Risk identification analysis method for electricity selling service

Abstract

The invention provides a risk identification analysis method for an electricity selling service, and the method comprises the following steps: obtaining a plurality of groups of first power data samples for the electricity selling service, carrying out the preprocessing of the first power data samples, and obtaining a standard power data sample; training a neural network model by using the first power data sample, and testing the neural network model through the standard power data sample to obtain a tested neural network model; acquiring multiple groups of second power data samples for risk identification, and preprocessing the multiple groups of second power data samples to obtain standard test data samples; inputting a standard test data sample into the trained neural network model to obtain corresponding first risk identification data; calculating second risk identification data in the second power data sample by using the first risk identification data; and according to the first risk identification data and the second risk identification data, judging a risk type set category to which the second power data sample belongs.

Description

Technical Field

[0001] The present invention belongs to the field of data analysis, and in particular relates to a risk identification and analysis method for electricity sales business. Background Art

[0002] The rapid growth of electricity sales in the modern electricity market presents both new opportunities and challenges for power companies. With intensified market competition, these companies face a variety of risks, including market, credit, operational, and policy risks. Therefore, establishing effective risk identification and analysis methods is crucial. These methods not only help companies identify potential risks but also provide a scientific basis for decision-making, ensuring sustainable development. However, current risk identification and analysis methods still suffer from several significant flaws in practical applications, limiting their effectiveness and reliability.

[0003] Existing risk identification methods often rely on historical data analysis, but in the rapidly changing electricity market, historical data cannot fully reflect future market dynamics and risk trends. Many companies overly rely on past market performance when conducting risk assessments, ignoring the impact of emerging market factors and policy changes. This leads to delayed and inaccurate risk assessments. Furthermore, traditional risk identification methods often use single qualitative or quantitative analysis methods, lacking comprehensiveness and systematicity, and are unable to fully assess the interplay between different risks. For example, the correlation between market risk and credit risk is often overlooked, which, in practice, can exacerbate a company's overall risk level. Secondly, current risk analysis tools and models are often complex and difficult to operate, and many companies lack the technical capabilities to effectively apply these tools. This complexity not only increases training costs for companies but also leads to human error in the risk identification process, compromising the accuracy of risk assessments.

[0004] Existing risk analysis systems are also relatively weak in real-time monitoring and data processing capabilities, making them unable to quickly respond to market changes and emergencies. In the electricity market, price fluctuations and demand changes are often volatile. Traditional methods cannot track and analyze these dynamic factors in real time, resulting in slow response by companies in addressing risks. Furthermore, many existing risk identification methods face obstacles in data integration and sharing. The electricity market involves multiple participants, including power generation companies, power sales companies, and regulatory agencies. Data from each party is often fragmented and difficult to integrate, leading to information asymmetry in the risk identification process. This information asymmetry prevents companies from having a comprehensive perspective when assessing risks, hindering the scientific nature of their decision-making. Furthermore, current risk management methods are insufficient to address emerging risks. With the reform of the electricity market and technological advancements, new risks such as cybersecurity risks, industry policy risks, market price risks, and assessment deviation risks have become increasingly prominent. Existing risk identification methods often fail to keep up to date and adapt to these new challenges, leaving companies helpless in the face of emerging risks.

[0005] Existing risk identification and analysis methods also lack sustainability and flexibility. Changing market conditions require companies to possess flexible risk management capabilities, but many traditional methods are often too rigid and lack the necessary adjustment mechanisms, making it difficult to meet the needs of companies in varying market environments. Therefore, to address these issues, there is an urgent need for a new risk identification and analysis method for electricity sales that comprehensively considers multiple risk factors, provides real-time monitoring and dynamic adjustment capabilities, and emphasizes data integration and sharing to enhance companies' risk management capabilities and decision-making in complex market environments. There is an urgent need for a risk identification and analysis method for electricity sales. Summary of the Invention

[0006] The present invention proposes a risk identification and analysis method for electricity sales business, which solves the accuracy and efficiency problems of risk identification and analysis in electricity sales business. By processing power data through a neural network model, it realizes the accurate identification and classification of potential risks.

[0007] The technical solution of the present invention is implemented as follows: a risk identification and analysis method for electricity sales business, the method comprising the following steps:

[0008] Acquire multiple groups of first power data samples for the power sales business, preprocess the first power data samples to obtain standard power data samples; train a neural network model using the first power data samples, and test the neural network model using the standard power data samples to obtain a tested neural network model;

[0009] Acquire multiple sets of second power data samples for risk identification, and preprocess the multiple sets of second power data samples to obtain standard test data samples; input the standard test data samples into the trained neural network model to obtain corresponding first risk identification data;

[0010] Calculating second risk identification data in the second power data sample using the first risk identification data; determining the risk type set category to which the second power data sample belongs based on the first risk identification data and the second risk identification data, and outputting third risk identification data for the risk;

[0011] Perform a comprehensive classification assessment on the risk type set in the third risk identification data output result, and store the assessment result in the risk knowledge base.

[0012] Traditional risk identification methods often rely on empirical rules or simple statistical analysis, lacking sufficient data support and intelligent processing capabilities. This approach results in inaccurate risk identification and a failure to promptly detect potential risks. This method, however, preprocesses multiple sets of power data samples and utilizes a neural network model for training and testing, making the risk identification process more scientific and systematic. Neural networks can learn complex patterns and features from large amounts of data, thereby improving the accuracy of risk identification.

[0013] The standardization and testing steps in this method enhance the model's generalization capabilities. Traditional methods often exhibit significant instability when faced with new data, resulting in poor risk identification. This method, however, standardizes the first and second power data samples to ensure consistency and comparability of the input data, thereby enhancing the model's adaptability and stability in practical applications. This standardization process, which is uncommon in existing technologies, can effectively reduce recognition errors caused by data inconsistencies.

[0014] This method innovates on multi-level risk identification analysis. By comparing and calculating primary and secondary risk identification data, it enables in-depth risk analysis and identifies a more detailed set of risk types. This multi-level risk analysis capability enables users to gain a more comprehensive understanding of potential risks and develop more effective response strategies. Traditional methods often only provide a single risk assessment, which is difficult to meet the needs of complex business environments. This method, however, provides a more comprehensive risk identification solution for electricity sales businesses.

[0015] The risk knowledge base established within this method provides crucial support for subsequent risk management and decision-making. By storing assessment results in the risk knowledge base, the system can continuously accumulate and update risk identification experience. This knowledge base not only helps improve the efficiency and accuracy of subsequent risk identification but also provides data support for decision-makers, enabling them to make more informed decisions based on historical data. Traditional methods often lack a systematic knowledge accumulation mechanism, which prevents them from fully utilizing historical experience in risk management. This method effectively addresses this deficiency through the establishment of a knowledge base.

[0016] This method emphasizes the integration of automation and intelligence in its implementation. By applying a neural network model, the entire risk identification process is highly automated, reducing manual intervention. This automated design not only improves identification efficiency but also reduces the risk of human error, ensuring the objectivity and accuracy of risk identification. Traditional methods typically rely on manual data analysis and judgment, which is inefficient and susceptible to subjective factors. This method, however, improves overall efficiency through intelligent means.

[0017] As a preferred embodiment, the first power data sample includes power sales business operation data and weather forecast data, and the second power data sample includes measured power data and measured meteorological data; during preprocessing, the first power data sample and the second power data sample are sequentially subjected to data noise filtering, missing data completion, statistical distribution test and normalization processing.

[0018] As a preferred embodiment, the process of preprocessing the first power data sample is to calculate the correlation between the original data from different sources and the same time node for the power sales business operation data; select a group of data with the highest correlation as the standard power data for cleaning processing, and use the interpolation method to fill the missing values of the non-standard data; perform anomaly detection on the filled data, and eliminate the outliers to obtain the processed first power data sample; for the original data from different sources in the meteorological forecast data, calculate the correlation between the original data from different sources and the same time node; select a group of data with the highest correlation as the standard power data for cleaning processing, and use the interpolation method to fill the missing values of the other data; perform anomaly detection on the filled data, and eliminate the outliers to obtain the processed first power data sample.

[0019] As a preferred embodiment, after the missing values of the first power data sample and the second power data sample are supplemented by the interpolation method, the following steps are also included: feature extraction of the first power data sample and the second power data sample; using a feature correlation analysis method to calculate the correlation between the original data from different sources; and based on the correlation, feature fusion of the original data.

[0020] As a preferred embodiment, during feature extraction, the correlation between original data from different sources is calculated through feature correlation analysis; based on the correlation, the original data is feature fused; based on the multimodal evaluation index of the fused data, a neural network feature fusion algorithm based on a support vector machine is used to fuse the multi-source data after feature extraction to obtain fused data.

[0021] As a preferred embodiment, the neural network model that has been tested is a neural network model based on a multi-layer perceptron; in the process of training the neural network model, the neural network model is evaluated and optimized by the mean absolute error and the coefficient of determination.

[0022] As a preferred embodiment, error weights are established based on the first risk identification data and the second risk identification data, multiple groups of second power data samples are classified and corresponding third risk identification data output results are given; the third risk identification data output results include output results corresponding to each group of second power data samples; the output results include the risk type set category to which the second power data sample belongs and the third risk identification data; the third risk identification data includes a classification threshold and a risk status assessment.

[0023] As a preferred embodiment, the risk type set of the second power data sample includes normal operating state, warning state, abnormal state and emergency state; the correspondence between the risk type set and the output result is calculated as follows: when 0<x≤x1, the second power data sample belongs to normal operating state; when x1<x≤x2, the second power data sample belongs to the warning state; when x2<x≤x3, the second power data sample belongs to the abnormal state; when x>x3, the second power data sample belongs to the emergency state; wherein, x is the value of the error weight, x1 is the threshold 1 of x, x2 is the threshold 2 of x, and x3 is the threshold 3 of x.

[0024] By employing the above technical solution, the present invention has the beneficial effect of: by applying a neural network model, the system can learn potential risk patterns from large amounts of power data and accurately identify various risk types. This accurate risk identification capability is crucial for electricity retailers, helping them to promptly identify potential problems and reduce the economic losses caused by risks.

[0025] The efficiency of this method improves the operational efficiency of electricity sales. Through automated data processing and risk analysis, companies can quickly respond to market changes and adjust operational strategies in a timely manner. This efficient processing capability can save companies significant labor and time costs, enabling them to maintain their advantage in a highly competitive market.

[0026] The established risk knowledge base provides companies with ongoing risk management support. By accumulating and analyzing historical risk data, companies can continuously optimize their risk identification models, improving the accuracy and efficiency of future risk identification. This continuous knowledge accumulation and optimization mechanism enables companies to more confidently navigate complex market environments.

[0027] This method's multi-level analytical capabilities enable enterprises to more comprehensively understand risks and develop more scientific response strategies. By meticulously categorizing risk types, enterprises can identify the sources and impacts of different risks and develop targeted preventative measures. This comprehensive risk management capability safeguards the sustainable development of enterprises and effectively reduces operational uncertainty. Intelligent design concepts enhance enterprises' technological proficiency and innovation capabilities. By introducing advanced neural network technology, enterprises can achieve a higher level of technological application in risk management, enhancing their market competitiveness. This technological innovation not only enhances enterprises' risk management capabilities but also lays a solid foundation for their subsequent technological development and business expansion. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0029] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0031] Example:

[0032] like Figure 1As shown, a risk identification and analysis method for electricity sales business. The risk identification and analysis method for electricity sales business has significant innovation and advantages over the existing technology in many aspects. First, the method obtains multiple groups of first electricity data samples and preprocesses them to ensure the standardization and consistency of the data. This process is often overlooked in the existing technology, resulting in uneven data quality during the model training process, affecting the accuracy and reliability of the model. By applying the first electricity data samples to the training of the neural network model, the method can establish a deep learning model that can adapt to the characteristics of the electricity market, while the existing technology mostly relies on traditional statistical methods and lacks the ability to deeply mine and learn complex data patterns.

[0033] In practical scenarios, the operational process of this approach is very clear. First, the electricity retail company collects a large number of electricity data samples from a variety of electricity users or market environments. Next, a data preprocessing step ensures that all data samples conform to a standard format, laying a solid foundation for subsequent model training. After training and testing, the neural network model effectively captures the underlying patterns in the electricity data and uses this to identify risks.

[0034] The next step is to obtain and preprocess multiple sets of second power data samples for risk identification. This process is similar to the first data processing process, but the focus is on identifying potential risks. By inputting the preprocessed standard test data samples into the trained neural network model, the corresponding first risk identification data can be quickly generated. This highly automated process greatly improves the efficiency of risk identification.

[0035] After obtaining the first risk identification data, the method further uses this data to calculate the second risk identification data for the second power data sample, forming a dynamic risk identification mechanism. Compared with existing technologies, this significantly improves the accuracy and timeliness of risk identification, as traditional methods often rely on static risk models and cannot adapt to the rapid changes in the market environment.

[0036] By comprehensively evaluating the risk type set contained in the output of the third-party risk identification data, power retailers can gain a clear understanding of the various risk factors in the current electricity market and store the assessment results in a risk knowledge base. This knowledge base not only provides a basis for subsequent decision-making but also serves as an important reference for future risk management and strategic adjustments.

[0037] This technical solution, by incorporating deep learning and data preprocessing technologies, constructs an efficient, accurate, and dynamic risk identification and analysis method. Compared to existing technologies, it can better cope with the complexity and uncertainty of the electricity market, thereby improving the security and decision-making capabilities of electricity sales operations. The implementation of this method will bring significant benefits and competitiveness to electricity sales companies in terms of risk control. This technical solution relates to a risk identification and analysis method for electricity sales operations, significantly improving the processing and risk identification capabilities of electricity data. Compared with existing technologies, this method is more systematic and intelligent. Existing technologies typically rely on rule-based risk assessment methods, which often fail to accurately capture potential risks in complex electricity data because they rely on manually set thresholds and rules and lack flexibility and adaptability. In contrast, this solution obtains multiple sets of first electricity data samples, preprocesses them, and generates standard electricity data samples. This data is then used to train a neural network model, which can automatically learn and identify risk patterns in the data. The tested neural network model has strong generalization capabilities and can adapt to changes in electricity data in different scenarios, improving the accuracy and efficiency of risk identification.

[0038] This application addresses the issue of relying on historical data through real-time data collection and analysis: Utilizing advanced IoT technology and data analysis platforms, it collects real-time information on electricity market transactions, price fluctuations, demand changes, and other information. Integrating machine learning algorithms, it deeply mines historical data while assigning greater weight to real-time data to more accurately predict future market dynamics and risk trends. Scenario simulation and forecasting: A scenario simulation model for the electricity market is constructed, considering various combinations of policy, economic, and technological factors. By simulating market dynamics under different scenarios, the risks faced by electricity retailers can be assessed and corresponding response strategies formulated.

[0039] This also addresses the inability to fully assess the interplay between different risks through a multi-dimensional risk assessment model: A multi-dimensional risk assessment model encompassing market risk, credit risk, operational risk, and policy risk is established. Using system dynamics, the interactions and impact paths between different risks are analyzed to form a risk network diagram.

[0040] Cross-risk early warning system: Based on the risk assessment model, a cross-risk early warning system is developed. When a risk indicator reaches the warning threshold, it automatically triggers the assessment of other related risks. Through the early warning system, enterprises can promptly identify and respond to potential risk combinations, reducing overall risk levels.

[0041] This addresses the issue of fragmented and difficult-to-integrate data from various parties. A data-sharing platform was established, inviting participation from various stakeholders, including power generation companies, power sales companies, and regulatory agencies. Data sharing protocols and standards were developed to ensure data accuracy, timeliness, and security. Data integration and analysis tools were developed to clean, integrate, and standardize data from various parties. Big data technologies and artificial intelligence algorithms were used to conduct in-depth data mining and analysis, generating valuable risk intelligence.

[0042] In a specific work scenario, power sales companies can regularly collect large amounts of electricity data samples. This data includes information such as user electricity usage, load variations, and peak and off-peak electricity prices. By preprocessing this data, the company can eliminate noise, fill in missing values, ensure data quality, and prepare for subsequent analysis. Next, the neural network model is trained using the preprocessed first power data sample. During the training process, the model automatically identifies key factors influencing power risk, thereby learning the potential risk characteristics of the power data.

[0043] After model training is complete, the power retailer can obtain multiple secondary power data samples for risk identification and perform the same preprocessing to generate standard test data samples. These samples are then fed into the trained neural network model, and the system generates corresponding primary risk identification data. This process enables the power retailer to monitor and analyze user electricity usage in real time, promptly identifying anomalies such as sudden increases in electricity consumption and abnormal loads, and thereby determine existing risks.

[0044] By utilizing the first risk identification data, the power sales company can calculate the second risk identification data for the second power data sample. This calculation process, based on the model's output, can provide an accurate risk assessment for each power data sample. Subsequently, combining the first and second risk identification data, the system can determine the risk type set category to which the second power data sample belongs and generate a third risk identification data output result specific to that risk. This data-driven risk identification method is more efficient and accurate than traditional manual assessment methods, enabling timely prevention of potential risks.

[0045] The system conducts a comprehensive classification and assessment of the risk types contained in the third-party risk identification data output. The results are stored in a risk knowledge base. This knowledge base not only records historical risk events and their resolution, but also provides empirical support for future risk identification. By continuously updating and maintaining the risk knowledge base, power retailers can gradually improve their ability to identify power risks, forming a continuous cycle of improvement.

[0046] This technical solution provides an efficient and intelligent risk identification and analysis method by introducing machine learning and data analysis technology, which significantly improves the risk management capabilities of the electricity sales business. The implementation of this solution can not only improve the work efficiency of the electricity sales company and reduce potential economic losses, but also provide users with safer and more reliable electricity services. As a preferred embodiment, the first power data sample includes power sales business operation data and weather forecast data, and the second power data sample includes measured power data and measured meteorological data; during preprocessing, the first power data sample and the second power data sample are sequentially subjected to data noise filtering, missing data completion, statistical distribution test and normalization processing.

[0047] The process of preprocessing the first power data sample is to calculate the correlation between the original data from different sources and the same time node for the power sales business operation data; select a group of data with the highest correlation as the standard power data for cleaning processing, and use the interpolation method to fill the missing values of the non-standard data; perform anomaly detection on the filled data, and eliminate the outliers to obtain the processed first power data sample; for the original data from different sources in the meteorological forecast data, calculate the correlation between the original data from different sources and the same time node; select a group of data with the highest correlation as the standard power data for cleaning processing, and use the interpolation method to fill the missing values of the other data; perform anomaly detection on the filled data, and eliminate the outliers to obtain the processed first power data sample.

[0048] In this technical solution, the preprocessing process for the first power data sample emphasizes the systematic and scientific nature of data correlation analysis and missing value filling. Compared with existing technologies, traditional data processing methods often lack effective correlation analysis for raw data from different sources, resulting in insufficient information utilization during data integration and difficulty in obtaining accurate standard power data. This solution calculates the correlation between raw data from different sources and the same time node, and selects the data with the highest correlation as the standard power data for cleaning. This process ensures that the selected data is more representative and accurate. For non-standard data, interpolation is used to fill missing values. Traditional methods rely solely on simple mean filling or median filling, which can easily introduce large errors. By performing anomaly detection on the completed data and removing outliers, this solution further improves data quality and ensures that the obtained first power data sample is more reliable. In a specific work scenario, the operator first collects multiple sets of power data samples, cleans and integrates the data through preprocessing steps, and then conducts subsequent analysis and prediction. This streamlined processing method effectively improves the efficiency and accuracy of data analysis.

[0049] Power sales business operational data typically covers various data directly related to the power sales business, including but not limited to: Electricity consumption data: users' actual electricity consumption, peak and valley values, and electricity load. This data is the foundational data for power sales operations and is used to formulate electricity consumption plans, adjust electricity consumption policies, and evaluate electricity efficiency. Electricity sales data: Data on the amount of electricity sold by power sales companies to users. This data reflects the company's sales performance and market share. Transaction data: Transaction data in the electricity market, such as electricity prices and transaction volumes. This data helps power sales companies understand market dynamics and competitive landscapes.

[0050] Measured power data refers to power data measured in real time by power monitoring equipment. It primarily includes: Real-time power data: Power data measured in real time by smart meters or other power monitoring equipment. This data reflects the actual operating status of the power system. Voltage and current data: Basic parameters of the power system, such as voltage and current, are crucial for its stability and security. Next, we analyze whether the data types of the two sample inputs are consistent.

[0051] The first electricity data sample includes electricity sales business operation data and weather forecast data. The electricity sales business operation data mainly belongs to electricity consumption data and electricity sales data (also includes some transaction data), while the weather forecast data belongs to external environment data, which is used to predict future weather conditions and the impact on electricity demand.

[0052] The second power data sample includes measured power data and measured meteorological data. The measured power data mainly belongs to real-time power data and power system operation data such as voltage and current data, while the measured meteorological data is weather data measured in real time by meteorological monitoring equipment.

[0053] From a data type perspective, the electricity sales data (particularly electricity consumption data) in the first power data sample and the measured power data (particularly real-time electricity consumption data) in the second power data sample are somewhat similar in nature, as they both reflect the power system's electricity consumption. However, while both weather forecast data and measured weather data are meteorological data, the former is a forecast value, while the latter is a measured value, resulting in differences in data accuracy and timeliness.

[0054] The electricity sales business operational data also includes transaction data and other data directly related to the electricity market, while the measured electricity data focuses more on the actual operating status of the power system. Therefore, strictly speaking, the data input of the two sets of samples are not completely consistent in type, but they both contain data related to electricity market operations and meteorological conditions, which provides a basis for learning and prediction of the neural network model. The electricity sales business operational data mainly corresponds to applied electricity data and electricity sales data (including some transaction data), while the measured electricity data mainly corresponds to real-time electricity data and power system operation data. The data input of the two sets of samples have certain similarities in type, but are not completely consistent.

[0055] After interpolation is used to fill missing values in the first and second power data samples, the method further includes the following steps: extracting features from the first and second power data samples; calculating the correlation between raw data from different sources using a feature correlation analysis method; and fusing features of the raw data based on the correlation. After interpolation is used to fill missing values in the first and second power data samples, feature extraction and feature fusion are further performed, thereby enhancing the depth and breadth of data analysis. This process is relatively uncommon in existing technologies, as traditional methods often focus on basic data processing and lack in-depth exploration of data features. By calculating the correlation between data from different sources through feature correlation analysis, potential data relationships can be discovered, providing a basis for subsequent decision-making. Fusing features of the raw data based on the correlation creates a more comprehensive feature set, making subsequent model training and prediction more accurate and effective. In practice, after completing data preprocessing, operators use feature extraction methods to extract key features and fuse them, laying the foundation for subsequent risk identification and analysis. This method can significantly improve the scientific and practical nature of the analysis.

[0056] During feature extraction, the correlation between original data from different sources is calculated through feature correlation analysis; based on the correlation, the original data is feature fused; based on the multimodal evaluation index of the fused data, a neural network feature fusion algorithm based on support vector machine is used to fuse the multi-source data after feature extraction to obtain fused data.

[0057] During the feature extraction process, the application of a neural network feature fusion algorithm based on support vector machines enables the effective fusion of multi-source data to obtain fused data. Compared with existing technologies, the introduction of this technology demonstrates a deep understanding and application capability of complex data processing. Traditional methods often lack the effective fusion of different features, resulting in poor model performance. However, this solution introduces advanced feature fusion algorithms to ensure that the advantages of data from different sources are fully utilized, thereby improving the accuracy and robustness of the model. After extracting features from the data, operators can achieve fusion through this algorithm to obtain richer feature information, thereby improving the depth and breadth of subsequent analysis, and can help identify complex power risks in practical applications.

[0058] The tested neural network model is based on a multi-layer perceptron. During the training process, the neural network model is evaluated and optimized using mean absolute error and coefficient of determination. This method uses mean absolute error and coefficient of determination to evaluate and optimize the model, a method that stands in stark contrast to the single-metric evaluation methods commonly used in the prior art. In the prior art, many systems rely solely on loss functions to evaluate model performance, lacking a comprehensive assessment of model accuracy. This solution, by combining multiple evaluation metrics, ensures a more scientific and comprehensive model training process. This process enables the model to be continuously optimized during training, finding the optimal parameter configuration and thus improving the final prediction results. In actual operation, after model training, operators analyze model performance using evaluation metrics and make adjustments and optimizations based on feedback to ensure that the model maintains efficient predictive capabilities when faced with real-world data.

[0059] As a preferred embodiment, error weights are established based on the first risk identification data and the second risk identification data, multiple groups of second power data samples are classified and corresponding third risk identification data output results are given; the third risk identification data output results include output results corresponding to each group of second power data samples; the output results include the risk type set category to which the second power data sample belongs and the third risk identification data; the third risk identification data includes a classification threshold and a risk status assessment.

[0060] This solution establishes error weights, classifies multiple groups of second power data samples, and outputs corresponding third risk identification data. The design of this process reflects systematic thinking on risk management. In existing technologies, risk identification often relies on static threshold judgments and cannot adapt to dynamically changing environments. However, this solution dynamically calculates error weights and classifies input data according to the weight values to form a set of risk types for normal operation, warning, abnormality, and emergency. This classification mechanism can promptly identify and respond to potential risks, ensuring that measures can be taken quickly to deal with them when risks occur. In specific work scenarios, operators can adjust business strategies in a timely manner based on the risk status fed back by the system to ensure business continuity and security, thereby effectively reducing the impact of risks. Through this series of steps, this technical solution achieves efficient identification and management of electricity sales business risks, improving the safety and reliability of overall operations.

[0061] The risk identification process typically involves multiple types of data used to assess and analyze potential risks. Based on the information you provided, "Third-level risk identification data includes classification thresholds and risk status assessments," we can infer that third-level risk identification data is used to further refine and quantify risk status. However, although you did not directly define primary and secondary risk identification data, we can reasonably interpret them based on the general risk identification process. Primary risk identification data generally refers to foundational risk data. This data serves as the starting point for risk identification and includes: Historical risk event records: past risk events, their impacts, and causes; Business operations data: Various data related to business operations, such as sales, costs, and customer satisfaction, which contain potential risk signals; External environmental data: Factors such as market trends, policy changes, and natural disaster warnings pose risks to the business. Through the collection, organization, and analysis of this data, a basic understanding of risk is formed.

[0062] Secondary risk identification data is preliminarily processed risk data. This data is derived through further analysis and mining based on primary risk identification data. It includes: Risk indicators: Calculated from basic data, these indicators include risk probability and risk impact. Risk trend analysis: Analyzing trends in historical risk events and business operations data to identify potential risk trends. Risk correlation analysis: Analyzing the correlations and mutual impacts between different risks. This data provides the basis for more in-depth risk assessments.

[0063] The risk type set of the second power data sample includes normal operating state, warning state, abnormal state and emergency state; the correspondence between the risk type set and the output result is calculated as follows: when 0<x≤x1, the second power data sample belongs to normal operating state; when x1<x≤x2, the second power data sample belongs to the warning state; when x2<x≤x3, the second power data sample belongs to the abnormal state; when x>x3, the second power data sample belongs to the emergency state; wherein, x is the value of the error weight, x1 is the threshold 1 of x, x2 is the threshold 2 of x, and x3 is the threshold 3 of x.

[0064] Determining x (the value of the error weight) is typically based on an assessment of the error's impact on the model or business, and can be determined through statistical analysis, empirical evaluation, or business logic. x1, x2, and x3 serve as thresholds for x, dividing the error into different intervals. These thresholds can be determined based on historical data distribution to ensure sufficient sample size for each interval; using key business nodes as thresholds based on business tolerance; or employing the technical staff's experience and opinions to determine them through expert experience. In short, the determination of x and x1, x2, and x3 requires comprehensive consideration of data characteristics, business requirements, and expert opinion to ensure their rationality and effectiveness.

[0065] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A risk identification and analysis method for electricity sales business, characterized in that: The method comprises the following steps: Acquire multiple groups of first power data samples for the power sales business, preprocess the first power data samples to obtain standard power data samples; train a neural network model using the first power data samples, and test the neural network model using the standard power data samples to obtain a tested neural network model; Acquire multiple sets of second power data samples for risk identification, and preprocess the multiple sets of second power data samples to obtain standard test data samples; input the standard test data samples into the trained neural network model to obtain corresponding first risk identification data; Calculating second risk identification data in the second power data sample using the first risk identification data; Determine, based on the first risk identification data and the second risk identification data, the risk type set category to which the second power data sample belongs, and output a third risk identification data result for the risk; Perform a comprehensive classification assessment on the risk type set in the third risk identification data output result, and store the assessment result in the risk knowledge base.

2. The risk identification and analysis method for electricity sales business according to claim 1, characterized in that: The first power data sample includes power sales business operation data and weather forecast data, and the second power data sample includes measured power data and measured meteorological data; during preprocessing, the first power data sample and the second power data sample are sequentially subjected to data noise filtering, missing data completion, statistical distribution test and normalization processing.

3. The risk identification and analysis method for electricity sales business according to claim 2, characterized in that: The process of preprocessing the first power data sample is to calculate the correlation between the original data from different sources and the same time node for the power sales business operation data; select a group of data with the highest correlation as the standard power data for cleaning processing, and use the interpolation method to fill the missing values of the non-standard data; perform anomaly detection on the filled data, and eliminate the outliers to obtain the processed first power data sample; for the original data from different sources in the meteorological forecast data, calculate the correlation between the original data from different sources and the same time node; select a group of data with the highest correlation as the standard power data for cleaning processing, and use the interpolation method to fill the missing values of the other data; perform anomaly detection on the filled data, and eliminate the outliers to obtain the processed first power data sample.

4. The risk identification and analysis method for electricity sales business according to claim 3, characterized in that: After the first power data sample and the second power data sample are supplemented with missing values by interpolation, the method further includes the following steps: extracting features from the first power data sample and the second power data sample; The feature correlation analysis method is used to calculate the correlation between original data from different sources; based on the correlation, the features of the original data are fused.

5. The risk identification and analysis method for electricity sales business according to claim 4, characterized in that: During feature extraction, the correlation between original data from different sources is calculated through feature correlation analysis; based on the correlation, the original data is feature fused; based on the multimodal evaluation index of the fused data, a neural network feature fusion algorithm based on support vector machine is used to fuse the multi-source data after feature extraction to obtain fused data.

6. The risk identification and analysis method for electricity sales business according to claim 1, characterized in that: The neural network model completed in the test is a neural network model based on a multi-layer perceptron; in the process of training the neural network model, the neural network model is evaluated and optimized by using the mean absolute error and the coefficient of determination.

7. The risk identification and analysis method for electricity sales business according to claim 1, characterized in that: Establishing error weights based on the first risk identification data and the second risk identification data, classifying multiple groups of second power data samples and providing corresponding third risk identification data output results; The third risk identification data output result includes an output result corresponding to each group of second power data samples; The output result includes the risk type set category to which the second power data sample belongs and the third risk identification data; The third risk identification data includes classification thresholds and risk status assessments.

8. The risk identification and analysis method for electricity sales business according to claim 7, characterized in that: The risk type set of the second power data sample includes normal operating state, warning state, abnormal state and emergency state; the correspondence between the risk type set and the output result is calculated as follows: when 0<x≤x1, the second power data sample belongs to normal operating state; when x1<x≤x2, the second power data sample belongs to the warning state; when x2<x≤x3, the second power data sample belongs to the abnormal state; when x>x3, the second power data sample belongs to the emergency state; wherein, x is the value of the error weight, x1 is the threshold 1 of x, x2 is the threshold 2 of x, and x3 is the threshold 3 of x.

Images