Method for evaluating operation of electricity spot market based on multi-factor evaluation model

Abstract

The application discloses a power spot market operation evaluation method based on a multi-factor evaluation model, relates to the technical field of power spot market operation evaluation, and comprises the following steps: performing initial market simulation based on back testing of historical data and obtaining a first state set; adopting a clustering algorithm to perform feature division on the first state set, taking the division result as a preset Bayesian network input signal, and constructing an operation evaluation decision model through the Bayesian network; deciding on a preset first action set according to the operation evaluation decision model to obtain a simulation alternation sequence; adopting a preset adversarial network to perform feature mapping on the simulation alternation sequence to obtain a projection offset feature; and supplementing the state transition probability of the preset Bayesian network according to the projection offset feature to optimize the operation evaluation decision model, thereby solving the problem of mode switching decision distortion caused by feature space mismatch in the traditional evaluation method.

Description

Technical Field

[0001] This invention relates to the field of electricity spot market operation technology, and more specifically, to an electricity spot market operation evaluation method based on a multi-factor evaluation model. Background Technology

[0002] The electricity spot market refers to a market where qualified operators conduct day-ahead, intraday, and real-time electricity trading. In this market, electricity prices are determined by a market clearing mechanism, typically based on the intersection of the supply and demand curves. A key characteristic of the electricity spot market is its high price volatility; electricity prices fluctuate with changes in supply and demand. Therefore, the stability and operational efficiency of the electricity spot market are crucial to the economic efficiency and security of the power system.

[0003] The electricity spot market should proceed with simulated trial operation, settlement trial operation, and formal operation in sequence. During this process, various factors, including market mechanisms, price discovery, trading rules, and the behavior of market participants, need to be comprehensively evaluated. Market operation involves complex resource allocation, market rule design, and supply-demand balance, especially given the increasing proportion of large-scale renewable energy sources (such as wind and solar power). Ensuring stable market operation is a significant challenge. The randomness, intermittency, and volatility of renewable energy sources exacerbate price fluctuations in the electricity market and increase the uncertainty of market decisions.

[0004] For example, the invention patent announcement CN113657705B discloses a method, device, and storage medium for assessing the impact of electricity spot market parameters, which solves the problem in existing technologies of being able to specifically construct and analyze important factors affecting the electricity market. This invention includes the following steps: S1: Establishing an assessment model, setting economic assessment indicators, electricity volume assessment indicators, operational assessment indicators, and other assessment indicators within the model; S2: Determining parameters that have a critical impact on the electricity market; S3: Determining the optimal assessment indicators; S4: Assessing the impact of parameters on the electricity market based on the number of optimal assessment indicators, and determining the intensity of the parameter's impact based on the quantitative data of the optimal assessment indicators. This invention has advantages such as accurately assessing the operational status of the electricity market and having a wide range of applications.

[0005] For example, the invention patent announcement CN111754076B discloses a method and equipment applicable to the evaluation of electricity wholesale market models. This method involves simulating the clearing of two or more market models for a target electricity market within the same regional power grid, and outputting comparative results based on the simulation clearing results of each market model, according to comparative indicators. These comparative indicators include one or more of the following: inter-provincial market connectivity, day-ahead optimization potential, total daily generation cost, and clean energy consumption. This disclosure also provides output equipment, storage equipment, and computing equipment for using the above method. This disclosure provides a specific method and equipment for evaluating the selection of electricity market models for the construction of regional electricity spot markets in my country.

[0006] The above-disclosed technical solutions have at least the following technical problems:

[0007] Existing technologies for assessing the operation of the electricity spot market, such as Monte Carlo simulation or agent-based simulation, require manual pre-setting of switching conditions, lacking the ability to dynamically switch assessment strategies and adjust assessment actions in a timely manner. Furthermore, because market conditions are dynamic, assessment actions that are incompatible with market conditions may be selected during strategy switching, leading to prediction failures or misjudgments of risk. To address these problems, this invention proposes a solution. Summary of the Invention

[0008] To overcome the aforementioned deficiencies of existing technologies, embodiments of the present invention provide a method for evaluating the operation of the electricity spot market based on a multi-factor evaluation model. This method constructs a market simulation model through backtesting of historical data and combines a dynamic Bayesian network to quantify power source uncertainty and market manipulation risks. It also introduces a confidence-driven dynamic switching mechanism to address the distortion in mode switching decisions caused by feature space mismatch in traditional methods.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] The electricity spot market operation assessment method based on a multi-factor evaluation model includes the following steps: conducting an initial market simulation based on backtesting of historical data to obtain a first state set; using a clustering algorithm to segment the first state set and using the segmentation results as input signals for a pre-defined Bayesian network, constructing an operation assessment decision model through the Bayesian network; making decisions on a pre-defined first action set according to the operation assessment decision model to obtain a simulated alternation sequence; using a pre-defined adversarial network to perform feature mapping on the simulated alternation sequence to obtain projection offset features; and supplementing the state transition probabilities of the pre-defined Bayesian network based on the projection offset features to optimize the operation assessment decision model.

[0011] In a preferred embodiment, the initial market simulation based on backtesting of historical data specifically involves: acquiring historical transaction data of the electricity spot market and constructing a first model; obtaining market clearing price fluctuations based on the first model to obtain an initial market simulation dataset and verifying the simulation data through backtesting; the simulation data verification includes calculating the error rate of the market simulation data and historical transaction data, and determining the validity of the market simulation data based on the error rate.

[0012] In a preferred embodiment, the first state set includes new energy state features, which are obtained as follows: First data on new energy is acquired and discretized to obtain a set of state variables; causal relationship learning is performed on the state variable set to construct a Bayesian network model; based on the Bayesian network model, the state variable set is divided into several fuzzy intervals to obtain a fuzzy rule base; based on the fuzzy rule base, the conditional probabilities between nodes in the Bayesian network model are calculated to obtain an initial conditional probability table; the initial conditional probability table is used as the initial parameters of the Bayesian network model and the parameters are optimized to obtain a probability distribution matrix; based on the probability distribution matrix, the probability value of the electricity price exceeding a preset first price threshold within a preset time window is calculated and used as a new energy state feature.

[0013] In a preferred embodiment, the first state set further includes operational risk characteristics, which are obtained as follows: the electricity spot market is divided into several sub-markets, and first data is obtained, including historical bidding capacity share data, bidding price data, and bidding behavior data for each sub-market; the Herfindahl-Hirschman index is calculated on the historical bidding capacity share data to obtain the market concentration assessment value for each sub-market; the price volatility of each sub-market is calculated based on the bidding price data; the behavior switching rate and behavior fluctuation frequency of each sub-market are calculated based on the bidding behavior data; preset weights are assigned to the behavior switching rate and behavior fluctuation frequency to obtain a behavior stability score; and the behavior stability score and the market concentration assessment value are fused to obtain the operational risk characteristics.

[0014] In a preferred embodiment, the first state set further includes game-theoretic features, which are obtained as follows: a game-theoretic tree model is constructed, and the game-theoretic tree paths are filtered according to the behavioral stability score; the price deviation and the cumulative behavioral cost of the remaining paths are calculated to obtain the path evaluation value; the remaining paths are filtered based on the path evaluation value to obtain a set of high-risk paths; the high-risk path set is subjected to anomaly evaluation to obtain abnormal bidding behavior features; and the abnormal bidding behavior features and the price volatility of the sub-market are fused according to a preset weight to obtain game-theoretic features.

[0015] In a preferred embodiment, the step of using a clustering algorithm to segment the first state set and using the segmentation result as the input signal of a preset Bayesian network to construct an operational evaluation decision model involves: extracting features from the first state set to obtain a first state vector, and calculating the state transition probability of the first state vector using a preset Bayesian network; compressing the first state vector using a dimensionality reduction algorithm and fusing it with the state transition probability to obtain a first feature space; clustering the first feature space using a clustering algorithm to obtain a first region and a second region, constructing a Monte Carlo simulation model library in the first region and an Agent-based market simulation model library in the second region; calculating the spatial distance between the centroids of the first region and the second region, and activating the model library based on the spatial distance to obtain the operational evaluation decision model.

[0016] In a preferred embodiment, the step of making a decision on a preset first action set according to the operational evaluation decision model to obtain a simulation alternation sequence, wherein the first action set includes Monte Carlo simulation and agent-based market simulation, specifically involves: dynamically dividing the preset first action set into states according to the operational evaluation decision model; if the current state is divided into a first region, loading the preset optimal parameter set of the Monte Carlo simulation mode library and starting the simulation; if the current state is divided into a second region, activating the preset behavior rule library of the agent-based mode library and dynamically adjusting the agent interaction frequency; obtaining the switching time point and sequence of the simulation to obtain the simulation alternation sequence.

[0017] In a preferred embodiment, the step of using a preset adversarial network to perform feature mapping on the simulated alternating sequence to obtain projection offset features specifically involves: constructing an adversarial network, which includes a generator and a discriminator; extracting features from the simulated alternating sequence to obtain Monte Carlo feature vectors and agent-based feature vectors; mapping the Monte Carlo feature vectors and agent-based feature vectors to a first feature space according to the generator to obtain Monte Carlo projection vectors and agent-based projection vectors; and performing error judgment according to the discriminator to obtain the projection offset features of the Monte Carlo projection vectors and agent-based projection vectors.

[0018] In a preferred embodiment, the step of correcting the state transition probabilities of a preset Bayesian network based on projection offset features to optimize the operational evaluation decision model specifically involves: obtaining the magnitude and direction of the projection offset features and discretizing the projection offset features to obtain the offset level; obtaining the state transition probabilities of a preset Bayesian network, dynamically adjusting the state transition path according to the offset level, and recalculating the state transition probabilities to obtain the corrected state transition probabilities; and dynamically applying the corrected state transition probabilities to the operational evaluation decision model to obtain the optimized operational evaluation decision model.

[0019] The technical effects and advantages of the electricity spot market operation evaluation method based on the multi-factor evaluation model of this invention are as follows:

[0020] 1. This invention ensures the accuracy of the initial state set by conducting market simulations based on historical data and employs a clustering algorithm to partition the feature space, optimizing the rationality of state classification and reducing the risk of feature space mismatch. Secondly, it utilizes a Bayesian network to construct an operational evaluation and decision-making model. Through systematic analysis of new energy state characteristics, operational risk characteristics, and game theory characteristics, it achieves multi-dimensional feature fusion, enabling the model to accurately perceive market conditions. Furthermore, during the decision-making process, this method alternates between Monte Carlo simulation and agent-based market simulation, and utilizes spatial distance calculation to activate an adaptive mode library, making simulation mode switching more targeted and reducing erroneous switching due to feature mismatch. Further, this method introduces an adversarial network to perform feature mapping on the alternating simulation sequences, using projection offset features to correct state transition probabilities, achieving dynamic realignment of the Bayesian network, thereby optimizing the evaluation and decision-making model and ensuring its high adaptability and accuracy in the face of market environment changes. Compared to traditional methods, this method can not only effectively avoid the mismatch problem in the feature space and improve the accuracy of mode switching, but also perform adaptive optimization during state transition, thereby improving the stability and reliability of the evaluation and providing more accurate and intelligent decision support for the formal operation of the electricity spot market. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the electricity spot market operation assessment method based on a multi-factor assessment model, as described in this invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1, Figure 1 This invention presents a method for evaluating the operation of the electricity spot market based on a multi-factor evaluation model, comprising the following steps:

[0024] S1, perform initial market simulation based on backtesting of historical data and obtain the first state set;

[0025] In this example, initial market simulation is conducted based on backtesting using historical data, specifically as follows:

[0026] Obtain historical transaction data from the electricity spot market and construct a first model;

[0027] The market clearing price fluctuation is obtained based on the first model, and an initial market simulation dataset is obtained. The simulation data is then verified using a backtesting method. The verification of the simulation data includes the calculation of the error rate of the market simulation data and historical transaction data. The validity of the market simulation data is determined by the error rate.

[0028] It's important to note that the initial market simulation based on historical data requires first acquiring historical transaction data from the electricity spot market and constructing a power output probability model and a load demand response model. Power output is influenced by factors such as weather, fuel supply, and equipment failures, and can be modeled using Markov chains or kernel density estimation methods. The load demand response model, however, incorporates user electricity consumption habits and price elasticity, constructed through regression analysis or machine learning methods to simulate electricity consumption trends under different price conditions. The significance of this process lies in building a dynamic supply and demand model based on real market behavior, thereby improving the reliability and adaptability of the simulation. Based on this, the constructed power output probability model and load demand response model are used to simulate market clearing prices, forming the initial market simulation dataset. Market clearing refers to calculating the market equilibrium price during the supply and demand matching process. For example, when wind power output increases, electricity prices may decrease, while the temporary withdrawal of thermal power units may lead to an increase in electricity prices. This market dynamic can be modeled using Monte Carlo simulation or agent-based simulation methods. By simulating market clearing price fluctuations under different scenarios, we can gain a more comprehensive understanding of market operating mechanisms and predict market price risks, thus providing support for market management and investment decisions.

[0029] Furthermore, backtesting is used to verify the error rate between simulated data and historical trading data, and statistical indicators such as mean squared error (MSE) and mean absolute percentage error (MAPE) are used to evaluate the accuracy of the simulation results. For example, if the error between the simulated market price and the historical market price is controlled within 5%, it indicates that the model can reflect market trends well. If the error is large, the modeling method needs to be optimized or the input assumptions adjusted. The significance of error assessment is to ensure the reliability of the simulation results, improve the credibility of the model, and make it more suitable for market analysis and forecasting. Finally, the validity of the initial market simulation is determined based on the error rate. When the backtesting error meets the preset threshold standard (e.g., the error is less than 10%), the simulation model can be considered valid and applied to subsequent market evaluation. If the error exceeds the threshold, the input parameters need to be optimized, the market behavior assumptions adjusted, or the simulation method improved, thereby improving the overall accuracy of the simulation system.

[0030] Furthermore, the uncertainties of new energy sources, operational risks, and game theory characteristics are selected as the first state set for the initial market simulation to assess the stability and operation of the electricity spot market. The uncertainties of new energy sources reflect the output volatility of renewable energy sources such as wind and solar power due to meteorological factors. This uncertainty may trigger sharp fluctuations in market prices, affecting the balance of electricity supply and demand. Therefore, probabilistic modeling is needed to quantify the impact of new energy sources on the market. Operational risks measure the stability of market structure and behavior. Factors such as market concentration, bidding behavior volatility, and price volatility directly relate to the degree of market competition and bidding strategies. Excessive market concentration or abnormal fluctuations in bidding behavior may lead to market manipulation and price distortions, thus affecting market fairness and efficiency. Game theory characteristics focus on the interaction between market participants. By analyzing the bidding behavior, price deviation, and market strategy evolution of market participants through game tree analysis, abnormal bidding behavior, high-risk paths, and the possibility of market manipulation can be identified, thereby improving the accuracy of market supervision. Therefore, comprehensively considering these three characteristics not only fully reflects the core uncertainties and sources of risk in market operation but also provides more accurate decision-making basis for market assessment, thus ensuring the stability, fairness, and sustainability of market operation.

[0031] In this example, the first state set includes the new energy state characteristics, which are obtained in the following way:

[0032] First data on new energy sources are acquired and discretized to obtain a set of state variables;

[0033] Learn causal relationships from the set of state variables and construct a Bayesian network model;

[0034] Based on the Bayesian network model, the set of state variables is divided into several fuzzy intervals to obtain a fuzzy rule base;

[0035] Based on the fuzzy rule base, the conditional probabilities between nodes in the Bayesian network model are calculated to obtain an initial conditional probability table.

[0036] The initial conditional probability table is used as the initial parameters of the Bayesian network model, and parameter optimization is performed to obtain the probability distribution matrix.

[0037] Based on the probability distribution matrix, the probability value of the electricity price exceeding the preset first price threshold within a preset time window is calculated and used as a feature of the new energy status.

[0038] It should be noted that acquiring primary data on renewable energy includes constructing a set of state variables based on key variables such as historical power prediction errors, renewable energy penetration rates, and electricity price volatility. A Bayesian network is then used for causal relationship learning to accurately characterize the uncertainty of renewable energy output and its impact on market prices. By fuzzily partitioning the state variables and establishing a fuzzy rule base, complex market dynamics can be handled more flexibly, avoiding misjudgments caused by data noise or sudden changes, and improving the model's adaptability. Based on the fuzzy rule base, the conditional probabilities of the Bayesian network are derived, enabling the model to effectively predict the market impact of renewable energy even under uncertain environments. Furthermore, parameter optimization yields a probability distribution matrix that better reflects actual market conditions, further enhancing prediction accuracy. In addition, calculating the probability that electricity prices will exceed a threshold within a preset time window using this probability distribution matrix provides a quantitative indicator for assessing the uncertainty risk of renewable energy. This helps market managers formulate more robust dispatch strategies, reduce the impact of renewable energy fluctuations on market prices, and improve the reliability of market simulations, making the assessment process more scientific and data-driven, thus providing strong support for the stable operation of the electricity spot market.

[0039] In this example, the first state set also includes operational risk characteristics, which are obtained as follows:

[0040] The electricity spot market is divided into several sub-markets, and first data is obtained, which includes historical bidding capacity share data, bidding price data and bidding behavior data for each sub-market;

[0041] The Herfindahl-Hirschman Index was calculated based on historical bidding capacity share data to obtain the market concentration assessment value for each sub-market.

[0042] Based on the bid price data, calculate the price volatility of each sub-market;

[0043] Based on bidding behavior data, calculate the behavior switching rate and behavior fluctuation frequency for each sub-market;

[0044] Preset weights are assigned to behavior switching rate and behavior fluctuation frequency to obtain a behavior stability score;

[0045] By fusing behavioral stability scores with market concentration assessment values, operational risk characteristics are obtained.

[0046] In this example, the formula for calculating the price volatility of each sub-market is as follows;

[0047]

[0048] in, For price volatility, For the number of bid data, Let be the price of the i-th bid. This is the preset benchmark price.

[0049] It should be noted that in the electricity spot market, the construction of operational risk characteristics requires a refined analysis in conjunction with the characteristics of sub-markets. Based on the proximity of trading and settlement times and the differences in market functions, the electricity spot market is typically divided into three sub-markets: the day-ahead market, the intraday market, and the real-time market. Each market has significantly different trading mechanisms and risk characteristics, specifically:

[0050] The day-ahead market: Electricity transactions conducted one day in advance, with generation plans and electricity prices determined based on load forecasts for the following 24 hours;

[0051] Intraday Market: Transactions conducted after the day-ahead market closes and before the real-time market opens (usually within the same day) to adjust generation plans;

[0052] Real-time market: Transactions conducted before the actual delivery of electricity (usually 15 minutes to 1 hour) to balance real-time supply and demand discrepancies.

[0053] Furthermore, by dividing the electricity spot market into multiple sub-markets and calculating the Herfindahl-Hirschman index to assess market concentration, it is possible to determine whether market competition is sufficient and avoid operational risks caused by unreasonable market structures. The standard deviation of bid prices measures market price volatility, helping to identify whether there are drastic price fluctuations and thus providing early warnings of anomalies that may affect market stability. Based on this, further analysis of bidding behavior data, by calculating behavior switching rates and the frequency of behavior fluctuations, assesses the adaptability and stability of market participants under different market conditions. Combining this with weighted behavioral stability scores makes the assessment of market bidding behavior more accurate. Finally, integrating market concentration and behavioral stability scores provides a comprehensive characterization of operational risks in market operation, allowing the assessment to not only focus on the rationality of the overall market structure but also delve into changes in the bidding patterns of individual market participants. This method provides a scientific basis for the formal operation of the spot market, ensuring the stability of market mechanisms under different operating conditions, and providing reliable data support for regulators to take timely intervention measures when market anomalies are detected, thereby reducing the systemic risks that may be faced during the formal operation of the market.

[0054] In this example, the first state set of the initial market simulation also includes game features, which are obtained as follows:

[0055] Construct a game tree model and filter the game tree paths based on the behavioral stability score;

[0056] Calculate the price deviation and cumulative behavioral cost of the remaining path to obtain the path evaluation value;

[0057] The remaining paths are filtered based on the path evaluation values ​​to obtain a set of high-risk paths;

[0058] Anomaly assessment is performed on the high-risk path set to obtain characteristics of abnormal bidding behavior;

[0059] By integrating the characteristics of abnormal bidding behavior and the price volatility of the sub-market according to preset weights, the game characteristics are obtained.

[0060] In this example, the formula for calculating price deviation is as follows:

[0061]

[0062] in, Let be the price deviation on the i-th remaining path. Let i be the price of the i-th remaining path at time t. Let t be the preset market benchmark price. Let t be the market price standard. This represents the number of time nodes contained in the remaining path.

[0063] In this example, the formula for calculating the cumulative path cost is as follows:

[0064]

[0065] in, The cumulative path cost for the i-th remaining path is calculated as follows: Let i be the quote for the i-th remaining path at time (t-1). Let be the behavior switching frequency of the i-th remaining path at time t.

[0066] It should be noted that by constructing a game tree model to simulate the bidding decision-making paths of market participants under different market conditions, and filtering the game tree paths based on behavioral stability scores, paths with high randomness or lack of representativeness are eliminated, thereby improving the accuracy and computational efficiency of the analysis. Next, the price deviation and cumulative behavioral cost of the remaining paths are calculated to quantify the rationality of market participants' strategies under specific environments, avoiding reliance solely on historical data and instead conducting dynamic evaluations based on the actual behavior of market participants. Based on these calculations, paths are filtered to extract a set of high-risk paths that may lead to significant market price fluctuations or affect market stability, thereby identifying market participants or trading patterns that may exhibit abnormal bidding behavior. Subsequently, further anomaly assessments are conducted on the high-risk path set to extract characteristics of abnormal bidding behavior, such as extreme changes in bid prices and frequent adjustments to bidding strategies, thereby identifying behavioral patterns that may affect market stability and ensuring that the bidding strategies of market participants remain predictable and rational in formal operation. Finally, the characteristics of abnormal bidding behavior are integrated with the price volatility of the sub-market according to preset weights, so that the game characteristics can fully reflect the correlation between the decision-making patterns of market participants and the overall market volatility, providing a more intuitive reference for market supervision and operation evaluation.

[0067] The advantage of this method lies in its focus not only on the overall market operation but also in its in-depth analysis of the specific decision-making behaviors of market participants, thereby improving the comprehensiveness and accuracy of the assessment. Compared to traditional market operation assessment methods, this method, based on game theory, dynamically tracks the strategic adjustments of market participants under different market environments and combines price volatility and abnormal bidding behavior characteristics to form a comprehensive assessment of market operation risks. During the operation of the electricity spot market, this method can help regulators identify potential market anomalies in advance, enabling more timely and effective regulatory measures. Furthermore, through multi-level path screening and feature fusion, this method avoids the biases that may arise from judging a single factor, making the assessment results more scientific and reliable. Ultimately, this method not only provides objective data support for the formal operation of the market but also serves as an important basis for future market optimization and adjustment, ensuring the stability, safety, and efficiency of the electricity spot market after its formal operation.

[0068] S2, a clustering algorithm is used to divide the first state set into features, and the division result is used as the input signal of the preset Bayesian network. The Bayesian network is used to build an evaluation and decision model.

[0069] In this example, a clustering algorithm is used to segment the first state set, and the segmentation result is used as the input signal for a pre-defined Bayesian network. The Bayesian network is then used to construct and run an evaluation decision model, specifically:

[0070] Feature extraction is performed on the first state set to obtain the first state vector, and the state transition probability of the first state vector is calculated using a preset Bayesian network;

[0071] The first state vector is compressed using a dimensionality reduction algorithm and then fused with the state transition probability to obtain the first feature space.

[0072] Clustering algorithms are used to cluster the first feature space to obtain a first region and a second region. A Monte Carlo simulation model library is constructed in the first region, and an Agent-based market simulation model library is constructed in the second region.

[0073] Calculate the spatial distance between the centroids of the first and second regions, and activate the pattern library based on the spatial distance to obtain the operational evaluation decision model.

[0074] It should be noted that the above method provides a more intelligent and adaptive mode-switching decision-making scheme in the assessment of the operation of the electricity spot market. First, a clustering algorithm is used to segment the first state set, enabling precise differentiation of market states and avoiding unreasonable assessments due to state ambiguity. Second, a Bayesian network is used to calculate state transition probabilities, allowing the evolution of market states to be described at the probabilistic level, improving the model's robustness. Simultaneously, a dimensionality reduction algorithm is introduced to compress the state vectors and integrate the state transition probabilities, reducing computational complexity while retaining the core information of market state changes, ensuring the effectiveness of the assessment. Subsequently, based on the clustering results, different states are divided into two regions: a Monte Carlo simulation model library and an Agent-based market simulation model library. This allows the assessment model to flexibly select the optimal simulation method according to the different characteristics of the market state, improving the accuracy of the assessment. Finally, by calculating the centroid spatial distance between the two regions, the corresponding model library is dynamically activated, achieving intelligent switching of simulation methods. The advantages of this method are: it avoids the limitation of traditional assessment methods that require manual pre-setting of switching conditions, and realizes data-driven mode switching; it effectively solves the problem of decision distortion caused by feature space mismatch and improves the adaptability of assessment; through Bayesian network modeling, the assessment is more in line with the randomness and uncertainty of the market; and it improves the stability and computational efficiency of the assessment model, making the simulation and prediction of the power market more accurate.

[0075] S3, based on the operation evaluation decision model, make a decision on the preset first action set to obtain the simulation alternation sequence;

[0076] In this example, based on the operational evaluation decision model, a decision is made on a preset first action set to obtain a simulation alternation sequence. The first action set includes Monte Carlo simulation and agent-based market simulation, specifically:

[0077] The preset first action set is dynamically divided into states according to the operation evaluation decision model. If the current state is divided into the first region, the preset optimal parameter set of the Monte Carlo simulation mode library is loaded and the simulation is started.

[0078] If the current state is classified into the second region, the preset behavior rule library of the Agent-based pattern library is activated and the interaction frequency of the intelligent agent is dynamically adjusted;

[0079] Obtain the switching time points and sequence of the simulation to obtain the simulation alternation sequence.

[0080] S4. A pre-defined adversarial network is used to perform feature mapping on the simulated alternating sequence to obtain projection offset features;

[0081] In this example, a pre-defined adversarial network is used to perform feature mapping on the simulated alternating sequence to obtain the projected offset features, specifically:

[0082] Construct an adversarial network, which includes a generator and a discriminator;

[0083] Feature extraction is performed on the simulated alternating sequence to obtain Monte Carlo feature vectors and agent-based feature vectors;

[0084] The generator maps the Monte Carlo feature vector and the Agent-based feature vector to the first feature space to obtain the Monte Carlo projection vector and the Agent-based projection vector.

[0085] Error judgment is performed based on the discriminator to obtain the projection offset features of the Monte Carlo projection vector and the Agent-based projection vector.

[0086] It should be noted that the generator in the adversarial network consists of fully connected layers and a multi-head attention mechanism, used to map the Monte Carlo feature vectors and agent-based feature vectors to a shared latent space; the discriminator adopts a convolutional neural network structure to determine the simulation type of the latent space feature source; the specific steps of adversarial training are as follows:

[0087] (1) Input the Monte Carlo simulation data into the generator to generate the Monte Carlo projection vector;

[0088] (2) Input the agent-based simulation data into the generator to generate agent-based projection vectors;

[0089] (3) Calculate the source discrimination error of the two types of projection features using a discriminator;

[0090] (4) Jointly optimize the loss function of the generator and the discriminator, wherein the loss function includes adversarial loss, feature reconstruction loss and spatial orthogonality constraint term.

[0091] It should be noted that feature mapping of the simulation alternation sequence through a pre-defined adversarial network offers significant advantages in intelligent adaptation and feature alignment. First, an adversarial network is constructed by introducing a generator and a discriminator, enabling the features of Monte Carlo simulation and agent-based market simulation to be mapped within a unified feature space, thereby reducing evaluation bias caused by differences in feature space. Second, feature extraction is performed on the simulation alternation sequence to generate Monte Carlo and agent-based feature vectors respectively, and the generator completes the mapping, ensuring consistency in expression between the two simulation modes within the same feature space, improving the comparability and reliability of simulation results. Furthermore, the discriminator performs error assessment on the mapped feature vectors, calculating projection offset features, allowing the system to quantify the deviations between different simulation methods and further optimize the model's state transition process, reducing decision distortion caused by feature mismatch during mode switching. This method not only improves the effectiveness of simulation data, making market evaluation more accurate under different simulation modes, but also enhances the adaptability and robustness of the market evaluation model through dynamic feature correction via adversarial training during decision optimization.

[0092] S5. Based on the projection offset characteristics, the state transition probabilities of the preset Bayesian network are supplemented to optimize the operation evaluation decision model;

[0093] In this example, the state transition probabilities of the preset Bayesian network are supplemented based on the projection offset characteristics to optimize the evaluation decision model, specifically as follows:

[0094] The magnitude and direction of the projection offset feature are obtained, and the projection offset feature is discretized to obtain the offset level;

[0095] Obtain the preset state transition probabilities of the Bayesian network, dynamically adjust the state transition path according to the offset level, and recalculate the state transition probabilities to obtain the corrected state transition probabilities.

[0096] The state transition probabilities after verification are dynamically applied to the operation evaluation decision model to obtain the optimized operation evaluation decision model.

[0097] It should be noted that the amplitude and direction of the projection offset feature are obtained, and the projection offset feature is discretized to obtain the offset level. The specific steps are as follows:

[0098] When the offset exceeds the first threshold, it is marked as a high-risk offset, triggering online fine-tuning of the generator parameters;

[0099] When the offset is between the second threshold and the first threshold, it is marked as a medium-risk offset and an anomaly log is recorded.

[0100] When the offset is below the second threshold, it is marked as normal fluctuation, and the current mapping parameters are maintained.

[0101] Furthermore, the offset level is embedded into the Bayesian network as a new observation node to construct a three-dimensional conditional probability relationship that includes state, action, and offset features. The original state transition probability is directionally adjusted according to the offset level: when the offset is high, the probability weight of unconventional transition paths (such as steady state → transition state) is enhanced; when the offset is low, the original probability distribution is maintained or fine-tuned. Combining the corrected transition probability and real-time risk indicators, an operation strategy with integrated anti-offset capability is generated. The simulation alternation path with low switching cost and significant risk suppression effect is selected first. The execution effect of the simulation decision (such as switching success rate and risk control rate) is fed back to the feature mapping module to dynamically calibrate the projection parameters of the adversarial network and the probability adjustment weights of the Bayesian network, forming an adaptive decision optimization closed loop.

[0102] It should be noted that by calculating the magnitude and direction of the projection offset features and discretizing them to form offset levels, the system can quantify simulation errors and provide targeted correction criteria. Secondly, the state transition paths of the Bayesian network are dynamically adjusted based on the offset levels, thereby reducing misjudgments of states caused by feature space offsets, improving the accuracy of the market assessment model, and recalculating the state transition probabilities after verification. This makes the market state transition process more consistent with real market laws, helping to reduce biases caused by inherent model assumptions and improving the rationality of assessment decisions. Furthermore, the optimized state transition probabilities are dynamically applied to the running assessment decision model, enabling it to continuously self-adjust according to changes in the market environment, enhancing the robustness and intelligence of market assessment. Compared to traditional fixed-parameter Bayesian networks, this method can perceive changes in market characteristics in real time and dynamically optimize state transition rules, making the assessment results more timely and accurate.

[0103] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0104] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0105] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0106] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0107] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0108] In conclusion, 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 within the protection scope of the present invention.

Claims

1. A method for assessing the operation of a power spot market based on a multi-factor assessment model, characterized in that, The method comprises the following steps: initial market simulation based on backtesting of historical data, and obtaining a first state set; adopting a clustering algorithm to divide the features of the first state set, taking the division result as a preset input signal of a Bayesian network, and constructing a running evaluation decision model through the Bayesian network; deciding on a preset first action set according to the running evaluation decision model to obtain a simulation alternation sequence; the first state set comprises new energy state features, and the specific obtaining method is as follows: obtaining new energy first data and performing discretization processing to obtain a state variable set; performing causal relationship learning on the state variable set to construct a Bayesian network model; based on the Bayesian network model, the state variable set is divided into a plurality of fuzzy intervals to obtain a fuzzy rule base; based on the fuzzy rule base, the conditional probability between nodes of the Bayesian network model is calculated to obtain an initial conditional probability table; taking the initial conditional probability table as the initial parameters of the Bayesian network model and performing parameter optimization to obtain a probability distribution matrix; based on the probability distribution matrix, the probability value of the electricity price exceeding a preset first price threshold within a preset time window is calculated and taken as the new energy state feature; the first state set further comprises an operating risk feature, and the specific obtaining method is as follows: dividing the power spot market into a plurality of submarkets and obtaining first data, the first data comprising historical bidding capacity proportion data, bidding price data and bidding behavior data of each submarket; performing H-H index calculation on the historical bidding capacity proportion data to obtain a market concentration degree evaluation value of each submarket; based on the bidding price data, the price volatility rate of each submarket is calculated; based on the bidding behavior data, the behavior switching rate and the behavior fluctuation frequency of each submarket are calculated; a preset weight is given to the behavior switching rate and the behavior fluctuation frequency to obtain a behavior stability score value; the behavior stability score value is fused with the market concentration degree evaluation value to obtain the operating risk feature; the first state set further comprises a game feature, and the specific obtaining method is as follows: constructing a game tree model, and screening the game tree paths according to the behavior stability score value; calculating the price deviation amount and the path cumulative behavior value on the remaining paths to obtain a path evaluation value; based on the path evaluation value, the remaining paths are screened to obtain a high-risk path set; performing abnormal evaluation on the high-risk path set to obtain an abnormal bidding behavior feature; fusing the abnormal bidding behavior feature and the price volatility rate of the submarket according to a preset weight to obtain the game feature; the method of adopting a clustering algorithm to divide the features of the first state set, taking the division result as a preset input signal of a Bayesian network, and constructing a running evaluation decision model through the Bayesian network, specifically comprises the following steps: performing feature extraction on the first state set to obtain a first state vector, and adopting a preset Bayesian network to calculate the state transition probability of the first state vector; adopting a dimension reduction algorithm to compress the first state vector and fuse the features with the state transition probability to obtain a first feature space; The first feature space is clustered by using a clustering algorithm to obtain a first region and a second region, and a Monte Carlo simulation mode library is constructed in the first region and an Agent-based market simulation mode library is constructed in the second region; the spatial distance of the centroids of the first region and the second region is calculated, and the mode libraries are activated according to the spatial distance to obtain a running evaluation decision model; The first action set includes Monte Carlo simulation and Agent-based market simulation, specifically: according to the running evaluation decision model, the preset first action set is dynamically divided into states, if the current state is divided into the first region, the preset optimal parameter set of the Monte Carlo simulation mode library is loaded and the simulation is started; If the current state is divided into the second region, the preset behavior rule library of the Agent-based mode library is activated and the interaction frequency of the agent is dynamically adjusted; the switching time point and sequence of the simulation are obtained to obtain a simulation alternation sequence; The simulation alternation sequence is feature-mapped by using a preset adversarial network to obtain a projection offset feature; The state transition probability of the preset Bayesian network is supplemented according to the projection offset feature, and the running evaluation decision model is optimized. 2.The method of claim 1, wherein, The backtest based on historical data is used to perform initial market simulation, specifically: The historical transaction data of the power spot market is obtained, and a first model is constructed; According to the first model, the market clearing price fluctuation is obtained, the initial market simulation data set is obtained, and the simulation data is verified by the backtest method; the simulation data verification includes calculation of the error rate of the market simulation data and the historical transaction data, and the market simulation data is determined to be effective by the error rate. 3.The method of claim 2, wherein, The simulation alternation sequence is feature-mapped by using a preset adversarial network to obtain a projection offset feature, specifically: An adversarial network is constructed, the adversarial network includes a generator and a discriminator; The simulation alternation sequence is feature-extracted to obtain a Monte Carlo feature vector and an Agent-based feature vector; The Monte Carlo feature vector and the Agent-based feature vector are mapped into a first feature space by the generator to obtain a Monte Carlo projection vector and an Agent-based projection vector; The error is judged by the discriminator to obtain the projection offset feature of the Monte Carlo projection vector and the Agent-based projection vector. 4.The method of claim 3, wherein, The state transition probability of the preset Bayesian network is supplemented according to the projection offset feature, and the running evaluation decision model is optimized, specifically: The amplitude and direction of the projection offset feature are obtained, and the projection offset feature is discretized to obtain an offset level; The state transition probability of the preset Bayesian network is obtained, the state transition path is dynamically adjusted according to the offset level, and the state transition probability is recalculated to obtain the supplemented state transition probability; The supplemented state transition probability is dynamically applied to the running evaluation decision model to obtain an optimized running evaluation decision model.

Images