An electricity transaction market manipulation quotation intelligent identification method and system
By extracting causal features of manipulated bids in the electricity market using a causal contrastive learning method, constructing a causal structure diagram and scoring it, the robustness and interpretability issues of existing technologies in identifying manipulated bids are resolved, achieving efficient and accurate identification of manipulative behavior and regulatory support.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ELECTRIC POWER RESEARCH INSTITUTE OF STATE GRID SHANDONG ELECTRIC POWER COMPANY
- Filing Date
- 2026-02-05
- Publication Date
- 2026-06-12
AI Technical Summary
Existing technologies are not robust enough in identifying manipulative bids in the electricity market, are easily circumvented, have insufficient model interpretability, are difficult to adapt to dynamic market changes, and lack the ability to model the causal relationship between the motivations for bidding behavior, the impact path, and the market feedback mechanism.
A causal contrastive learning mechanism is introduced, and feature representations with causal invariance are extracted through a behavioral encoder to construct a causal structure graph of pricing behavior. A manipulation risk scoring model is used for evaluation to generate a latent causal representation with causal invariance and policy robustness.
It improves the accuracy and stability of identifying manipulation behavior, has real-time performance and interpretability, adapts to multi-scenario deployment, supports multi-source data environments, and meets the needs of power market supervision.
Smart Images
Figure CN122199014A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power trading supervision technology, specifically to a method and system for intelligent identification of price manipulation in the power trading market. Background Technology
[0002] The spot and medium-to-long-term electricity trading mechanisms have been gradually improved during the electricity market reform, and the scale and types of market participants have continued to expand. With the increasing proportion of renewable energy and intensified fluctuations in electricity supply and demand, electricity market price formation increasingly relies on the bidding behavior and competitive strategies of multiple participants. Against this backdrop, identifying and preventing potential price manipulation in the market, and ensuring fair trading and reasonable prices, has become an important task for electricity trading centers and regulatory agencies.
[0003] In existing technologies, the identification of abnormal pricing and manipulation mainly employs statistical thresholding, rule matching, and traditional machine learning classification methods. For example, monitoring is conducted by setting thresholds for price deviation ranges, upper limits of price fluctuations, or transaction differences. However, these methods typically rely on prior experience parameters, making it difficult to adapt to changes in market conditions. Furthermore, they have weak identification capabilities for the concealment and flexibility of abnormal strategies, resulting in limited effectiveness in complex, multi-player game scenarios.
[0004] In recent years, some studies have attempted to apply deep learning and temporal modeling methods to price behavior recognition, such as using models like LSTM, Transformer, and Graph Neural Networks (GNN) to construct subject behavior features and perform anomaly classification. However, these models are generally based on correlation learning frameworks and are susceptible to market random fluctuations, noise injection, and adversarial perturbations, which can lead to false positives or false negatives in real trading scenarios. At the same time, deep models lack interpretability and are difficult to meet the requirements of regulatory agencies for audit tracing and manipulation evidence chains.
[0005] Furthermore, in the actual market, some entities deliberately evade detection by simulating normal fluctuations, shifting strategies across time periods, and using multiple accounts to coordinate false quotes. This makes it difficult for traditional anomaly detection methods to distinguish between "strategic quotes" and "manipulative quotes." Current technology lacks the ability to model the causal relationship between the motivations, impact paths, and market feedback mechanisms of quote behavior. The identification of manipulative behavior is more focused on whether the surface features are abnormal, rather than whether the underlying intentions are abnormal.
[0006] In summary, existing technologies have problems such as weak robustness, susceptibility to circumvention, insufficient model interpretability, and difficulty in adapting to dynamic market changes when dealing with the identification of manipulative pricing in the electricity market. Summary of the Invention
[0007] To address the aforementioned problems, this invention proposes an intelligent identification method and system for manipulating bids in the power trading market. It introduces a causal contrastive learning mechanism to extract causal invariant feature representations from bidding behavior and maintains a novel detection model with high identification stability when facing strategy disturbances, noise injection, and adversarial manipulation. This model can effectively support the needs of power market supervision and risk prevention.
[0008] According to some embodiments, the present invention adopts the following technical solution: A method for intelligently identifying price manipulation in the electricity trading market, comprising: Obtain data on the pricing behavior and market status of target market participants during the current trading cycle; The pricing behavior data and market state data are input into the trained behavior encoder to obtain a potential causal representation; Based on potential causal representation, the manipulation risk score of the target market entity in the current trading cycle is calculated using the trained manipulation risk scoring model. The behavioral encoder is trained by using causal contrastive learning to encode the bidding behavior data of the same subject under different market conditions. By constraining the similarity between positive and negative samples, a potential causal representation with causal invariance and policy robustness is generated.
[0009] Furthermore, the bidding behavior data is a sequence consisting of electricity prices and corresponding bidding quantities at different bidding tiers during the trading cycle; The market status data includes system load, peak demand indicators, meteorological characteristics, and market supply and demand deviation signals during the trading cycle.
[0010] Furthermore, the behavior encoder is expressed by the formula: in, For market entities During the trading cycle Potential causal representation, For market entities During the trading cycle Price quotation behavior data, For the trading cycle Market status data.
[0011] Furthermore, the manipulation risk scoring model is expressed by the formula: in, For market entities During the trading cycle Manipulation risk score, For the Sigmoid function, These are the model parameters.
[0012] Furthermore, the aforementioned causal contrastive learning specifically includes: Based on pricing behavior data under different market conditions, positive and negative sample pairs are constructed for the same market participant; By using a behavior encoder, the price quotation behavior data and market state data in the positive and negative sample pairs are encoded separately to obtain the potential causal representation of the positive and negative samples. The similarity of the latent causal representations of positive and negative sample pairs is calculated, and a contrastive loss function is constructed based on the similarity. The training process of the behavior encoder is constrained by minimizing the contrastive loss.
[0013] Furthermore, the construction of positive and negative samples for the same market entity specifically involves: Randomly sample batches of data from the dataset. ; From the sampled batch data, two samples of the same market entity under different market conditions are selected to form a positive sample pair. ,in, , These are two samples of the same market entity under different market conditions; For the sample Generate offset behavior with added policy perturbations , forming negative sample pairs .
[0014] According to some embodiments, the present invention adopts the following technical solution: A smart system for identifying price manipulation in the electricity trading market includes: The data acquisition module is configured to acquire pricing behavior data and market status data of the target market participants in the current trading period; The behavior encoding module is configured to input pricing behavior data and market state data into the trained behavior encoder to obtain a latent causal representation; The risk scoring module is configured to: calculate the manipulation risk score of the target market participant in the current trading period based on the potential causal representation and using the trained manipulation risk scoring model. The behavioral encoder is trained by using causal contrastive learning to encode the bidding behavior data of the same subject under different market conditions. By constraining the similarity between positive and negative samples, a potential causal representation with causal invariance and policy robustness is generated.
[0015] Furthermore, the bidding behavior data is a sequence consisting of electricity prices and corresponding bidding quantities at different bidding tiers during the trading cycle; The market status data includes system load, peak demand indicators, meteorological characteristics, and market supply and demand deviation signals during the trading cycle.
[0016] Furthermore, the behavior encoder is expressed by the formula: in, For market entities During the trading cycle Potential causal representation, For market entities During the trading cycle Price quotation behavior data, For the trading cycle Market status data.
[0017] Furthermore, the manipulation risk scoring model is expressed by the formula: in, For market entities During the trading cycle Manipulation risk score, For the Sigmoid function, These are the model parameters.
[0018] Furthermore, the aforementioned causal contrastive learning specifically includes: Based on pricing behavior data under different market conditions, positive and negative sample pairs are constructed for the same market participant; By using a behavior encoder, the price quotation behavior data and market state data in the positive and negative sample pairs are encoded separately to obtain the potential causal representation of the positive and negative samples. The similarity of the latent causal representations of positive and negative sample pairs is calculated, and a contrastive loss function is constructed based on the similarity. The training process of the behavior encoder is constrained by minimizing the contrastive loss.
[0019] Furthermore, the construction of positive and negative samples for the same market entity specifically involves: Randomly sample batches of data from the dataset. ; From the sampled batch data, two samples of the same market entity under different market conditions are selected to form a positive sample pair. ,in, , These are two samples of the same market entity under different market conditions; For the sample Generate offset behavior with added policy perturbations , forming negative sample pairs .
[0020] According to some embodiments, the present invention adopts the following technical solution: A computer program product includes a computer program that, when executed by a processor, performs the following steps: Obtain data on the pricing behavior and market status of target market participants during the current trading cycle; The pricing behavior data and market state data are input into the trained behavior encoder to obtain a potential causal representation; Based on potential causal representation, the manipulation risk score of the target market entity in the current trading cycle is calculated using the trained manipulation risk scoring model. The behavioral encoder is trained by using causal contrastive learning to encode the bidding behavior data of the same subject under different market conditions. By constraining the similarity between positive and negative samples, a potential causal representation with causal invariance and policy robustness is generated.
[0021] Furthermore, the bidding behavior data is a sequence consisting of electricity prices and corresponding bidding quantities at different bidding tiers during the trading cycle; The market status data includes system load, peak demand indicators, meteorological characteristics, and market supply and demand deviation signals during the trading cycle.
[0022] Furthermore, the behavior encoder is expressed by the formula: in, For market entities During the trading cycle Potential causal representation, For market entities During the trading cycle Price quotation behavior data, For the trading cycle Market status data.
[0023] Furthermore, the manipulation risk scoring model is expressed by the formula: in, For market entities During the trading cycle Manipulation risk score, For the Sigmoid function, These are the model parameters.
[0024] Furthermore, the aforementioned causal contrastive learning specifically includes: Based on pricing behavior data under different market conditions, positive and negative sample pairs are constructed for the same market participant; By using a behavior encoder, the price quotation behavior data and market state data in the positive and negative sample pairs are encoded separately to obtain the potential causal representation of the positive and negative samples. The similarity of the latent causal representations of positive and negative sample pairs is calculated, and a contrastive loss function is constructed based on the similarity. The training process of the behavior encoder is constrained by minimizing the contrastive loss.
[0025] Furthermore, the construction of positive and negative samples for the same market entity specifically involves: Randomly sample batches of data from the dataset. ; From the sampled batch data, two samples of the same market entity under different market conditions are selected to form a positive sample pair. ,in, , These are two samples of the same market entity under different market conditions; For the sample Generate offset behavior with added policy perturbations , forming negative sample pairs .
[0026] According to some embodiments, the present invention adopts the following technical solution: A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, perform the following steps: Obtain data on the pricing behavior and market status of target market participants during the current trading cycle; The pricing behavior data and market state data are input into the trained behavior encoder to obtain a potential causal representation; Based on potential causal representation, the manipulation risk score of the target market entity in the current trading cycle is calculated using the trained manipulation risk scoring model. The behavioral encoder is trained by using causal contrastive learning to encode the bidding behavior data of the same subject under different market conditions. By constraining the similarity between positive and negative samples, a potential causal representation with causal invariance and policy robustness is generated.
[0027] Furthermore, the bidding behavior data is a sequence consisting of electricity prices and corresponding bidding quantities at different bidding tiers during the trading cycle; The market status data includes system load, peak demand indicators, meteorological characteristics, and market supply and demand deviation signals during the trading cycle.
[0028] Furthermore, the behavior encoder is expressed by the formula: in, For market entities During the trading cycle Potential causal representation, For market entities During the trading cycle Price quotation behavior data, For the trading cycle Market status data.
[0029] Furthermore, the manipulation risk scoring model is expressed by the formula: in, For market entities During the trading cycle Manipulation risk score, For the Sigmoid function, These are the model parameters.
[0030] Furthermore, the aforementioned causal contrastive learning specifically includes: Based on pricing behavior data under different market conditions, positive and negative sample pairs are constructed for the same market participant; By using a behavior encoder, the price quotation behavior data and market state data in the positive and negative sample pairs are encoded separately to obtain the potential causal representation of the positive and negative samples. The similarity of the latent causal representations of positive and negative sample pairs is calculated, and a contrastive loss function is constructed based on the similarity. The training process of the behavior encoder is constrained by minimizing the contrastive loss.
[0031] Furthermore, the construction of positive and negative samples for the same market entity specifically involves: Randomly sample batches of data from the dataset. ; From the sampled batch data, two samples of the same market entity under different market conditions are selected to form a positive sample pair. ,in, , These are two samples of the same market entity under different market conditions; For the sample Generate offset behavior with added policy perturbations , forming negative sample pairs .
[0032] According to some embodiments, the present invention adopts the following technical solution: An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to perform the following steps: Obtain data on the pricing behavior and market status of target market participants during the current trading cycle; The pricing behavior data and market state data are input into the trained behavior encoder to obtain a potential causal representation; Based on potential causal representation, the manipulation risk score of the target market entity in the current trading cycle is calculated using the trained manipulation risk scoring model. The behavioral encoder is trained by using causal contrastive learning to encode the bidding behavior data of the same subject under different market conditions. By constraining the similarity between positive and negative samples, a potential causal representation with causal invariance and policy robustness is generated.
[0033] Furthermore, the bidding behavior data is a sequence consisting of electricity prices and corresponding bidding quantities at different bidding tiers during the trading cycle; The market status data includes system load, peak demand indicators, meteorological characteristics, and market supply and demand deviation signals during the trading cycle.
[0034] Furthermore, the behavior encoder is expressed by the formula: in, For market entities During the trading cycle Potential causal representation, For market entities During the trading cycle Price quotation behavior data, For the trading cycle Market status data.
[0035] Furthermore, the manipulation risk scoring model is expressed by the formula: in, For market entities During the trading cycle Manipulation risk score, For the Sigmoid function, These are the model parameters.
[0036] Furthermore, the aforementioned causal contrastive learning specifically includes: Based on pricing behavior data under different market conditions, positive and negative sample pairs are constructed for the same market participant; By using a behavior encoder, the price quotation behavior data and market state data in the positive and negative sample pairs are encoded separately to obtain the potential causal representation of the positive and negative samples. The similarity of the latent causal representations of positive and negative sample pairs is calculated, and a contrastive loss function is constructed based on the similarity. The training process of the behavior encoder is constrained by minimizing the contrastive loss.
[0037] Furthermore, the construction of positive and negative samples for the same market entity specifically involves: Randomly sample batches of data from the dataset. ; From the sampled batch data, two samples of the same market entity under different market conditions are selected to form a positive sample pair. ,in, , These are two samples of the same market entity under different market conditions; For the sample Generate offset behavior with added policy perturbations , forming negative sample pairs .
[0038] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This paper introduces causal contrastive learning to model the underlying intentions of electricity pricing behavior for the first time. Traditional methods for identifying pricing behavior generally rely on surface-level temporal or correlational features, making it difficult to characterize the decision-driven mechanisms behind market participants' pricing. This invention constructs a causal structure diagram of pricing behavior based on market mechanisms and transaction feedback paths, incorporating "market state → pricing behavior → price result" into a unified modeling framework to capture the causal driving chain of behavior. By constructing positive and negative sample pairs (normal pricing and strategy-induced pricing by the same entity), the model is guided to learn causal invariant representations, thereby gaining the ability to understand the essence of manipulative behavior.
[0039] 2. A robust contrastive encoder and discrimination mechanism for recognizing manipulation behavior are proposed. The behavior encoder designed in this invention can automatically learn the low-dimensional semantic representation of the subject's pricing behavior, and strengthen the semantic boundary under different strategies by comparing loss functions, so that the model can still maintain its recognition ability when facing market fluctuations, price disturbances and disguised behavior. Compared with traditional classification models, this mechanism significantly improves the detection accuracy and robustness against disturbances of covert manipulation behavior.
[0040] 3. Construct an efficient scoring mechanism that balances real-time performance with interpretability. This invention uses a lightweight scoring model to evaluate the causal representation of behavior, which can quickly output manipulation risk scores within the real-time trading cycle and set alarm thresholds for monitoring. Because the model has a clear structure and traceable path, it is convenient for the trading supervision platform to perform visual auditing and evidence chain extraction, thus meeting the dual requirements of "identification + interpretation" for intelligent supervision.
[0041] 4. Supports multi-scenario deployment and adapts to the complex systems and multi-source data environment of the electricity market. The model of this invention is adapted to different trading mechanisms in the electricity market, such as spot, intraday, and monthly bidding. It can automatically construct a causal relationship diagram based on the behavioral characteristics of participants and supports the fusion input of multi-source data (load, weather, fluctuation indicators, etc.). It has strong cross-regional and cross-system migration capabilities and is easy to deploy and integrate into existing market systems. Attached Figure Description
[0042] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0043] Figure 1 Here is a flowchart of an intelligent method for identifying manipulated bids in the power trading market, as described in Example 1. Figure 2 This is a graph showing the comparison of recognition accuracy between the proposed method and the comparative method under different disturbance intensities in a simulated market environment in Example 1. Figure 3 The graph shows the risk score stability of this method under different types of manipulation strategies (such as tiered pricing disturbances and false orders) in Example 1. Figure 4 This is a comparison chart of the interpretability visualization results when using causal contrastive learning and traditional supervised classification methods to identify the target subject's price quotation sequence in Example 1; Figure 5 This is a bar chart comparing the identification performance indicators of the method in Example 1 on the test set after deployment in different electricity market mechanisms (spot, day-ahead, intraday). Detailed Implementation
[0044] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0045] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0046] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0047] Example 1 One embodiment of the present invention provides a method for intelligent identification of price manipulation in the power trading market, comprising: Step S1: Obtain pricing behavior data and market status data of the target market participants in the current trading cycle; Step S2: Input the bidding behavior data and market state data into the trained behavior encoder to obtain the latent causal representation; Step S3: Based on the potential causal representation, calculate the manipulation risk score of the target market entity in the current trading cycle using the trained manipulation risk scoring model; The behavioral encoder is trained by using causal contrastive learning to encode the bidding behavior data of the same subject under different market conditions. By constraining the similarity between positive and negative samples, a potential causal representation with causal invariance and policy robustness is generated.
[0048] Furthermore, the bidding behavior data is a sequence consisting of electricity prices and corresponding bidding quantities at different bidding tiers during the trading cycle; The market status data includes system load, peak demand indicators, meteorological characteristics, and market supply and demand deviation signals during the trading cycle.
[0049] Furthermore, the behavior encoder is expressed by the formula: in, For market entities During the trading cycle Potential causal representation, For market entities During the trading cycle Price quotation behavior data, For the trading cycle Market status data.
[0050] Furthermore, the manipulation risk scoring model is expressed by the formula: in, For market entities During the trading cycle Manipulation risk score, For the Sigmoid function, These are the model parameters.
[0051] Furthermore, the aforementioned causal contrastive learning specifically includes: Based on pricing behavior data under different market conditions, positive and negative sample pairs are constructed for the same market participant; By using a behavior encoder, the price quotation behavior data and market state data in the positive and negative sample pairs are encoded separately to obtain the potential causal representation of the positive and negative samples. The similarity of the latent causal representations of positive and negative sample pairs is calculated, and a contrastive loss function is constructed based on the similarity. The training process of the behavior encoder is constrained by minimizing the contrastive loss.
[0052] Furthermore, the construction of positive and negative samples for the same market entity specifically involves: Randomly sample batches of data from the dataset. ; From the sampled batch data, two samples of the same market entity under different market conditions are selected to form a positive sample pair. ,in, , These are two samples of the same market entity under different market conditions; For the sample Generate offset behavior with added policy perturbations , forming negative sample pairs .
[0053] As one embodiment, the present invention provides an intelligent identification method for manipulating bids in the power trading market. This method extracts causally invariant feature representations from bidding behavior and employs a novel detection model that maintains high identification stability when facing strategy disturbances, noise injection, and adversarial manipulation. This effectively supports the needs of power market supervision and risk prevention. Figure 1 As shown, the specific implementation process is as follows: I. Data Representation and Symbol Explanation Suppose there are N market participants in the electricity market. Each market participant has bidding behavior data in each trading period t, and each trading period t has corresponding market status data and the final settlement price, as shown below: (1) Bidding behavior data , which is a sequence of price quote curves, can be expressed by the formula: in, For market entities In the trading period t, the first Electricity pricing in tiered pricing schemes. For market entities During the trading cycle The The volume of each price tier. For the trading cycle The total number of price tiers.
[0054] (2) Market status data, expressed by the formula: in, For the trading cycle The system load, For the trading cycle Peak demand indicators For the trading cycle meteorological characteristics, For the trading cycle This is a signal of a market supply-demand imbalance.
[0055] (3) The final settlement electricity price for trading period t As feedback information reflecting the market consequences of pricing behavior, it is used to constrain and discriminate potential causal representations, thereby ensuring that the learned representations can characterize strategic behaviors that have a substantial impact on market price formation. Specifically: Based on settlement electricity price Calculate risk score Risk score As a label to supervise learning; specifically, if the settlement electricity price is used during the transaction cycle t. relative benchmark price If the deviation of subject i is greater than the deviation of the previous trading period, then the behavior of subject i in period t is labeled as high risk; otherwise, it is labeled as low risk. Each level of risk has a preset risk score, expressed by the formula: in, , Representing the trading period t, respectively The settlement electricity price, , These represent the benchmark clearing price (a constant value provided by the electricity trading market) for the corresponding trading period. This is an indicator function that takes the value 1 when the condition inside the parentheses is true, and 0 otherwise.
[0056] II. Behavior Encoder Based on Causal Structure This embodiment constructs a causal chain of "market state → pricing behavior → price outcome", i.e., a causal structure: Its corresponding structural equation model (SEM) can be expressed as: in, This indicates the subject's pricing strategy intention, ( , Noise and unobservable influencing factors are included. For the same trading period The bidding behavior of all market participants, where N is the number of market participants.
[0057] Because of the main body's pricing strategy intention These are abstract, unobservable latent variables, and their true impact needs to be reflected through market settlement results. Therefore, this embodiment constructs a behavioral encoder to encode the pricing behavior of entities under a given market state and estimate the potential causal representation. To quantify the pricing strategy intent of the main entity Thus establishing market conditions With pricing behavior The relationship between them, as expressed by the behavior encoder, is: Furthermore, it remains unchanged in the face of market disturbances and strategic manipulation.
[0058] III. Using causal contrastive learning to train the behavior encoder To extract causal invariant representations, this embodiment introduces a contrastive learning framework, specifically: For market entities Construct positive and negative sample pairs, where the positive sample pairs represent behaviors under different normal states, denoted as: Negative sample pairs represent behaviors suspected of being manipulated or perturbed. ,in It is the policy offset behavior simulated by the perturbation generator.
[0059] The latent causal representation of three samples is estimated using a behavior encoder, expressed by the formula: Based on the estimated latent causal representation, cosine similarity is used to measure the similarity between pairs of samples, expressed by the formula: in, This represents the potential causal representation of the two samples to be measured.
[0060] Based on the above similarity metric, a contrastive loss function (InfoNCE) is constructed, expressed by the formula: in, This is the temperature coefficient.
[0061] This loss constraint maintains causal consistency for the same subject in different situations, while creating a differentiated stretch for manipulative behavior.
[0062] IV. Manipulation Risk Scoring Model The potential causal representation is input into the manipulation risk scoring model to score the manipulation risk, expressed by the formula: in, To manipulate risk scores, This is the Sigmoid function.
[0063] Set risk threshold When manipulating risk scores Then determine the market entity There is a tendency to manipulate during trading period t.
[0064] V. Model Training and Inference Process Algorithm 1: Training Process Input: Historical multi-source dataset Batch size ,temperature Balance coefficient Output: Behavior Encoder Manipulating risk scoring models Risk threshold step: 1. Initialize model parameters and ; 2. Repeat training until convergence: a. Randomly sample batches of data from the dataset. ; b. Construct positive sample pairs (same entity, different market conditions): c. Construct negative sample pairs (adding policy perturbations): d. Extract representations using a behavior encoder: e. Calculate the contrast loss: f. Calculate the risk score: g. If existing labels exist, calculate the supervision loss: in, For the predicted risk score, The risk score is used as a label.
[0065] h. Combination loss: i. Right Perform backpropagation and parameter update; 3. Use the validation set behavioral score distribution to set the threshold. ; 4. Output the trained model .
[0066] Algorithm 2: Online Manipulation Risk Reasoning and Alerts Input: New cycle data ,Model threshold .
[0067] Output: Risk score With alarm tags.
[0068] 1. Use a behavior encoder to extract latent causal representations: 2. Calculate the risk score using a manipulation risk scoring model: 3. If If the condition is met, an "operation risk warning" will be output; otherwise, it will be considered normal.
[0069] 4. Return the score and results.
[0070] This embodiment presents a comprehensive effectiveness verification experiment of the proposed method. Figure 2 This is a comparison of the recognition accuracy of our method and the comparison method under different perturbation intensities. Figure 3 The stability curves of the risk score of this method are shown for different types of manipulation strategies (such as tiered pricing disturbances and false orders). Figure 4 This is a visualization comparing the interpretability of interpretable results when using causal contrastive learning and traditional supervised classification methods to identify target subject price quote sequences. Figure 5 The bar charts show the performance of the proposed method on the test set after deployment in different electricity market mechanisms (spot, day-ahead, and intraday). Table 1 shows the comparison of the recognition accuracy and robustness of the proposed method with existing technologies in typical scenarios. Table 2 shows the trend of risk score changes (mean ± standard deviation) under different types of price manipulation. Table 3 shows the deployment effect of the proposed method under different electricity trading mechanisms (practicality evaluation).
[0071] Table 1. Comparison of Recognition Accuracy and Robustness in Experiments Table 2: Trend of Risk Score Change Table 3 Deployment Results From the above experimental results, it can be seen that the intelligent identification method for power trading manipulation pricing based on causal contrastive learning proposed in this embodiment has the following advantages: (1) Significantly improves the accuracy and stability of identifying manipulative pricing behavior. By constructing a causal relationship modeling and contrastive representation learning mechanism, the proposed method can accurately model the motivations and feedback paths behind bidding behavior, thereby effectively identifying bidding behavior that has been packaged, hidden, or strategically perturbed. Experimental results show that in challenging scenarios such as manipulation strategy injection, price perturbation, and time-shifting, the method outperforms traditional statistical monitoring and deep classification models in terms of recognition accuracy and recall, with an overall accuracy improvement of 12% to 38%.
[0072] (2) It has excellent model robustness and can resist various manipulation attacks or data disturbances. This embodiment employs a causal contrastive loss function and introduces various perturbation simulation strategies during the encoder training phase, thereby improving the model's robustness to external noise interference, random mutation behavior, and spoofing manipulation. In adversarial example testing, this method maintains a false positive rate of less than 5% and a correct recognition rate of over 90% in most scenarios.
[0073] (3) Enhance explainability to facilitate regulatory auditing and accountability. Because this method employs a causal path structure and representation of potential behavioral intentions, it not only outputs whether an anomaly is observed, but also provides causal explanations for the causes of the anomaly, including visualized pricing behavior trajectories and causal impact diagrams of market states, supporting post-event review and automatic evidence chain generation. This feature is significantly superior to traditional black-box neural network models and is suitable for market supervision systems that require compliant and reliable outputs.
[0074] (4) Support for multi-source data fusion and adaptation to heterogeneous market mechanisms This embodiment supports the integration of various heterogeneous data, such as transaction data, load forecasting, weather factors, and supply-demand deviations, into a unified modeling system, demonstrating strong data adaptability. Furthermore, it can be flexibly deployed under different electricity market mechanisms (such as spot, day-ahead, intraday, and bilateral contracts) and can operate in parallel without modifying market rules.
[0075] (5) Low computational resource consumption, suitable for real-time deployment The behavioral encoder and risk scoring model used in this embodiment have a lightweight structure, and the inference latency can be controlled within seconds, making it suitable for near real-time applications in actual power trading systems. Furthermore, this method can be embedded into existing price analysis modules without requiring additional data acquisition equipment or complex system modifications, demonstrating good engineering feasibility.
[0076] Example 2 One embodiment of the present invention provides an intelligent identification system for price manipulation in the power trading market, comprising: The data acquisition module is configured to acquire pricing behavior data and market status data of the target market participants in the current trading period; The behavior encoding module is configured to input pricing behavior data and market state data into the trained behavior encoder to obtain a latent causal representation; The risk scoring module is configured to: calculate the manipulation risk score of the target market participant in the current trading period based on the potential causal representation and using the trained manipulation risk scoring model. The behavioral encoder is trained by using causal contrastive learning to encode the bidding behavior data of the same subject under different market conditions. By constraining the similarity between positive and negative samples, a potential causal representation with causal invariance and policy robustness is generated.
[0077] Furthermore, the bidding behavior data is a sequence consisting of electricity prices and corresponding bidding quantities at different bidding tiers during the trading cycle; The market status data includes system load, peak demand indicators, meteorological characteristics, and market supply and demand deviation signals during the trading cycle.
[0078] Furthermore, the behavior encoder is expressed by the formula: in, For market entities During the trading cycle Potential causal representation, For market entities During the trading cycle Price quotation behavior data, For the trading cycle Market status data.
[0079] Furthermore, the manipulation risk scoring model is expressed by the formula: in, For market entities During the trading cycle Manipulation risk score, For the Sigmoid function, These are the model parameters.
[0080] Furthermore, the aforementioned causal contrastive learning specifically includes: Based on pricing behavior data under different market conditions, positive and negative sample pairs are constructed for the same market participant; By using a behavior encoder, the price quotation behavior data and market state data in the positive and negative sample pairs are encoded separately to obtain the potential causal representation of the positive and negative samples. The similarity of the latent causal representations of positive and negative sample pairs is calculated, and a contrastive loss function is constructed based on the similarity. The training process of the behavior encoder is constrained by minimizing the contrastive loss.
[0081] Furthermore, the construction of positive and negative samples for the same market entity specifically involves: Randomly sample batches of data from the dataset. ; From the sampled batch data, two samples of the same market entity under different market conditions are selected to form a positive sample pair. ,in, , These are two samples of the same market entity under different market conditions; For the sample Generate offset behavior with added policy perturbations , forming negative sample pairs .
[0082] Example 3 One embodiment of the present invention provides a computer program product, including a computer program that, when executed by a processor, performs the following steps: Obtain data on the pricing behavior and market status of target market participants during the current trading cycle; The pricing behavior data and market state data are input into the trained behavior encoder to obtain a potential causal representation; Based on potential causal representation, the manipulation risk score of the target market entity in the current trading cycle is calculated using the trained manipulation risk scoring model. The behavioral encoder is trained by using causal contrastive learning to encode the bidding behavior data of the same subject under different market conditions. By constraining the similarity between positive and negative samples, a potential causal representation with causal invariance and policy robustness is generated.
[0083] Furthermore, the bidding behavior data is a sequence consisting of electricity prices and corresponding bidding quantities at different bidding tiers during the trading cycle; The market status data includes system load, peak demand indicators, meteorological characteristics, and market supply and demand deviation signals during the trading cycle.
[0084] Furthermore, the behavior encoder is expressed by the formula: in, For market entities During the trading cycle Potential causal representation, For market entities During the trading cycle Price quotation behavior data, For the trading cycle Market status data.
[0085] Furthermore, the manipulation risk scoring model is expressed by the formula: in, For market entities During the trading cycle Manipulation risk score, For the Sigmoid function, These are the model parameters.
[0086] Furthermore, the aforementioned causal contrastive learning specifically includes: Based on pricing behavior data under different market conditions, positive and negative sample pairs are constructed for the same market participant; By using a behavior encoder, the price quotation behavior data and market state data in the positive and negative sample pairs are encoded separately to obtain the potential causal representation of the positive and negative samples. The similarity of the latent causal representations of positive and negative sample pairs is calculated, and a contrastive loss function is constructed based on the similarity. The training process of the behavior encoder is constrained by minimizing the contrastive loss.
[0087] Furthermore, the construction of positive and negative samples for the same market entity specifically involves: Randomly sample batches of data from the dataset. ; From the sampled batch data, two samples of the same market entity under different market conditions are selected to form a positive sample pair. ,in, , These are two samples of the same market entity under different market conditions; For the sample Generate offset behavior with added policy perturbations , forming negative sample pairs .
[0088] Example 4 One embodiment of the present invention provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the following steps: Obtain data on the pricing behavior and market status of target market participants during the current trading cycle; The pricing behavior data and market state data are input into the trained behavior encoder to obtain a potential causal representation; Based on potential causal representation, the manipulation risk score of the target market entity in the current trading cycle is calculated using the trained manipulation risk scoring model. The behavioral encoder is trained by using causal contrastive learning to encode the bidding behavior data of the same subject under different market conditions. By constraining the similarity between positive and negative samples, a potential causal representation with causal invariance and policy robustness is generated.
[0089] Furthermore, the bidding behavior data is a sequence consisting of electricity prices and corresponding bidding quantities at different bidding tiers during the trading cycle; The market status data includes system load, peak demand indicators, meteorological characteristics, and market supply and demand deviation signals during the trading cycle.
[0090] Furthermore, the behavior encoder is expressed by the formula: in, For market entities During the trading cycle Potential causal representation, For market entities During the trading cycle Price quotation behavior data, For the trading cycle Market status data.
[0091] Furthermore, the manipulation risk scoring model is expressed by the formula: in, For market entities During the trading cycle Manipulation risk score, For the Sigmoid function, These are the model parameters.
[0092] Furthermore, the aforementioned causal contrastive learning specifically includes: Based on pricing behavior data under different market conditions, positive and negative sample pairs are constructed for the same market participant; By using a behavior encoder, the price quotation behavior data and market state data in the positive and negative sample pairs are encoded separately to obtain the potential causal representation of the positive and negative samples. The similarity of the latent causal representations of positive and negative sample pairs is calculated, and a contrastive loss function is constructed based on the similarity. The training process of the behavior encoder is constrained by minimizing the contrastive loss.
[0093] Furthermore, the construction of positive and negative samples for the same market entity specifically involves: Randomly sample batches of data from the dataset. ; From the sampled batch data, two samples of the same market entity under different market conditions are selected to form a positive sample pair. ,in, , These are two samples of the same market entity under different market conditions; For the sample Generate offset behavior with added policy perturbations , forming negative sample pairs .
[0094] Example 5 One embodiment of the present invention provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to perform the following steps: Obtain data on the pricing behavior and market status of target market participants during the current trading cycle; The pricing behavior data and market state data are input into the trained behavior encoder to obtain a potential causal representation; Based on potential causal representation, the manipulation risk score of the target market entity in the current trading cycle is calculated using the trained manipulation risk scoring model. The behavioral encoder is trained by using causal contrastive learning to encode the bidding behavior data of the same subject under different market conditions. By constraining the similarity between positive and negative samples, a potential causal representation with causal invariance and policy robustness is generated.
[0095] Furthermore, the bidding behavior data is a sequence consisting of electricity prices and corresponding bidding quantities at different bidding tiers during the trading cycle; The market status data includes system load, peak demand indicators, meteorological characteristics, and market supply and demand deviation signals during the trading cycle.
[0096] Furthermore, the behavior encoder is expressed by the formula: in, For market entities During the trading cycle Potential causal representation, For market entities During the trading cycle Price quotation behavior data, For the trading cycle Market status data.
[0097] Furthermore, the manipulation risk scoring model is expressed by the formula: in, For market entities During the trading cycle Manipulation risk score, For the Sigmoid function, These are the model parameters.
[0098] Furthermore, the aforementioned causal contrastive learning specifically includes: Based on pricing behavior data under different market conditions, positive and negative sample pairs are constructed for the same market participant; By using a behavior encoder, the price quotation behavior data and market state data in the positive and negative sample pairs are encoded separately to obtain the potential causal representation of the positive and negative samples. The similarity of the latent causal representations of positive and negative sample pairs is calculated, and a contrastive loss function is constructed based on the similarity. The training process of the behavior encoder is constrained by minimizing the contrastive loss.
[0099] Furthermore, the construction of positive and negative samples for the same market entity specifically involves: Randomly sample batches of data from the dataset. ; From the sampled batch data, two samples of the same market entity under different market conditions are selected to form a positive sample pair. ,in, , These are two samples of the same market entity under different market conditions; For the sample Generate offset behavior with added policy perturbations , forming negative sample pairs .
[0100] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0101] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0102] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.
Claims
1. A method for intelligently identifying price manipulation in the electricity trading market, characterized in that, include: Obtain data on the pricing behavior and market status of target market participants during the current trading cycle; The pricing behavior data and market state data are input into the trained behavior encoder to obtain a potential causal representation; Based on potential causal representation, the manipulation risk score of the target market entity in the current trading cycle is calculated using the trained manipulation risk scoring model. The behavioral encoder is trained by using causal contrastive learning to encode the bidding behavior data of the same subject under different market conditions. By constraining the similarity between positive and negative samples, a potential causal representation with causal invariance and policy robustness is generated.
2. The intelligent identification method for manipulating bids in the power trading market as described in claim 1, characterized in that, The bidding behavior data is a sequence consisting of electricity prices and corresponding bidding quantities at different bidding tiers during the trading cycle; The market status data includes system load, peak demand indicators, meteorological characteristics, and market supply and demand deviation signals during the trading cycle.
3. The intelligent identification method for manipulating bids in the power trading market as described in claim 1, characterized in that, The behavior encoder is expressed by the formula: in, For market entities During the trading cycle Potential causal representation, For market entities During the trading cycle Price quotation behavior data, For the trading cycle Market status data.
4. The intelligent identification method for price manipulation in the power trading market as described in claim 1, characterized in that, The manipulation risk scoring model is expressed by the following formula: in, For market entities During the trading cycle Manipulation risk score, For the Sigmoid function, These are the model parameters.
5. The intelligent identification method for price manipulation in the power trading market as described in claim 1, characterized in that, The aforementioned causal comparison learning specifically includes: Based on pricing behavior data under different market conditions, positive and negative sample pairs are constructed for the same market participant; By using a behavior encoder, the price quotation behavior data and market state data in the positive and negative sample pairs are encoded separately to obtain the potential causal representation of the positive and negative samples. The similarity of the latent causal representations of positive and negative sample pairs is calculated, and a contrastive loss function is constructed based on the similarity. The training process of the behavior encoder is constrained by minimizing the contrastive loss.
6. The intelligent identification method for manipulating bids in the power trading market as described in claim 5, characterized in that, The construction of positive and negative samples for the same market entity is specifically as follows: Randomly sample batches of data from the dataset. ; From the sampled batch data, two samples of the same market entity under different market conditions are selected to form a positive sample pair. ,in, , These are two samples of the same market entity under different market conditions; For the sample Generate offset behavior with added policy perturbations , forming negative sample pairs .
7. A smart identification system for price manipulation in the electricity trading market, characterized in that, include: The data acquisition module is configured to acquire pricing behavior data and market status data of the target market participants in the current trading period; The behavior encoding module is configured to input pricing behavior data and market state data into the trained behavior encoder to obtain a latent causal representation; The risk scoring module is configured to: calculate the manipulation risk score of the target market participant in the current trading period based on the potential causal representation and using the trained manipulation risk scoring model. The behavioral encoder is trained by using causal contrastive learning to encode the bidding behavior data of the same subject under different market conditions. By constraining the similarity between positive and negative samples, a potential causal representation with causal invariance and policy robustness is generated.
8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the intelligent identification method for manipulating bids in the power trading market as described in any one of claims 1-6.
9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement a method for intelligent identification of manipulated pricing in the power trading market as described in any one of claims 1-6.
10. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform a method for intelligent identification of price manipulation in the power trading market as described in any one of claims 1-6.