk: debt-to-asset ratio, historical default rate, credit rating score and accounts receivable turnover rate.

[0112] Operational risk ultimately determines system stability score, information security incident frequency, employee operational error rate and emergency response time as quantitative indicators.

[0113] In some embodiments, analyzing the probability relationship between the first influencing factors of the target risk in combination with multiple preset risk models includes:

[0114] Market risk uses LSTM neural network to predict the probability of price fluctuations and capture time series characteristics;

[0115] Credit risk is optimized through support vector machine combined with logistic regression to calculate the probability of default;

[0116] Operational risk is modeled with the help of Bayesian networks, and the conditional probability relationship between multiple factors is analyzed.

[0117] In some embodiments, the market risk uses the LSTM neural network to predict the probability of price fluctuations including:

[0118] Build an LSTM model based on the characteristics of the data, define the input layer, LSTM hidden layer and output layer, and set the key parameters of the model;

[0119] The data is divided into training set and validation set according to the time window, and the model is trained through the back propagation algorithm, and the hyperparameters are adjusted using the validation set;

[0120] After the model training is completed, the test data is used for prediction, and the mean square error and mean absolute error are used to evaluate the prediction results.

[0121] In some embodiments, the credit risk is optimized by a support vector machine combined with logistic regression, and the calculation of the probability of default includes:

[0122] Collect credit data of transaction partners, including financial status and historical default records;

[0123] Standardizing the credit data;

[0124] Select an appropriate kernel function and penalty parameter C, and build an SVM model;

[0125] Train the SVM model, use the collected data to calculate the support vector, and determine the location of the optimal hyperplane;

[0126] Combined with the parameters of the logistic regression optimization model, the test data is input into the model for prediction, and the performance of the model is evaluated through the confusion matrix and AUC indicators, and the default probability of each sample is output.

[0127] In some embodiments, operational risk is modeled using a Bayesian network, and analyzing the conditional probability relationship between multiple factors includes:

[0128] Based on the operation process of the trading center, define the nodes and directed edges of the Bayesian network, where each node represents a variable that affects the operational risk;

[0129] Collect historical data, estimate the parameters of the probability distribution of each node, and use maximum likelihood estimation or expectation maximization algorithm to optimize the conditional probability;

[0130] Use Bayesian inference algorithms to calculate the probability of operational risk in specific scenarios and analyze the contribution of different variables to risk;

[0131] By continuously updating data and models, and conducting targeted risk control and prevention based on model results.

[0132] (1) Market risk prediction model based on LSTM

[0133] LSTM (Long Short-Term Memory) is a recurrent neural network (RNN) that controls the storage, forgetting and output of information by introducing input gates, forgetting gates and output gates to solve the gradient vanishing problem of traditional RNN. LSTM is good at capturing long-term dependencies in time series and is suitable for predicting fluctuations in electricity prices in market risks. It can discover the inherent laws of price changes by learning from historical data, thereby accurately predicting future market price trends and capturing risk signals in market fluctuations.

[0134] First, build an LSTM model based on the characteristics of the data, define the input layer, LSTM hidden layer, and output layer, and set the key parameters of the model, such as learning rate, time step, batch size, etc. Next, divide the data into training set and validation set according to the time window, and train the model through the back propagation algorithm, and use the validation set to adjust the hyperparameters to avoid overfitting. After the model training is completed, use the test data for prediction, and use the mean square error (MSE), mean absolute error (MAE) and other indicators to evaluate the prediction results to ensure that the model has high accuracy and generalization ability.

[0135] (2) Credit risk prediction model based on support vector machine

[0136] Support vector machine (SVM) is a classification algorithm based on statistical learning theory, which is suitable for classification tasks that process complex and high-dimensional data. In credit risk prediction, SVM divides the credit status of the transaction object into two categories: "default" and "non-default" by finding the optimal hyperplane. In order to improve the classification accuracy, SVM also combines logistic regression to optimize the weights, improve the processing ability of small samples and unbalanced data, and enhance the robustness and prediction accuracy of the model. First, collect the credit data of the transaction object, including variables such as financial status and historical default records, and standardize the data to ensure that data of different dimensions are comparable. Next, select an appropriate kernel function (such as linear kernel or RBF kernel) and penalty parameter C, and build an SVM model. In the model training stage, the support vector is calculated using the collected data to determine the location of the optimal hyperplane. Subsequently, the parameters of the model are optimized in combination with logistic regression to improve the flexibility of the classification boundary and the stability of the model. After the model is trained, the test data is input into the model for prediction, and the performance of the model is evaluated by confusion matrix, AUC and other evaluation indicators, and the default probability of each sample is output to provide support for credit risk management.

[0137] (3) Bayesian-based operational risk prediction model

[0138] Bayesian network is a tool based on probabilistic graph model, which is used to represent and calculate the conditional dependencies between random variables. It constructs a graph structure through nodes (variables) and directed edges (dependencies), and uses Bayesian theorem to quantify the associations between different operational risk factors. In the operational risk management of power trading centers, Bayesian network can model the key links and their dependencies in the operation process, infer the risk probability of each link based on historical data, and provide support for risk identification and prevention.

[0139] First, based on the operational process of the trading center, define the nodes and directed edges of the Bayesian network, where each node represents a variable that affects operational risk, such as human error, system failure, process violation, etc. Then, collect historical data, estimate the parameters of the probability distribution of each node, and use the maximum likelihood estimation or expectation maximization (EM) algorithm to optimize the conditional probability. After the network is built, use the Bayesian inference algorithm to calculate the probability of operational risk in a specific scenario and analyze the contribution of different variables to the risk. By continuously updating data and models, the trading center can identify and warn of potential operational risks, and carry out targeted risk control and prevention based on the model results.

[0140] In some embodiments, performing risk level classification based on the optimal weight combination in the weight combination, and outputting risk level data includes:

[0141] Genetic algorithms are used to generate a new generation of individuals through selection, crossover and mutation operations, and the performance of each group of weights is evaluated based on the fitness function, with the goal of minimizing the risk error;

[0142] After multiple generations of evolution, the algorithm converges to the optimal weight combination, thereby deriving reasonable weights for each risk factor and outputting risk level data.

[0143] Specifically include:

[0144] (1) Initialize the population

[0145] In the first step, several individuals (i.e., multiple risk weight combinations) are generated, which form the initial population. Each individual represents a different weight distribution method for market risk, credit risk, and operational risk. In order to ensure the diversity of solutions, each individual in the population has a random weight value.

[0146] (2) Fitness evaluation

[0147] Individuals in each generation need to be evaluated for fitness. The level of fitness represents the performance of the individual. In this application, the fitness is determined by the error performance of each risk prediction model, ensuring that the optimization goal of the risk weight combination is to minimize the combined error of market, credit and operational risks.

[0148] (3) Select operation

[0149] In this stage, individuals with higher fitness are selected to enter the next generation through methods such as "roulette" or "tournament selection". Individuals with higher fitness have a greater probability of being selected for reproduction. This step ensures that the genes of high-quality individuals can be passed on to future generations.

[0150] (4) Crossover operation

[0151] The selected individuals are paired with each other and crossover is performed with a certain probability. The crossover process is similar to gene recombination, exchanging some gene fragments of the parent individuals to generate new offspring individuals. For example, a market risk weight can be exchanged with the credit risk weight of another individual. Through this recombination, the new generation of individuals can inherit the good characteristics of the parents and have a new gene combination.

[0152] (5) Mutation Operation

[0153] In order to maintain the diversity of the population and prevent the algorithm from falling into the local optimal solution, there is a certain probability that individuals in each generation will mutate. The mutation operation generates new solution space exploration opportunities by randomly changing the risk weight of an individual. For example, the weight value of a market risk can be slightly adjusted to explore a better weight distribution method.

[0154] (6) Generate a new generation of individuals and evaluate their fitness

[0155] After crossover and mutation operations, a new generation of population is generated. The individuals of the new generation of population are evaluated for fitness again to ensure that the optimization direction remains consistent. Through the improvement of fitness, each generation will have a better risk weight combination than the previous generation.

[0156] (7) Multi-generational evolution and convergence

[0157] The above selection, crossover and mutation operations are repeated many times until the stopping condition is met. The stopping condition is usually when a preset number of generations (such as 100 generations) is reached or the fitness does not improve within a few generations. When the algorithm converges, the best individual in the current population is the best risk weight combination.

[0158] (8) Comprehensive risk level classification

[0159] Based on the optimized risk weight combination, the probability values ​​of market risk, credit risk and operational risk are weighted and combined to obtain a comprehensive risk score. According to this score, the risks of the power trading center are divided into different levels, 0-0.3 for low risk, 0.3-0.7 for medium risk, and 0.7-1 for high risk, to achieve scientific assessment and effective management of trading risks, and provide reference for the operational decision-making of the trading center.

[0160] like Figure 2As shown, in some embodiments, the present application provides a risk assessment method for a power trading center based on multiple risk superposition, which is configured in the risk assessment system for a power trading center based on multiple risk superposition of the above embodiment, and the method includes:

[0161] The input terminal obtains the risk basic data and sends it to the processing terminal;

[0162] The processing terminal receives the risk basic data;

[0163] Using an adversarial network method, filling abnormal values ​​and missing values ​​in the risk basic data to obtain first risk data;

[0164] Determine the correlation between the influencing factors and the risks based on the first risk data by using the grey correlation analysis method, and screen out the first influencing factors of the multiple target risks;

[0165] Analyze the probability relationship between the first influencing factors of the target risk in combination with multiple preset risk models;

[0166] Based on the genetic algorithm, the multi-objective optimization of the target risk is performed to obtain a weighted combination of multiple target risks;

[0167] Performing risk level classification based on the optimal weight combination in the weight combinations, and outputting risk level data;

[0168] sending the risk level data to a display device;

[0169] The display device generates a target image according to the risk level data, and displays the target image on a display of the display device; wherein the target image is used to characterize the distribution of the risk level of each target risk.

[0170] The present invention uses generative adversarial network (GAN) technology to process data, which can accurately identify and repair outliers and missing values ​​in the data, ensure the integrity and consistency of basic data, and improve the accuracy of subsequent risk prediction models.

[0171] Based on the multi-model combination of LSTM neural network, support vector machine and Bayesian network, accurate prediction of market risk, credit risk and operational risk is achieved. By capturing time series characteristics, logistic regression optimization and conditional probability modeling, the reliability of various risk predictions is improved from multiple angles.

[0172] Genetic algorithms are used to optimize the weights of various risk factors, minimize risk errors, and achieve a comprehensive assessment of the risk superposition effect. After multiple generations of iterative optimization, the algorithm eventually converges to the optimal weight combination, ensuring the scientificity and rationality of the risk level division.

[0173] The risk assessment technology of the present invention can not only realize the accurate prediction of various risks, but also divide the risks of the trading center into low, medium and high levels through the calculated comprehensive risk scores, providing a scientific basis for the risk warning and management decision-making of the trading center.

[0174] Example

[0175] Step S1: Based on the responsibilities of the power trading center, sort out the main risks of the power trading center and further identify risk factors, including market risk, credit risk and operational risk.

[0176] Market risk mainly reflects the fluctuation of market price and trading volume, which affects the income stability of the trading center. It is determined by screening 10 factors, including market price volatility, trading volume volatility, supply and demand volatility, energy price index, interest rate changes, exchange rate changes, industry competition index, policy change frequency, market capacity and inflation rate.

[0177] Credit risk involves the default possibility and financial health of the counterparty. We analyze the credit risk from the debt-to-asset ratio, credit rating score, historical default rate, cash flow coverage, contract fulfillment rate, accounts receivable turnover, debt size, transaction history score, financing cost and net profit margin.

[0178] Operational risks mainly come from system failures, human errors or insufficient process management. We evaluate the system stability score, information security incident frequency, employee error rate, emergency response time, process complexity index, system upgrade frequency, employee turnover rate, training coverage, technical support time and management compliance score.

[0179] In step S2, basic data is collected, and an adversarial network method is used to fill in outliers and missing values ​​to achieve data processing.

[0180] Collect data related to market risk, credit risk, operational risk, etc. Specific variables include quantitative indicators such as price volatility, transaction amount, user credit score, number of operations, etc. These data may come from historical transaction records, market conditions, user credit evaluation systems, etc.

[0181] The data may contain noise, outliers, or missing values, and need to be preprocessed to ensure the accuracy of the analysis.

[0182] GAN consists of two neural networks: the generator and the discriminator. The generator is responsible for generating realistic data, while the discriminator is responsible for distinguishing between real data and generated data. The two compete with each other, and the generator can eventually generate data close to the real data. For outlier detection and missing value filling, the core idea of ​​GAN is to make abnormal data or missing data consistent with normal data by generating data.

[0183] Specific steps:

[0184] Data preparation: Collect and organize historical data and relevant influencing factor data to ensure the accuracy and completeness of the data.

[0185] Constructing adversarial networks: Designing and constructing generative adversarial networks that are suitable for data characteristics, including the structure and parameter settings of the generator and discriminator.

[0186] Training the adversarial network: Use normal data to train the adversarial network, so that the generator can learn the distribution characteristics of the data, and the discriminator can continuously improve its ability to distinguish between real data and generated data.

[0187] Outlier detection: Detect outliers in historical data and related influencing factor data through statistical methods.

[0188] Outlier interpolation: The detected outliers are input into the trained generator, which will generate new data similar to the normal data distribution to replace the outliers.

[0189] Specifically, performing outlier interpolation processing on historical cost input data and related influencing factor data through an adversarial network further includes: generating a reconstructed sample for adversarial use through a generator; distinguishing the distance between the adversarial sample and the original sample through a discriminator, wherein the original sample is any sample data of the historical cost data and the related influencing factor data; and reducing the distance between the reconstructed sample G(z) and the original sample x using a reconstruction loss function, wherein the reconstruction loss function is represented by the following formula:

[0190] L r (z) = ||x·mG(z)·m||2;

[0191] The discriminant loss function is used to maximize the output probability of the discriminator to ensure that the reconstructed sample G(z) conforms to the distribution of the original sample. The discriminant loss function is expressed by the following formula:

[0192] L d (z) = -D(G(z));

[0193] A filling loss function is generated according to the reconstruction loss function and the discriminant loss function, and the filling loss function is minimized based on the reconstructed samples and the original samples to generate missing values ​​for interpolation. The filling loss function and the missing value are respectively expressed by the following formulas:

[0194] L inputation (z) = L r (z)+λ·L d (z);

[0195] x imputed =x·m+(1-m)·G(z * );

[0196] Where x represents the original sample; m is a mask matrix used to indicate the missing parts in the historical cost input data and related influencing factor data; G(z) represents the reconstructed sample, Lr(z) represents the distance between the generated sample and the original sample; Ld(z) represents the discriminant loss function, which is used to evaluate the output probability of the discriminator; D(G(z)) is the output of the discriminator, which represents the probability that the reconstructed sample G(z) is discriminated as the original sample; Limputation(z) is a comprehensive loss function composed of the reconstruction loss function and the discriminant loss function; λ is a hyperparameter used to control the ratio between the reconstruction loss and the discriminant loss; ximputed represents the padded data sample; x·m represents the non-missing part in the original sample; (1-m)·G(z) represents the corresponding missing part in the reconstructed sample; G(z * ) represents filling the missing values ​​of the original sample.

[0197] In step S3, a metric analysis is further performed on the influencing factors, and the correlation between the influencing factors and the risks is determined by the grey correlation analysis method, so as to screen the main influencing factors of different risks.

[0198] Grey correlation analysis is an analytical method used to evaluate the similarity and correlation between multiple variables. It is suitable for small sample scenarios with large uncertainty and incomplete data. Its core principle is to determine the correlation between variables and the target variable by comparing the changing trends between them. The more similar the trends are, the higher the correlation is. In the risk assessment of the power trading center, this method can quantify the relationship between different influencing factors and risk categories, and identify key factors that are highly correlated with market, credit and operational risks.

[0199] The specific steps are as follows: determine the analysis sequence, assume that the dependent variable data constitutes the reference sequence, and the individual variable data constitutes the comparison sequence Xi′ (i=1, 2, ...n), and the n+1 data sequences form the following matrix.

[0200]

[0201] Where Xi ′=(x i ′(1),x i ′(2),...,x i ′(N)) T ,i=0,1,2,...,n, N is the length of the variable sequence.

[0202] For each influencing factor Xi′ and effective asset size Y (reference sequence), the data is first dimensionally normalized (standardized or initialized). The method usually adopted is to normalize the range of each dimensional data so that the minimum value of each data sequence becomes 0 and the maximum value becomes 1.

[0203]

[0204]

[0205] Calculate the correlation coefficient r ij : The correlation coefficient r between the i-th influencing factor and risk Y ij , can be measured by calculating the similarity of the relative change rates of the two, the formula is as follows.

[0206]

[0207] Calculate the grey relational degree G i : Correlation G i is the correlation coefficient r ij The average value reflects the long-term correlation between the i-th influencing factor and the risk. The calculation formula is as follows:

[0208]

[0209] After calculating the grey correlation of all influencing factors, they are sorted according to the correlation. The factors with high correlation indicate a stronger correlation with the risk. Finally, the first four factors are selected as the input parameters of the prediction model.

[0210] The market risk finally selected four quantitative factors: market price volatility, trading volume volatility, supply and demand volatility and energy price index.

[0211] Four quantitative indicators were finally selected for credit risk: debt-to-asset ratio, historical default rate, credit rating score and accounts receivable turnover rate.

[0212] Operational risk ultimately determines system stability score, information security incident frequency, employee operational error rate and emergency response time as quantitative indicators.

[0213] In step S4, the market risk is combined with the risk type respectively, using the LSTM neural network to predict the probability of price fluctuations and capture the time series characteristics; the credit risk is optimized by the support vector machine combined with logistic regression to calculate the probability of default; the operational risk is modeled with the help of Bayesian network to analyze the conditional probability relationship between multiple factors.

[0214] (1) Market risk prediction model based on LSTM

[0215] LSTM neural network is designed to effectively transmit and express information in long time series without causing useful information from a long time ago to be forgotten.

[0216] The input of the forget gate is the hidden state h of the previous moment t-1 and the current input x t , output f t Used to control which information in memory cells should be forgotten at a certain moment, f is the bias vector of the forget gate. The relationship between input and output is as follows:

[0217] f t =σ(W f [h t-1 ,x t ]+b f );

[0218] In the LSTM neural network, the input of the input gate is the hidden state h of the previous moment. t-1 and the current input x t , output i t Used to determine the information in the current input that should be added to the memory cell, b i is the bias vector of the input gate. The relationship between its input and output is:

[0219] i t =σ(W i [h t-1 ,x t ]+b i );

[0220] In addition, the input of the memory cell is the hidden state h of the previous moment t-1 and the current input x t , the output is the current cell C t Used to update memory cells, b c is the bias vector of the output gate.

[0221]

[0222]

[0223] Where:

[0224] ——External output information of the memory unit module at time t.

[0225] When using the SAO algorithm to optimize LSTM parameters, the algorithm is first initialized to determine the number of clusters N of molecules and the dimension dim of the problem. Then, a batch of N rows and dim columns of molecules are randomly generated.

[0226]

[0227] Where:

[0228] Z——position information matrix of all molecules;

[0229] Z min , Z max ——the upper and lower limits of the molecular position;

[0230] Z rand ——Random numbers, used to randomly generate different molecular positions;

[0231] Z 1,1 ——The position of the first molecule in the first dimension.

[0232] The snow melting optimization algorithm uses a dual population mechanism to divide the molecular population P into two sub-populations P a and P b The number of molecules in the population is N a and N b , the expression is:

[0233]

[0234]

[0235] N b (t+1)=N b (t) -1;

[0236] Input layer: takes the price data of the previous n days as input.

[0237] Hidden layer: Several LSTM units to capture time-dependent characteristics.

[0238] Output layer: Output the probability of market risk occurring.

[0239] Use historical data for training.

[0240] The mean square error (MSE) is selected as the loss function to optimize the model parameters.

[0241] The model is tested using the validation dataset and the prediction error is calculated.

[0242] Output the probability of market risk occurring in the next period.

[0243] (2) Credit risk prediction model based on support vector machine

[0244] Support vector machine (SVM) is a classification algorithm based on statistical learning theory, which is suitable for classification tasks that process complex and high-dimensional data. In credit risk prediction, SVM divides the credit status of transaction objects into two categories: "default" and "non-default" by finding the optimal hyperplane. In order to improve classification accuracy, SVM also combines logistic regression to optimize weights, improve the ability to handle small samples and unbalanced data, and enhance the robustness and prediction accuracy of the model.

[0245] Given a sample set {(x i ,y i ),i=1,2,···,l}, where x i ∈R n ,y i ∈{+1,-1}. The principle of support vector machine for solving nonlinear function optimization is to solve the high-dimensional feature space by mapping the nonlinearity into the high-dimensional space. First, a decision function (classification hyperplane) should be constructed to classify the sample data. The classification hyperplane can maximize the blank area on both sides of the hyperplane while ensuring the classification accuracy, so as to achieve the optimal classification of linearly separable problems. Linear separability means that sample points belonging to different categories are separated by one or several straight lines.

[0246] If there is a classification hyperplane:

[0247] WX+b=0;

[0248] Satisfy the constraints:

[0249] y i (WX+b)-1≥0,i=1,2,···,l;

[0250] The training set is said to be linearly separable. Where W is the weight vector, X is the input vector, and b∈R is the bias.

[0251] Use p to represent the distance between the hyperplane and the nearest sample. The purpose of the support vector machine is to find a hyperplane with the largest separation margin, which is the optimal hyperplane.

[0252] like Figure 3 As shown, we need to determine w and b when p is maximum. The prediction function obtained is:

[0253] Y = f(x) = sgn(WX+b);

[0254] Its generalization ability is the best. Among them, sgn(·) is the sign function. That is, the classification error rate of the original sample should be minimized.

[0255] From analytic geometry, we know that the distance from any point in the sample space to the optimal hyperplane is

[0256]

[0257] Simplify:

[0258] g(x)=r||W0||=W0X+b0;

[0259] Normalize the function so that |g(x0)|=1. This is the support vector, which is closest to the classification decision surface and plays a leading role in SVM.

[0260] So the algebraic distance from the support vector to the optimal hyperplane is

[0261]

[0262] Therefore, it can be considered that maximizing the separation margin is equivalent to minimizing the norm of the weight vector ∥W∥. Therefore, the classification hyperplane that minimizes ∥W∥ is the optimal hyperplane.

[0263] The classification machine is an algorithm that transforms the optimization problem of solving the optimal classification surface by the maximum margin method into its dual problem, thereby solving the original classification problem by solving the relatively simple dual problem.

[0264] Using the Lagrange optimization method and Wofle duality theory, the above problem can be transformed into its dual problem, that is, maximizing the universal function:

[0265]

[0266]

[0267] Among them, α i is the Lagrange multiplier corresponding to the sample.

[0268] Solving the above problem, the optimal classification function is:

[0269]

[0270] Among them, nsv is the number of support vectors, and b is the classification threshold, which can be obtained by taking the median of any pair of support vectors in the two categories.

[0271] By introducing the relaxation term ξ iIt can realize the generalized classification surface and solve the problem that the training samples are linearly inseparable, that is, it compromises the minimum misclassified samples and the maximum classification interval. For nonlinear problems, it can be transformed into a linear problem in a high-dimensional space through nonlinear transformation, and then the optimal classification surface is sought in this high-dimensional space. Only the inner product operation between samples (x i ·x j ) is involved, so only the inner product operation needs to be performed in the high-dimensional space, and the inner product operation can be realized by the function in the original space. According to the Hilbert-Schmidt principle, as long as the kernel function K(x i ·x j ) satisfies Mercer's condition, which corresponds to the inner product in a certain exchange space.

[0272] Therefore, the kernel function K(x i ·x j ) instead of the inner product in the formula, we can realize linear classification after some nonlinear transformation, and the optimal classification function becomes:

[0273]

[0274] In short, SVM maps the input vector to a high-dimensional feature space through some pre-selected nonlinear mapping, and constructs the optimal classification hyperplane in this feature space. In form, the SVM classification function is similar to a neural network, and the output is a linear combination of intermediate nodes, and each intermediate node corresponds to a support vector, such as Figure 4 The number of intermediate nodes of a neural network must be selected based on experience or comparative experiments, and different logics will have a great impact on network performance; while the number of intermediate nodes of an SVM is automatically determined by calculation.

[0275] Vapnik generalized the results obtained in SVM classification theory by introducing the ε-insensitive loss function, making it applicable to function fitting. With this generalization, the SVM method becomes a general method for expressing functions in high-dimensional space. It can be used for various function estimation problems, including density estimation problems and solving linear operator equations. i and The following optimization problem is constructed:

[0276]

[0277]

[0278] Among them, the constant C is the penalty coefficient.

[0279] The optimization problem in the dual space is:

[0280]

[0281]

[0282] Among them, α i , is the Lagrange multiplier.

[0283] Solve this problem and get the optimal Lagrange multiplier α i , So we get the fitting function:

[0284] (3) Bayesian-based operational risk prediction model

[0285] Bayesian network is a tool based on probabilistic graph model, which is used to represent and calculate the conditional dependencies between random variables. It constructs a graph structure through nodes (variables) and directed edges (dependencies), and uses Bayesian theorem to quantify the associations between different operational risk factors. In the operational risk management of power trading centers, Bayesian network can model the key links and their dependencies in the operation process, infer the risk probability of each link based on historical data, and provide support for risk identification and prevention.

[0286] The Bayesian network consists of two parts: the Bayesian network structure and the conditional probability table. The Bayesian network structure clearly reflects the relationship between data variables, and the conditional probability table reflects the degree of influence between nodes in each relationship.

[0287] Given a set of random variables, γ = {X1, X2, ..., Xn}, where Xi is an m-dimensional vector. The Bayesian network describes the joint conditional probability distribution on γ. The Bayesian network is defined as follows:

[0288] B=<G,θ> ;

[0289] G is a directed acyclic graph whose vertices correspond to random variables X1, X2, ..., Xn in a finite set γ. Its arcs represent a functional dependency. If there is an arc from variable Y to X, then Y is the parent or direct predecessor of X, and X is the successor of Y. Once its two parent nodes are given, each variable in the graph is independent of the non-successors of the node in the graph. All parent variables of Xi in the graph G are represented by the set Pa(Xi).

[0290] Identify the operational risk-related factors of the power trading center, such as information system failure, operational process errors, equipment damage, human errors, etc.

[0291] Treat each risk factor as a node in the network and determine the dependencies between nodes based on domain knowledge or historical data. For example, information system failure may affect the normal operation of the operation process, thereby increasing operational risks.

[0292] Draw the above dependencies into a directed acyclic graph (DAG) to ensure that there is no cyclic structure.

[0293] θ represents a set of parameters used to quantize the network. For each value xi of Xi and Pa(Xi), there is a parameter as follows:

[0294] θ xi |P a (X i )=P(xi|P a (X i ));

[0295] This parameter specifies the conditional probability of the event xi occurring given Pa(Xi). In fact, a Bayesian network gives the joint conditional probability distribution on the variable set γ, that is,

[0296]

[0297] Calculate the probability of overall operational risk based on the updated network structure and conditional probability table.

[0298] Use Bayesian reasoning tools or algorithms to automatically perform inferences between multiple nodes and ultimately derive the predicted probability of operational risk occurring.

[0299] In step S5, a genetic algorithm is used to generate a new generation of individuals through selection, crossover and mutation operations, and the performance of each group of weights is evaluated based on the fitness function, with the goal of minimizing the risk error. After multiple generations of evolution, the algorithm converges to the optimal weight combination, thereby deriving the reasonable weight of each risk factor and achieving an accurate division of the comprehensive risk level. Finally, based on the optimized comprehensive risk score, the risks of the trading center are divided into different levels.

[0300] The genetic algorithm realizes the multi-objective optimization of market risk, credit risk and operational risk, accurately calculates the weight combination of each risk, and finally realizes the classification of comprehensive risk levels of power trading centers. This method has the characteristics of self-adaptation and self-optimization, effectively solves the limitations of traditional methods in complex multi-objective problems, and provides a scientific and efficient tool for risk management of power trading centers. Genetic algorithm is an optimization algorithm that simulates the biological evolution process. By simulating the mechanisms of natural selection, gene recombination and mutation, it gradually approaches the optimal solution in a complex solution space. It is suitable for solving multi-objective and multi-constrained nonlinear problems. In the risk assessment of power trading centers, genetic algorithms are used to optimize the weight combination of market risk, credit risk and operational risk, ensure the reasonable allocation of various risks, and realize the classification of comprehensive risk levels. The genetic algorithm gradually improves the fitness of the final individual through the population evolution of the initial solution, and reaches the optimal state after multiple generations of evolution.

[0301] (1) Input system data, parameter values ​​and variable values. The input microhabitat genetic algorithm parameters include the size of the individual population, the number of genes contained in the individual, the crossover rate, the mutation rate and the maximum genetic generation.

[0302] (2) Use binary to encode chromosomes.

[0303] (3) Randomly generate initial population household seals.

[0304] (4) Verify constraints.

[0305] (5) Calculate the fitness function value of each chromosome in the population. Use the fitness function to evaluate the quality of each weight combination. The goal of the fitness function is to minimize the risk error. The formula min R = R e +R r +R g , so that the difference between the comprehensive risk and the target value (such as historical data or risk threshold) is minimized.

[0306]

[0307] In the formula,

[0308] GF—fitness function;

[0309] F—corrected value of the objective function;

[0310] F max —For a given larger constant, the maximum value of the objective function F obtained so far can be selected or obtained based on experimental data.

[0311] (6) Arrange the chromosomes in descending order according to their fitness function values, and remember the first LK chromosomes (M<LK).

[0312] (7) Perform proportional selection operation on the population P(tv) to obtain P(tv). The fitness function value proportional method is used as the selection strategy to implement the selection operation.

[0313] (8) Perform a single-point crossover operation on P(tv) to obtain P(tv). The crossover operation can change the topological structure of the two networks, thereby obtaining a new structure. The basic steps to achieve crossover are:

[0314] 1) Select the parent chromosome: The process is: repeat the following process from ai=1 to LK: generate a random number r from [0,1] ai , if r ai ≤P C , then select chromosome ai as the parent for crossover. C is the crossover rate, usually between 0.6 and 0.9, reflecting the idea of ​​information exchange in nature.

[0315] 2) Select the crossover point: Use one-point crossover. The specific operation is to randomly select a crossover point in a chromosome, and the two chromosomes are partially exchanged before or after the point to generate a new chromosome.

[0316] (9) Perform two-point mutation operation on P″(tv) to obtain P″′(tv).

[0317] (10) Calculate the fitness of P″′(tv).

[0318] (11) Perform the niche elimination algorithm. Combine the LK chromosomes obtained in step (9) and the M chromosomes memorized in step (6) to obtain a new population containing (LK+M) chromosomes. For these (LK+M) chromosomes, two chromosomes are obtained according to the following calculation formula: X ai and X aj The Hamming distance between:

[0319]

[0320] When ||X ai -X aj ||<DL (L is the specified distance), compare chromosome X ai and X aj The fitness of the chromosomes is calculated by the penalty function Fmin(X ai ,X aj )=Penalty.

[0321] (12) Arrange the new fitness of the (N+M) chromosomes in descending order and remember the first M chromosomes.

[0322] (13) Implement the optimal preservation strategy.

[0323] (14) Determine whether the termination condition is met. If the termination condition is met, go to step (15), otherwise update the evolutionary generation counter Gen = Gen + 1, take the first N chromosomes in the arrangement of step (11) as the new next generation population and then go to step (7).

[0324] (15) Decode and output the optimization calculation results. When the maximum number of iterations is reached or the fitness value is no longer significantly improved, the algorithm terminates and outputs the optimal weight combination.

[0325] (16) Comprehensive risk level classification

[0326] Based on the optimized risk weight combination, the probability values ​​of market risk, credit risk and operational risk are weighted and combined to obtain a comprehensive risk score. According to this score, the risks of the power trading center are divided into different levels, 0-0.3 for low risk, 0.3-0.7 for medium risk, and 0.7-1 for high risk, to achieve scientific assessment and effective management of trading risks, and provide reference for the operational decision-making of the trading center.

[0327] It can be seen from the above embodiments that the present application provides a risk assessment system and method for a power trading center based on multiple risk superpositions, the system comprising an input terminal, a processing terminal and a display device; the input terminal obtains risk basic data and sends it to the processing terminal; the processing terminal uses an adversarial network method to fill in the abnormalities and missing values ​​of the risk basic data to obtain the first risk data; through the gray correlation analysis method, the correlation between the influencing factors and the risks is determined based on the first risk data, and the first influencing factors of multiple target risks are screened out; the probability relationship between the first influencing factors of the target risks is analyzed in combination with a variety of preset risk models; the target risk is optimized based on the genetic algorithm to obtain a weight combination of multiple target risks; the risk level is divided based on the optimal weight combination in the weight combination, and the risk level data is output; the risk level data is sent to the display device. The present invention can not only realize the accurate prediction of multiple risks, but also divide the risks of the trading center into low, medium and high levels through the calculated comprehensive risk scores, providing a scientific basis for the risk warning and management decision of the trading center.

[0328] Similar parts between the embodiments provided in this application can be referenced to each other. The specific implementation methods provided above are only a few examples under the overall concept of this application and do not constitute a limitation on the scope of protection of this application.

Claims

1. A risk assessment system for power trading centers based on multi-risk superposition, characterized in that: include: An input terminal, a processing terminal and a display device; the input terminal and the display device are respectively connected to the communication interface of the processing terminal; The input terminal is configured to: obtain risk basic data and send it to the processing terminal; The processing terminal is configured to: Receiving the risk basic data, and using an adversarial network method to fill abnormal values ​​and missing values ​​in the risk basic data to obtain first risk data; Determine the correlation between the influencing factors and the risks based on the first risk data by using the grey correlation analysis method, and screen out the first influencing factors of the multiple target risks; Analyzing the probability relationship between the first influencing factors of the target risk in combination with a plurality of preset risk models, and determining the weight of the target risk based on the probability relationship; Based on the genetic algorithm, the multi-objective optimization of the target risk is performed to obtain a weighted combination of multiple target risks; Performing risk level classification based on the optimal weight combination in the weight combinations, and outputting risk level data; sending the risk level data to the display device; The display device is configured to generate a target image according to the risk level data, and display the target image on a display of the display device; wherein the target image is used to characterize the distribution of the risk level of each target risk.

2. The risk assessment system for power trading center based on multi-risk superposition according to claim 1 is characterized in that: The processing terminal includes a storage unit, a calculation unit and a control unit, wherein the storage unit and the calculation unit are electrically connected to the control unit respectively; the storage unit is used to store the calculated intermediate values ​​of the evaluation system; The calculation unit is used to perform solution calculations; the control unit is used to send control instructions.

3. The risk assessment system for power trading center based on multi-risk superposition according to claim 1 is characterized in that: Using adversarial network methods, fill in abnormal values ​​and missing values ​​of the risk basic data, including: Receiving the risk basic data; the risk basic data includes market risk data, credit risk data and operational risk data; Performing outlier detection on the risk basic data based on a preset neural network model, and replacing or correcting the outlier; The locations of missing values ​​in the risk basic data are marked based on a preset neural network model, and data is filled in at the locations.

4. The risk assessment system for power trading center based on multi-risk superposition according to claim 1 is characterized in that: By using the grey correlation analysis method, the correlation between the influencing factors and the risks is determined based on the first risk data, and the first influencing factors of multiple target risks are screened out, including: performing standardization processing on the first risk data; Taking the target risk as the reference sequence and the data of each influencing factor as the comparison sequence, the difference between each comparison sequence and the reference sequence is calculated; Based on the maximum and minimum values ​​of all differences, the grey relational coefficient is calculated, and the resolution coefficient is adjusted to control the sensitivity; By averaging the correlation coefficients of each factor, the grey correlation degree is obtained; Sort each factor from high to low according to the degree of correlation, and select the first influencing factor with the highest correlation with each type of risk.

5. The risk assessment system for power trading center based on multi-risk superposition according to claim 4 is characterized in that: Analyzing the probability relationship between the first influencing factors of the target risk in combination with multiple preset risk models includes: Market risk uses LSTM neural network to predict the probability of price fluctuations and capture time series characteristics; Credit risk is optimized through support vector machine combined with logistic regression to calculate the probability of default; Operational risk is modeled with the help of Bayesian networks, and the conditional probability relationship between multiple factors is analyzed.

6. The risk assessment system for power trading center based on multi-risk superposition according to claim 5 is characterized in that: Market risks use LSTM neural networks to predict the probability of price fluctuations including: Build an LSTM model based on the characteristics of the data, define the input layer, LSTM hidden layer and output layer, and set the key parameters of the model; The data is divided into training set and validation set according to the time window, and the model is trained through the back propagation algorithm, and the hyperparameters are adjusted using the validation set; After the model training is completed, the test data is used for prediction, and the mean square error and mean absolute error are used to evaluate the prediction results.

7. The risk assessment system for power trading center based on multi-risk superposition according to claim 5 is characterized in that: Credit risk is optimized through support vector machine combined with logistic regression, and the default probability is calculated including: Collect credit data of transaction partners, including financial status and historical default records; Standardizing the credit data; Select an appropriate kernel function and penalty parameter C, and build an SVM model; Train the SVM model, use the collected data to calculate the support vector, and determine the location of the optimal hyperplane; Combined with the parameters of the logistic regression optimization model, the test data is input into the model for prediction, and the performance of the model is evaluated through the confusion matrix and AUC indicators, and the default probability of each sample is output.

8. The risk assessment system for power trading center based on multi-risk superposition according to claim 5 is characterized in that: Operational risk is modeled with the help of Bayesian networks, and the conditional probability relationship between multiple factors is analyzed, including: Based on the operation process of the trading center, define the nodes and directed edges of the Bayesian network, where each node represents a variable that affects the operational risk; Collect historical data, estimate the parameters of the probability distribution of each node, and use maximum likelihood estimation or expectation maximization algorithm to optimize the conditional probability; Use Bayesian inference algorithms to calculate the probability of operational risk in specific scenarios and analyze the contribution of different variables to risk; By continuously updating data and models, and conducting targeted risk control and prevention based on model results.

9. The risk assessment system for power trading center based on multi-risk superposition according to claim 7 is characterized in that: The risk level classification is performed based on the optimal weight combination in the weight combination, and the output risk level data includes: Genetic algorithms are used to generate a new generation of individuals through selection, crossover and mutation operations, and the performance of each group of weights is evaluated based on the fitness function, with the goal of minimizing the risk error; After multiple generations of evolution, the algorithm converges to the optimal weight combination, thereby deriving reasonable weights for each risk factor and outputting risk level data.

10. A risk assessment method for a power trading center based on multi-risk superposition, characterized in that: In the risk assessment system for a power trading center based on multi-risk superposition as described in any one of claims 1 to 9, the method comprises: The input terminal obtains the risk basic data and sends it to the processing terminal; The processing terminal receives the risk basic data; Using an adversarial network method, filling in abnormal values ​​and missing values ​​in the risk basic data to obtain first risk data; By using the grey correlation analysis method, the correlation between the influencing factors and the risks is determined based on the first risk data, and the first influencing factors of the multiple target risks are screened out; Analyze the probability relationship between the first influencing factors of the target risk in combination with multiple preset risk models; Based on the genetic algorithm, the multi-objective optimization of the target risk is performed to obtain a weighted combination of multiple target risks; Performing risk level classification based on the optimal weight combination in the weight combinations, and outputting risk level data; sending the risk level data to a display device; The display device generates a target image according to the risk level data, and displays the target image on a display of the display device; wherein the target image is used to characterize the distribution of the risk level of each target risk.

Images