Transaction risk early warning system and method for electric power marketization reform

Abstract

The invention provides a transaction risk early warning system and method for electric power marketization reform. Relates to the technical field of electric power systems and electric power marketization reform, and comprises a risk data acquisition module used for acquiring multi-source data in the whole electric power marketization transaction process in real time; the risk data fusion module is used for performing time sequence synchronization, format standardization and heterogeneous data fusion on the multi-source data to generate a unified risk feature data set; and the risk assessment module is used for calculating market multi-dimensional risk indexes in real time through the risk feature data set and a Gaussian mixture model. According to the trading risk early warning system and method for electric power marketization reform, the false alarm rate and the missing report rate are remarkably reduced, and full-chain and all-weather monitoring and intelligent warning of price risks, performance risks, credit risks and market manipulation risks are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system and power market reform, in particular to a transaction risk early warning system and method for power market reform. BACKGROUND

[0002] At present, with the continuous deepening of power market reform, China's power transaction has developed from a single wholesale market to a multi-level and multi-variety comprehensive transaction system including spot market, auxiliary service market and long-term contract market. In order to cope with the characteristics of increasing transaction subjects, diversified transaction methods and sharp price fluctuations, the existing technology mainly adopts the following risk control means: first, the price prediction model based on time series model, neural network and other methods, which focuses on the prediction of intraday or monthly price trend; second, post-audit and risk control report, which assesses the risks that have occurred by manually reviewing transaction bills and contracts; third, professional risk control platform, which is mainly built by large power trading institutions or power grid enterprises, relying on static threshold and manual rules to centrally manage market data and transaction data and issue early warnings.

[0003] However, the above technical means still have obvious deficiencies: first, most systems remain at the post-analysis level and cannot realize real-time response to price mutations or transaction abnormalities; second, existing risk control focuses on price risk and lacks comprehensive monitoring of contract performance risk, credit risk and market manipulation risk throughout the transaction chain, making it difficult to meet the needs of current market participants and regulatory authorities for efficient and accurate risk early warning. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides a transaction risk early warning system and method for power market reform, which solves the problem of how to quickly predict and present risk situation in real time through Gaussian mixture model and deep reinforcement learning.

[0005] To achieve the above purpose, the present application realizes the following technical scheme: a transaction risk early warning system for power market reform, comprising: a risk data acquisition module for acquiring multi-source data in the whole process of power market transaction in real time; a risk data fusion module for time sequence synchronization, format standardization and heterogeneous data fusion of the multi-source data to generate a unified risk feature data set; a risk assessment module for real-time calculation of market multi-dimensional risk indicators through the risk feature data set and Gaussian mixture model; a dynamic threshold self-adaptive module for constructing a deep reinforcement learning agent, which generates and optimizes risk early warning thresholds online through a policy network to minimize false alarm rate and missed alarm rate; An early warning decision module is configured to compare the market multi-dimensional risk indicators with the risk early warning threshold in real time and generate a risk early warning signal. An early warning alarm module is configured to visually present the risk early warning signal and historical risk evolution trend and provide a multi-level and multi-dimensional risk report.

[0006] Preferably, the multi-source data includes market clearing price data, contract performance information, metering data and transaction subject credit rating data.

[0007] Preferably, the market multi-dimensional risk indicators include price risk, performance risk, credit risk and market manipulation risk.

[0008] Preferably, the Gaussian mixture model calculates the risk score through the following model formula: , wherein, is the fused risk feature vector, is the total number of Gaussian components, is the weight of the kth component in the jth risk model, satisfying , is the multivariate Gaussian probability density function with mean and covariance matrix as parameters.

[0009] Preferably, the dynamic threshold adaptive module adopts a deep reinforcement learning algorithm to construct a deep reinforcement learning agent, and the deep reinforcement learning algorithm optimizes the early warning threshold online through the following model formula: , wherein, is a state vector containing multi-dimensional risk indicators and false alarm and missed alarm historical feedback, is an increasing or decreasing action for each risk category threshold, is a learning rate, is a discount factor, is an immediate reward given according to the current early warning result, is the next state after the action , is a value function of the state-action pair corresponding to the agent at time t.

[0010] Preferably, the multi-level and multi-dimensional risk report is an integrated visual analysis of real-time indicator values and historical evolution trends of price risk, performance risk, credit risk and market manipulation risk classified into three levels of low risk, medium risk and high risk.

[0011] Preferably, the risk feature dataset and market multi-dimensional risk indicators are automatically signed and hash-verified by smart contracts when written into the distributed ledger to ensure data tamper-proofing and traceability.

[0012] A transaction risk early warning method for power market reform, comprising: S1. Real-time acquisition of multi-source data in the whole process of power market transaction, the multi-source data comprising market clearing price data, contract performance information, metering data and transaction subject credit rating data; S2. Time sequence synchronization and format standardization of the multi-source data to generate a risk feature dataset; S3. Based on the risk feature dataset, real-time risk indicators of price risk, performance risk, credit risk and market manipulation risk are calculated respectively by using a Gaussian mixture model; S4. A deep reinforcement learning agent is constructed, and the early warning threshold is updated online by a deep Q network to minimize the false alarm rate and the missed alarm rate; S5. The real-time risk indicators are compared with the early warning threshold one by one to generate a risk early warning signal.

[0013] The application provides a transaction risk early warning system and method for power market reform. The transaction risk early warning system and method for power market reform comprehensively improve the real-time performance and coverage of power transaction risk early warning. Through the close cooperation of the five modules of "risk data collection-fusion-evaluation-decision", the market clearing price, contract performance, metering and credit rating and other multi-source heterogeneous data can be synchronized and fused online at the second level, and the multi-dimensional risk indicators are calculated in real time based on the Gaussian mixture model. Especially, the dynamic threshold self-adaptive mechanism of deep reinforcement learning is introduced, so that the early warning threshold can be continuously self-optimized with market fluctuations and historical false alarm feedback, which significantly reduces the false alarm rate and the missed alarm rate, and realizes the all-chain and all-weather monitoring and intelligent alarm of price risk, performance risk, credit risk and market manipulation risk.

[0014] In addition, the present scheme also performs outstandingly in system architecture and data security. On the one hand, the modules are seamlessly integrated through standardized interfaces, which is convenient for subsequent function expansion and cross-platform deployment; on the other hand, the risk feature data and evaluation results are written into the blockchain distributed ledger, and are automatically signed and hash-verified by smart contracts, so that the historical records are tamper-proofing and traceable, meeting the regulatory compliance and audit requirements. The visual multi-level and multi-dimensional risk report provides intuitive and reliable decision support for market participants and regulatory departments, and improves the risk management level of the whole industry under the background of power market reform. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1is a flowchart for realizing the application. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the application.

[0017] As Figure 1 shown, the embodiment of the application provides a transaction risk early warning system and method for power market reform, comprising a risk data acquisition module for acquiring multi-source data in the whole process of power market transaction in real time. The multi-source data comprises market clearing price data, contract performance information, metering data and transaction subject credit rating data.

[0018] A risk data fusion module is configured to perform time sequence synchronization, format standardization and heterogeneous data fusion on the multi-source data, and generate a unified risk feature data set.

[0019] The specific implementation is as follows: Risk data acquisition: In this embodiment, the risk data acquisition module works in parallel through four data channels: Market clearing price: The interface of the provincial power trading center is called once every second to obtain the real-time clearing price of the day.

[0020] For example, at 14:00 on June 30, 2025, the clearing price collected by the system is 356.75 yuan / MWh.

[0021] Contract performance information:

[0022] The performance rate of each signed contract is queried from the contract management system every 5 seconds.

[0023] For example, the performance rate of contract number C12345 at the same time is 93.2%.

[0024] Metering data: The total meter reading is collected once every second through the SCADA system, and converted to megawatt hours.

[0025] For example, the sampling value at 14:00 on June 30, 2025 is 123.4567 MWh.

[0026] Transaction subject credit rating: The subject credit score is updated from a third-party credit rating agency every 24 hours, and is mapped to the 0.00-1.00 interval in proportion.

[0027] For example, the credit score of the main body G456 is 1.00 after the daily update.

[0028] Each type of raw data described above has a UTC standard timestamp and is synchronously pushed to the downstream fusion module through the enterprise internal message bus.

[0029] Risk data fusion: The risk data fusion module processes the collected multi-source data according to the following process: Time alignment and filling: The above four types of data are aligned by timestamp per second. If the contract performance rate or other data cannot be collected at a certain second, the last valid value is used for filling to ensure continuous data sequence without gaps.

[0030] Format standardization: The aligned price, performance rate, metering reading, and credit score are uniformly converted into floating-point number format with decimal points to ensure consistent field names and formats in the whole system.

[0031] Weighted sampling and state estimation: The system performs weighted sampling on the standardized multi-dimensional data based on the particle filtering idea, and updates the weight combined with observation to finally adaptively estimate the current state of the four indicators.

[0032] For example, the risk features output after fusion processing are: price 356.80 yuan / MWh, performance rate 93.3%, metering 123.46 MWh, and credit score 1.00.

[0033] The risk assessment module is used to calculate market multi-dimensional risk indicators in real time through the risk feature data set and the Gaussian mixture model. The market multi-dimensional risk indicators include price risk, performance risk, credit risk, and market manipulation risk.

[0034] The Gaussian mixture model calculates the risk score through the following model formula: , wherein, is the fused risk feature vector, is the total number of Gaussian components, is the weight of the kth component in the jth risk model, satisfying , is the multivariate Gaussian probability density function with mean and covariance matrix as parameters.

[0035] The specific implementation is as follows: Input risk feature vector: At the sampling time of 14:00 on June 30, 2025, the four-dimensional vector output by the fusion module is: Electricity price: 356.80 yuan / MWh.

[0036] Performance rate: 0.933, 93.3%.

[0037] Meter reading: 123.46 MWh.

[0038] Credit score: 1.00.

[0039] Model parameter settings: For price risk, the pre-trained GMM uses 3 Gaussian components, with the following parameter settings: Component weights: 0.40, 0.35, 0.25.

[0040] Component mean: 1) [350.0, 0.920, 120.0, 0.950]; 2) [360.0, 0.940, 125.0, 1.000]; 3) [370.0, 0.960, 130.0, 1.050].

[0041] The covariance matrix is in diagonal form, and the diagonal elements (variances) are respectively: 1) [10.0², 0.02², 5.0², 0.03²]; 2) [8.0², 0.015², 4.0², 0.02²]; 3) [12.0², 0.025², 6.0², 0.04²];

[0042] Component probability density calculation: Substitute the input vector into each Gaussian component in turn, and the calculated probability density values are approximately: Component 1: 0.0020; Component 2: 0.0030; Component 3: 0.0010.

[0043] Weighted sum and risk score: Weighted sum according to weight: 0.40 × 0.0020 + 0.35 × 0.0030 + 0.25 × 0.0010 ≈ 0.0021.

[0044] The system takes the negative logarithm of this weighted sum, and the price risk score is approximately 6.17.

[0045] Other risk indicator processing: The GMMs with the same structure but parameters adjusted according to the characteristics of each risk are used to calculate the scores of "counterparty risk", "credit risk" and "market manipulation risk" respectively, for example: The counterparty risk score ≈ 4.85.

[0046] The credit risk score ≈ 2.30.

[0047] The market manipulation risk score ≈ 5.92.

[0048] The output multi-dimensional risk indicators: The risk assessment module finally outputs four real-time risk scores as inputs for the subsequent dynamic threshold adaptation and early warning decision-making.

[0049] The dynamic threshold adaptation module is used to build a deep reinforcement learning agent that generates and optimizes risk warning thresholds online through a policy network to minimize false alarm rates and missed alarm rates.

[0050] The dynamic threshold adaptation module uses a deep reinforcement learning algorithm to build a deep reinforcement learning agent that optimizes the warning thresholds online through the following model formula: , wherein, is a state vector containing multi-dimensional risk indicators and historical feedback of false alarms and missed alarms, is an action of increasing or decreasing the threshold value for each risk category, is the learning rate, is the discount factor, is the immediate reward given according to the results of this warning, is the next state after the action , is the value function of the state-action pair corresponding to the agent at time t.

[0051] The early warning decision-making module is used to compare the multi-dimensional risk indicators of the market with the risk warning thresholds in real time and generate risk warning signals.

[0052] The early warning alert module is used to visually present the risk warning signals and historical risk evolution trends and provide multi-level and multi-dimensional risk reports.

[0053] The multi-level and multi-dimensional risk reports are integrated visual analyses that display the real-time indicator values and historical evolution trends of price risk, counterparty risk, credit risk and market manipulation risk in three levels of low risk, medium risk and high risk.

[0054] The risk characteristic data set and the market multi-dimensional risk indicators are automatically signed and hashed by smart contracts when written into the distributed ledger to ensure data integrity and traceability.

[0055] The specific implementation is as follows: Application scenario: dynamic threshold self-adaptation and early warning process in a high false alarm rate scenario.

[0056] In the stage of smooth operation of the electricity market but initial parameter optimization of the monitoring system, a large number of false alarms are prone to occur due to low threshold setting. At this time, the dynamic threshold self-adaptation module detects that the false alarm rate in the past week is continuously higher than 30%, and the agent automatically increases the risk warning thresholds of various types in the strategy network: Action execution: the thresholds of price risk, performance risk, credit risk and manipulation risk are respectively increased by 5% to 10%.

[0057] Threshold update: after online updating, the price risk threshold is increased from the original 6.0 to 6.6, the performance risk threshold is increased from 5.0 to 5.5, and so on.

[0058] The early warning decision module then compares the multi-dimensional risk indicators calculated in real time with the new thresholds. Due to the increase of the thresholds, the original low-value fluctuations no longer trigger an alarm, and an early warning signal is only sent out when the indicators exceed “price risk ≥ 6.6” or “manipulation risk ≥ 5.5”.

[0059] After receiving the signal, the early warning module only reminds a few moments when there is a real deviation in the risk situation board, and displays the trend of “false alarm rate from 30% to 15%” in the form of a line chart in the multi-level report, helping the operation and maintenance personnel to verify and confirm the threshold adjustment effect.

[0060] A transaction risk early warning method for electricity market reform, comprising: S1. Real-time acquisition of multi-source data in the whole process of electricity market transaction, the multi-source data including market clearing price data, contract performance information, metering data and transaction subject credit rating data.

[0061] S2. Time sequence synchronization and format standardization of the multi-source data to generate a risk feature data set.

[0062] S3. Based on the risk feature data set, the real-time risk indicators of price risk, performance risk, credit risk and market manipulation risk are calculated respectively by using a Gaussian mixture model.

[0063] S4. Constructing a deep reinforcement learning agent to update the early warning thresholds online through a deep Q network to minimize the false alarm rate and the missed alarm rate.

[0064] S5. Comparing the real-time risk indicators with the early warning thresholds one by one to generate a risk early warning signal.

[0065] Embodiment two

[0066] Different from the first embodiment, the application scenario of this embodiment is dynamic threshold self-adaptation and early warning process in the high false negative rate scenario.

[0067] When the power grid faces large-scale renewable energy access and dramatic increase in transaction fluctuations, the system may produce false negatives due to excessively high thresholds in the early stage. The dynamic threshold self-adaptation module detects that the false negative rate has exceeded 20% in the past 3 days, and the agent automatically reduces the risk thresholds to improve sensitivity: Action execution: reduce all risk category thresholds by 8%~12%.

[0068] Threshold update: after adjustment, the price risk threshold is reduced from 7.2 to 6.5, and the manipulation risk threshold is reduced from 6.0 to 5.3, etc.

[0069] The early warning decision module uses the reduced thresholds to compare real-time risk indicators and capture edge anomalies earlier. When the price risk indicator rises from 6.4 to 6.6, an early warning is triggered immediately; when the manipulation risk indicator, which was not detected before, reaches 5.4, an alarm is also timely.

[0070] The early warning alarm module presents the comparison of new and old thresholds and the improvement of early warning timeliness in a multi-dimensional report in the form of a column chart, and highlights the improvement effect of "false negative rate from 20% to 8%", helping market supervision parties to intuitively evaluate the accuracy of the alarm improved by threshold self-adaptation.

[0071] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A transaction risk early warning system for power market reform, characterized in that, The application relates to a risk early warning method for an electricity market, which comprises the following steps: a risk data collection module is used to collect multi-source data in a whole process of electricity market transaction in real time; a risk data fusion module is used to perform time sequence synchronization, format standardization and heterogeneous data fusion on the multi-source data, and generate a unified risk feature data set; a risk assessment module is used to calculate market multi-dimensional risk indexes in real time through the risk feature data set and a Gaussian mixture model; a dynamic threshold self-adaptive module is used to construct a deep reinforcement learning agent, and the deep reinforcement learning agent generates and optimizes a risk early warning threshold value online through a policy network; an early warning decision module is used to compare the market multi-dimensional risk indexes in real time according to the risk early warning threshold value, and generate a risk early warning signal; an early warning alarm module is used to visually present the risk early warning signal and a historical risk evolution trend, and provide a multi-level and multi-dimensional risk report.

2. The transaction risk early warning system for power market reform according to claim 1, characterized in that: The multi-source data comprises market clearing price data, contract performance information, metering data and transaction subject credit rating data.

3. The transaction risk early warning system for power market reform according to claim 1, characterized in that: The market multi-dimensional risk indexes comprise price risk, performance risk, credit risk and market manipulation risk.

4. The transaction risk early warning system for power market reform according to claim 1, characterized in that: The Gaussian mixture model calculates a risk score through the following model formula: ， wherein, is the fused risk feature vector, is the total number of Gaussian components, is the weight of the k-th component in the j-th risk model, satisfying , is a multivariate Gaussian probability density function with mean and covariance matrix parameters.

5. The transaction risk early warning system for power market reform according to claim 1, characterized in that: The dynamic threshold self-adaptive module adopts a deep reinforcement learning algorithm to construct a deep reinforcement learning agent, and the deep reinforcement learning algorithm optimizes the early warning threshold value online through the following model formula: ， wherein, is a state vector comprising multi-dimensional risk indicators and false positive / negative history feedbacks, is an increase / decrease action on each risk category threshold, is a learning rate, is a discount factor, is an immediate reward given according to the current warning result, is an action next state after the action is a value function of the state-action pair corresponding to the agent at time t.

6. The transaction risk early warning system for power market reform according to claim 1, characterized in that: The multi-level and multi-dimensional risk report is an integrated visual analysis of real-time index values and historical evolution trends of price risk, performance risk, credit risk and market manipulation risk, which are classified into three levels of low risk, medium risk and high risk.

7. The transaction risk early warning system for power market reform according to claim 1, characterized in that: The risk feature data set and the market multi-dimensional risk indexes are automatically signed and hash-verified through a smart contract when being written into a distributed ledger.

8. A transaction risk early warning method for power market reform, characterized in that, The application further provides a risk early warning method for an electricity market, which comprises the following steps: S1. multi-source data in a whole process of electricity market transaction is acquired in real time, and the multi-source data comprises market clearing price data, contract performance information, metering data and transaction subject credit rating data; S2. time sequence synchronization and format standardization are performed on the multi-source data, and a risk feature data set is generated; S3. real-time risk indexes of price risk, performance risk, credit risk and market manipulation risk are calculated through a Gaussian mixture model based on the risk feature data set; S4. a deep reinforcement learning agent is constructed, and a deep Q network is used to update an early warning threshold value online; S5. the real-time risk indexes are compared with the early warning threshold value one by one, and a risk early warning signal is generated.

Images