An electricity spot day-ahead market auxiliary bidding method

Abstract

The application discloses a kind of power spot day-ahead market auxiliary quotation method, it is related to electric power system automation technical field, and this method is through multi-source data fusion preprocessing, collects power grid operation, meteorological, market and carbon market data, is handled by space-time generation confrontation network and natural language processing model, forms multidimensional dataset;Topological modeling is carried out using graph neural network embedding power grid power balance constraint, and spatiotemporal characteristics are extracted in combination with time series neural network and attention mechanism;A double-layer optimization model is constructed to realize futures and spot market collaborative decision-making, and a carbon-electricity joint optimization and dynamic position adjustment mechanism is simultaneously integrated.The application solves the problems of traditional quotation method such as data isolation, inaccurate prediction and insufficient risk control, improves the accuracy of quotation and market competitiveness, provides an auxiliary quotation scheme that takes into account both revenue and risk for power generation enterprises, and is suitable for quotation strategy optimization in the power spot day-ahead market.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system automation, in particular to a power spot market auxiliary bidding method. BACKGROUND

[0002] Under the background of deepening power system reform, the construction of power spot market is gradually promoted, and the traditional power market bidding mode faces many challenges. In the prior art, the power market bidding method mainly relies on marginal cost pricing, historical data statistics and the experience of traders, and there are problems such as data isolation, insufficient prediction accuracy and weak risk control.

[0003] Specifically, on the one hand, the relevant data of the power market is scattered in the power grid operation monitoring system, the weather forecast system, the market trading platform and other independent systems, forming a data island, which is difficult to realize effective integration and comprehensive utilization, resulting in a lack of comprehensive data support for bidding decision-making. On the other hand, traditional prediction models, such as time series models, are difficult to accurately capture the nonlinear characteristics of power market price fluctuations and the influence of complex factors such as new energy output fluctuations, resulting in low power price prediction accuracy.

[0004] Therefore, the present application is proposed. SUMMARY

[0005] The purpose of the present application is to provide a power spot market auxiliary bidding method to solve the problems of data integration difficulty, low prediction accuracy and insufficient risk control mentioned in the background.

[0006] To solve the above technical problems, the present application provides a power spot market auxiliary bidding method, comprising the following steps:

[0007] S1 multi-source data fusion preprocessing: collect power grid operation data, weather data, market data and carbon market data, fill in missing values through a spatio-temporal generative adversarial network, synchronize the weather data at a 15-minute cycle, use a natural language processing model to extract strategy keywords from market declaration text, and form a spatio-temporally aligned multi-dimensional data set;

[0008] S2 power grid topology constraint modeling: construct a power grid topology graph containing node attributes and edge weights, embed power balance constraints through a graph neural network, and output node price prediction values that satisfy physical laws;

[0009] S3 spatio-temporal feature engineering and decision optimization: use a time series neural network combined with an attention mechanism to capture the temporal fluctuations of electricity prices and the spatial correlation of nodes, consider the transmission delay effect to generate a multi-dimensional feature matrix, calculate the optimal futures contract ratio and unit combination plan based on a double-layer optimization model, and output the final bidding strategy through iteration;

[0010] S4 Carbon-Electricity Joint Optimization: Real-time monitoring of carbon price, when exceeding the preset threshold, adjusting the bidding amount of thermal power units in high and low carbon emission periods, and synchronously generating carbon quota trading suggestions; through multi-source data deep fusion, power grid topology constraint modeling and cross-market collaborative optimization, solving the problems of data isolation, inaccurate prediction and insufficient risk control in traditional bidding methods, effectively improving the accuracy of bidding and market competitiveness, and realizing the maximization of power generation enterprise revenue and controllable risk.

[0011] Further, in S1, the collected power grid operation data includes SCADA real-time bus load, transmission line capacity and unit operation state, market data includes historical day-ahead electricity price and competitor declaration text, and carbon market data includes regional carbon quota price and enterprise carbon asset holding; ensuring the comprehensiveness and diversity of data sources, providing a basis for subsequent accurate analysis.

[0012] Further, in S1, the spatio-temporal generative adversarial network generates missing power grid operation data based on adjacent site historical data, and the natural language processing model uses the BERT model to perform entity recognition on the declaration text to extract strategy keywords such as "conservative bidding" and "aggressive quantity grabbing"; improve data integrity and availability, enhance insight into competitor strategies, and optimize bidding strategy formulation.

[0013] Further, in S2, the node attributes of the power grid topology graph include bus load prediction value and unit type, and the edge weight is set as the transmission line capacity limit; through power balance constraint, the model output conforms to the physical law of power grid, and the electricity price prediction result is obtained; make the electricity price prediction more in line with the actual power grid operation situation, improve the prediction accuracy, and reduce the bidding deviation.

[0014] Further, in S3, the futures layer of the double-layer optimization model calculates the optimal futures contract proportion based on option pricing, and the spot layer generates unit combination plan through mixed integer programming, and the two-layer model iterates at least 3 rounds to realize the collaborative optimization of risk and return; realize the linkage of spot and futures markets, effectively hedge price risk, and improve the comprehensive income of power generation enterprises.

[0015] Further, in S3, the time series neural network adopts the Transformer-LSTM architecture to capture the time series characteristics of electricity price, the attention mechanism calculates the node spatial correlation weight and introduces a 1-2 hour lag time window; accurately capture the spatio-temporal variation law of electricity price, and provide a more reliable basis for bidding strategy.

[0016] Further, in S4, the carbon price preset threshold is 100 yuan / ton, by adjusting the bidding amount of thermal power units in high carbon emission periods such as start-up and shutdown stages, and preferentially declaring low carbon emission periods such as full-load operation; realize carbon-electricity collaborative optimization, reduce carbon trading cost, and improve the economic benefit of enterprises in the carbon market environment.

[0017] Further, S5 dynamic position adjustment is further included: real-time monitoring of spot price fluctuations is realized by using deep reinforcement learning, position adjustment actions are triggered when the price deviation exceeds 10%, and the futures position ratio is dynamically adjusted by 5%-20%; the strategy is adjusted in time when the market price fluctuates, potential losses are reduced, and the flexibility and adaptability of the bidding strategy are enhanced.

[0018] Compared with the prior art, the beneficial effects of the present application are:

[0019] 1. Through multi-source data deep fusion and intelligent modeling, a full-chain technology system covering data processing, electricity price prediction and decision optimization is constructed. The missing values of the power grid data are filled by the space-time generative adversarial network, and the multi-source data are synchronized, the market strategy keywords are extracted by combining the BERT model, the multi-dimensional data set is formed, the data island problem of the traditional method is solved, and a reliable data foundation is laid for bidding decision. At the same time, the power grid power balance constraint is embedded by means of the graph neural network, the electricity price space-time characteristics are captured by combining the Transformer-LSTM and the attention mechanism, and the internal correlation between the power grid operation and the price fluctuation is accurately reflected.

[0020] 2. In terms of decision optimization and risk control, the present application innovatively constructs a double-layer optimization model to realize the coordination of spot and futures markets, and dynamically adjusts the position by means of deep reinforcement learning, thereby synchronously realizing the bidding target of "high yield and low fluctuation". In addition, the carbon-electricity joint optimization module automatically adjusts the thermal power bidding period when the carbon price exceeds the threshold, and combines the carbon quota transaction suggestion, thereby effectively responding to the new market demand under the "double carbon" target.

[0021] 3. The method realizes cross-platform migration through Docker containerized deployment, meets the timeliness requirement of day-ahead market bidding, significantly improves the bidding accuracy and market competitiveness of power generation enterprises, and provides a supplementary bidding solution for power market main bodies, which takes into account the economy and controllable risk, and has important practical value for promoting the coordinated development of the electricity spot market and the carbon market. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 It is a principle block diagram of a power spot day-ahead market supplementary bidding method. DETAILED DESCRIPTION

[0023] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0024] Please refer to Figure 1The application provides a technical solution: a power spot day-ahead market auxiliary bidding method. The power spot day-ahead market auxiliary bidding method given in the scheme uses multi-source data fusion, graph neural network modeling, space-time feature engineering, and double-layer optimization decision-making technologies to realize accurate bidding, control risks, and improve the income of market subjects.

[0025] Step 1: Multi-source data collection and pre-processing

[0026] 1. Collect multi-dimensional raw data, mainly including:

[0027] Power grid operation data: Obtain real-time bus load data (15 minutes per record) from SCADA system, transmission line capacity (like IEEE118 node system parameters), and also include unit operation state (output upper limit, start-stop cost).

[0028] Weather data: Wind speed and light intensity data from numerical weather prediction, 1 hour per record, resolution is 1km x 1km.

[0029] Market data: Historical day-ahead electricity price for the past 3 years (15 minutes per record), and competitor declaration text (usually PDF format of bidding strategy description).

[0030] Carbon market data: Regional carbon quota price (daily data), and the number of carbon assets owned by the enterprise.

[0031] 2. The necessity of data cleaning and synchronization:

[0032] If SCADA data is missing, use space-time generative adversarial network (ST-GAN) to fill it. For example, if wind power output data is missing, generate appropriate filling values based on the wind speed and historical output curve of the adjacent 3 sites.

[0033] Interpolate weather data from 1 hour per record to 15 minutes per record to make it match the time of electricity price data.

[0034] Use BERT model to do entity recognition on market declaration text to find out the keywords in competitor bidding strategy, such as "conservative bidding" and "aggressive volume grabbing".

[0035] For example, take a power generation enterprise in Guangdong as an example, collect data on XX month XX day in 20XX, SCADA system records 288 bus load data of Pearl River Basin (15 minutes per record), and the result has 12 missing due to communication failure; NWP data shows that the average wind speed is 6m / s and the light intensity is 400W / m 2The competitor A's declaration text mentions "priority protection of basic power", and the competitor B mentions "pursuit of high price period". Through ST-GAN, 12 pieces of missing output data are generated, and after interpolation processing, 288 pieces of full-period aligned data set are formed.

[0036] It should be noted here that the data of the electricity market has the characteristic of strong space-time heterogeneity (the grid data is synchronized every 15 minutes, and the meteorological data is synchronized every 1 hour), and the traditional interpolation method cannot grasp the nonlinear relationship in the data. ST-GAN can improve the data integrity rate by learning the data distribution through the generation of an adversarial mechanism, which provides accurate input data for the subsequent modeling.

[0037] Step 2: Build a grid topology graph neural network model (GNN)

[0038] 1. First, build a grid topology graph:

[0039] The bus of the IEEE118 node system is regarded as the node of the graph (a total of 118 nodes), and the transmission line is the edge with weight, and the weight is the line transmission capacity limit value, unit is MW.

[0040] Add attributes to the node: bus load prediction value (data processed in step 1), unit type (thermal power / wind power / solar power, different types of units have different marginal costs).

[0041] 2. In order to make the model more consistent with the actual grid operation, the direct current flow equation constraint is added to the message passing mechanism of GNN, so that the node price output by the model must satisfy the power balance condition:

[0042] ;

[0043] Here is the line susceptance, is the node phase angle, is the unit injection power, is the load.

[0044] 3. Train the GNN model:

[0045] The input is the time-space aligned data generated in step 1 (node attributes plus edge weights).

[0046] The output is the 24-hour future price prediction value of each node every 15 minutes (a total of 96 time points).

[0047] The loss function uses mean square error (MSE) plus the error of the flow equation constraint (weight 0.3).

[0048] For example, in the case of Guangdong-Guangxi cross-provincial power transmission, there are 50 key nodes in the power grid topology, of which 30 are in Guangdong and 20 are in Guangxi. The edge weight is determined according to the actual transmission line capacity, such as the Guangdong-Guangxi DC line capacity of 500 MW. During training, when the model predicts that the electricity price of Guangdong node 1 is 0.6 yuan / kWh, it is verified through the power flow equation that the power injection of Guangxi node 20 does not satisfy the balance condition, and the predicted value is automatically adjusted to 0.62 yuan / kWh, so that the power deviation rate is less than 5%. After 100 rounds of training, the electricity price prediction error of this region is significantly lower than that of the traditional method.

[0049] It should be noted that the existing public document (CN119358180A) only uses GNN for power flow calculation and does not combine it with electricity price prediction. This step of the present scheme embeds power grid operation rules into the model through a physical constraint layer, solving the problem of missing spatial features, such as the impact of transmission congestion on node electricity prices. Compared with GNN models without constraints, the prediction accuracy is improved.

[0050] Step 3: Spatio-temporal attention feature engineering

[0051] 1. Modeling in the time dimension:

[0052] Use the Transformer encoder to process the electricity price sequence of 96 time points to capture long-term dependencies, such as the price difference between weekdays and weekends.

[0053] Combine bidirectional LSTM to handle short-term price mutations and extract 15-minute-level price volatility features, such as the price spikes caused by the midday load peak.

[0054] 2. The necessity of modeling in the spatial dimension:

[0055] Calculate the spatial correlation weight between nodes through the attention mechanism to find key influencing nodes. For example, calculate the electricity price influence weight of Guangdong node 1 on Guangxi node 20. If it exceeds 0.1, it is determined that the two nodes are strongly correlated.

[0056] Introduce a lag window with a maximum lag of 4 time points, i.e., 1 hour, to capture the delay effect of cross-provincial power transmission, such as the impact of Guangxi hydropower output changes on Guangdong electricity prices.

[0057] 3. When fusing features, concatenate the GNN output node electricity price prediction value (spatial feature) and the spatio-temporal features extracted by Transformer-LSTM to form a multi-dimensional feature matrix containing 118 nodes x 96 time points.

[0058] For example: on a certain day in summer 20XX, GNN predicts that the hydropower output of Guangxi node 20 will peak at 14:00, and through the spatio-temporal attention mechanism calculation, it is found that this output change will affect the electricity price of Guangdong node 1 with a lag of 2 time points (30 minutes). The model automatically adjusts the predicted price of Guangdong node 1 at 14:30 from 0.58 yuan / kWh to 0.63 yuan / kWh, with an error of only 1.2% from the actual market clearing price. Without considering the lag effect, the traditional model has an error of 8.5%.

[0059] Here we can see that the spatio-temporal coupling of electricity prices is very strong, and cross-regional power transmission requires physical transmission time. Traditional time series models (such as ARIMA) only process single-node data, ignoring spatial correlations. This step of the program dynamically allocates node influence weights through the spatio-temporal attention mechanism, reducing the prediction error of the Guangdong-Guangxi cross-provincial market by 15%, which verifies the necessity of spatial feature modeling.

[0060] Step 4: Multi-modal data fusion and decision graph generation

[0061] 1. When processing unstructured data, use the BERT model to perform sentiment analysis on competitors' declaration texts to generate their strategy tendency vectors (dimension 100). For example, "aggressive volume grabbing", the value of the "price sensitivity" dimension in the corresponding vector will be greater than 0.8.

[0062] 2. Construct a "volume-price benchmark graph":

[0063] Take unit output (x-axis, unit: MW) and electricity price (y-axis, unit: yuan / kWh) as coordinates, and mark historical bid data as colored scatter points (green represents profit points, red represents loss points).

[0064] Use DBSCAN algorithm to divide the bidding strategy area into conservative area (low price-low risk), aggressive area (high price-high risk), and balanced area (medium price-medium yield).

[0065] 3. When fusing decision features, concatenate GNN spatio-temporal features (96x118), BERT strategy vectors (1x100), and volume-price area labels (1x3) as input features, and input them into a multi-layer perceptron (MLP) to generate an initial bidding scheme.

[0066] Specifically, there is a thermal power unit (output upper limit 500 MW), and the electricity price prediction value at 14:00 the next day obtained in step 3 is 0.65 yuan / kWh, which is at the boundary of the aggressive zone. At the same time, the BERT analysis shows that competitor A tends to be conservative in pricing (strategy vector "price sensitivity" 0.3) during this period, and competitor B tends to be aggressive (0.7). The model automatically adjusts the price of this period to 0.63 yuan / kWh (slightly lower than the predicted value), falling into the equilibrium zone, which not only avoids direct competition with B, but also has more advantages than A's conservative pricing. Finally, the winning electricity quantity in this period is 300 MW, which is 50 MW more than the traditional experience pricing.

[0067] Of course, the traditional method only uses structured data (such as historical electricity prices) and does not make good use of the semantic information of market declaration text. This step of the technical solution extracts the strategy of competitors through BERT, and then makes visual decision-making combined with the quantity-price benchmark chart, which improves the market adaptability of the pricing scheme and solves the problem of strategy blindness caused by "data silos".

[0068] Step 5: Solve the double-layer optimization decision-making model

[0069] Upper layer: futures market risk hedging layer

[0070] 1. When calculating the optimal futures contract ratio, input the following parameters into the Black-Scholes option pricing model:

[0071] Spot price volatility σ = 20% (calculated from historical data in step 3).

[0072] Risk-free interest rate r = 3% (yield).

[0073] Futures contract expiration time T = 1 day.

[0074] The objective function is to maximize the expected return minus 1.5 times the conditional value at risk (CVaR), balancing the return and risk:

[0075] ;

[0076] Here is the futures contract quantity, is the futures price, is the initial price of the spot;

[0077] Lower layer: unit combination layer in the spot market

[0078] 2. When generating the unit combination plan, input the initial pricing scheme in step 4, unit parameters (start-up and shutdown cost 100,000 yuan / time, coal consumption curve), and grid constraints (node capacity limit). Use a mixed integer programming (MIP) model, and the objective function is to maximize the day-ahead revenue:

[0079] ;

[0080] wherein is period output, is the offer price, is the generation cost, is the start-stop state);

[0081] 3. Double-layer iterative optimization, the optimal futures ratio q calculated by the upper layer is fed back to the lower layer to adjust the risk tolerance of the spot offer; the unit combination result generated by the lower layer (like start-stop plan) revises the CVaR calculation of the upper layer, and after 3 rounds of iteration, it can converge.

[0082] For example: A certain electricity selling company in Zhejiang used this step in XX month of 20XX: the upper layer calculated the optimal futures contract ratio as 40% of the spot transaction volume (corresponding to a futures price of 0.55 yuan / kWh and a spot forecast average price of 0.6 yuan / kWh); the lower layer's MIP model adjusted the offer strategy according to this ratio, generating more power in high-price periods (forecast > 0.65 yuan / kWh) and stopping in low-price periods (< 0.5 yuan / kWh). Finally, the company's day-ahead winning power ratio increased from 65% to 82%, and the degree of electricity yield increased by 0.03 yuan compared to the pure spot strategy, and the yield volatility decreased from 25% to 12%.

[0083] Obviously: the existing public document (CN202411554063) only optimizes the single spot market and does not consider futures hedging. The steps of this program realize cross-market coordination through a double-layer model, and through Monte Carlo simulation verification, in the scenario of price fluctuations, such as ±30% fluctuation, compared with single-layer optimization, the comprehensive yield can be improved and the risk can be reduced, which proves the creative value of cross-market coordination.

[0084] Step 6: Dynamic adjustment and execution of output

[0085] 1. Real-time market monitoring is important, through Kafka real-time receiving of power grid SCADA data (delay less than 500ms), updating the electricity price forecast value every 15 minutes (using the model of step 2-step 4 for rolling prediction).

[0086] 2. Use deep Q network (DQN) to dynamically adjust spot and futures positions, the state space includes current position, real-time electricity price deviation (current price / forecast price-1), carbon quota price fluctuation; the action space has 5 actions such as increasing 10% spot position and reducing 10% futures position, the reward function is the immediate income plus 0.9 times the future income expectation.

[0087] 3. Carbon-electricity joint optimization: If the real-time carbon price is >100 yuan / ton, and the unit is a thermal power plant, automatically reduce the bidding amount during high-carbon emission periods (such as start-stop stages) and preferentially bid during low-carbon periods (full-load operation). The final output of the final bidding scheme includes 24-hour unit output bids every 15 minutes, futures position ratio, and carbon quota trading recommendations.

[0088] For example: On XX / XX / 20XX at 10:00, the Guangdong spot price is updated to 0.7 yuan / kWh, which is higher than the predicted value of 0.65 yuan / kWh in step 5. The DQN model triggers the "reduce futures position" action, reducing the futures ratio from 40% to 30% and releasing 10% of the electricity for spot trading. At the same time, because the current carbon price is 120 yuan / ton, the model suggests that this thermal power plant increase the bid by 0.02 yuan / kWh during the 14:00-16:00 full-load operation period (utilizing the low-carbon advantage to obtain green electricity premium), and the final winning price for this period is 0.72 yuan / kWh, which is 120,000 yuan more than the original scheme, and the net carbon quota revenue also increases by 50,000 yuan.

[0089] Of course, market prices fluctuate and carbon policies change too quickly, and traditional static bidding cannot cope. This step of the scheme adjusts the strategy in real time through DQN, and when the city's electricity spot market is piloted, it encounters extreme price fluctuations (>0.8 yuan / kWh), with a loss reduction of 40%. Combined with carbon-electricity joint optimization, the profit volatility of comprehensive energy service providers can be reduced by 30-40%, which verifies the practical usefulness of dynamic adjustment of the scheme.

[0090] In summary: In terms of data processing and modeling, the combination of spatio-temporal generative adversarial networks (ST-GAN) and BERT natural language processing models solves the problem of strong heterogeneity and difficulty in integrating multi-source data in the electricity market, forming a multi-dimensional data set aligned in time and space, and providing comprehensive data support for bidding decisions. At the same time, for the first time, graph neural networks (GNN) are combined with power grid power balance constraints to build a topological model that takes into account the physical laws of the grid, resulting in lower prediction error than traditional methods, breaking the bottleneck of traditional prediction models that cannot accurately reflect the actual operating state of the grid.

[0091] In terms of decision optimization mechanism, a two-level optimization model is innovatively constructed, including a futures risk hedging layer and a spot unit combination layer, achieving coordinated decision-making between the electricity spot market and the futures market. Through the combination of option pricing and mixed integer programming, the proportion of winning electricity for power generation enterprises is increased, and the revenue volatility is reduced, effectively solving the problem of insufficient risk control in traditional bidding methods. In addition, dynamic adjustment of the position is introduced through deep reinforcement learning, which triggers strategy modification when market price fluctuations exceed 10%, reducing losses in extreme scenarios and significantly enhancing the market adaptability of the bidding strategy.

[0092] In response to the new demand of "double carbon" target, a carbon-electricity combined optimization module is proposed. When the carbon price exceeds 100 yuan / ton, the bidding quantity of high and low carbon emission period of thermal power unit is automatically adjusted. Combined with carbon quota transaction, the comprehensive income of electricity is increased by 3%-5%, realizing the collaborative optimization of electricity market bidding and carbon market, and providing a new solution for the bidding decision of power generation enterprises under the "double carbon" target.

Claims

1. A method for auxiliary quotation in the electricity spot day-ahead market, characterized by: The following steps are involved: S1 Multi-source Data Fusion Preprocessing: This collects grid operation data, meteorological data, market data, and carbon market data, fills in missing data values ​​using a spatiotemporal generative adversarial network, synchronizes meteorological data in 15-minute cycles, and uses a natural language processing model to extract strategic keywords from market application texts to form a spatiotemporally aligned multidimensional dataset. S2 Grid Topology Constraint Modeling: Construct a grid topology graph with node attributes and edge weights, embed grid power balance constraints through a graph neural network, and output node electricity price forecasts that meet physical laws. Node attributes in the grid topology graph include bus load forecasts and unit types, and edge weights are set to transmission line capacity limits. Power balance constraints are used to force the model to output electricity price forecasts that comply with the physical laws of the grid. S3 spatiotemporal feature engineering and decision optimization: This approach uses a time-series neural network combined with an attention mechanism to capture the temporal fluctuations in electricity prices and the spatial correlation between nodes. It also considers the effects of transmission delays to generate a multidimensional feature matrix. Based on a two-layer optimization model, it calculates the optimal futures contract ratio and unit combination plan, and iteratively outputs the final quotation strategy. S4 Carbon-Electricity Joint Optimization: Real-time monitoring of carbon prices. When the preset threshold is exceeded, the quoted amount of the thermal power unit during high and low carbon emission periods is adjusted, and carbon quota trading recommendations are generated simultaneously.

2. The auxiliary quotation method for the electricity spot day-ahead market according to claim 1, characterized in that: In S1, the collected grid operation data include SCADA real-time bus load, transmission line capacity and unit operation status, market data include historical day-ahead electricity prices and competitor declaration texts, and carbon market data include regional carbon quota prices and corporate carbon asset holdings.

3. The auxiliary quotation method for the electricity spot day-ahead market according to claim 1, characterized in that: In S1, the spatiotemporal generative adversarial network generates missing power grid operation data based on the historical data of adjacent sites, and the natural language processing model uses the BERT model to perform entity recognition on the declaration text and extract the keywords of the "conservative quotation" and "aggressive volume grabbing" strategies.

4. The auxiliary quotation method for the electricity spot day-ahead market according to claim 1, characterized in that: In S3, the futures layer of the two-layer optimization model calculates the optimal futures contract ratio based on option pricing, and the spot layer generates a unit commitment plan through mixed integer programming.

5. The auxiliary quotation method for the electricity spot day-ahead market according to claim 1, characterized in that: In S3, the time series neural network uses the Transformer-LSTM architecture to capture the time series characteristics of electricity prices, and the attention mechanism calculates the spatial association weights of nodes and introduces a 1-2 hour lag time window.

6. The auxiliary quotation method for the electricity spot day-ahead market according to claim 1, characterized in that: In S4, the preset threshold of carbon price is RMB 100 / ton. By adjusting the quotation amount of high carbon emission period during the start-up and shutdown phases of thermal power units, priority is given to declaring low-carbon period during full-load operation.

7. The auxiliary quotation method for the electricity spot day-ahead market according to claim 1, characterized in that: It also includes S5 dynamic position adjustment: using deep reinforcement learning to monitor spot price fluctuations in real time, triggering position adjustment actions when the price deviation exceeds 10%.

Images