Thermal power spot market quotation optimization method based on multi-dimensional data fusion

Abstract

The application discloses a thermal power spot market quotation optimization method based on multi-dimensional data fusion, and relates to the technical field of power spot market transaction; the method comprises the following steps: collecting multi-source data and performing pretreatment; constructing a dynamic marginal cost model of a unit full load interval; extracting a unit operation state feature; generating a power price probability scenario; optimizing a segmented quotation strategy dependent on a state, solving an optimal segmented quotation curve; adopting a two-stage rolling optimization and correction architecture to generate and update the quotation strategy in a day-ahead market and an intra-day-real-time rolling correction stage; and finally executing the quotation strategy and updating model parameters based on a feedback signal. Through the fusion of unit operation data, market clearing data and environmental data, the combination of dynamic cost modeling and power price uncertainty quantification, the end-to-end closed-loop optimization of the thermal power spot market quotation is realized, the expected income of the thermal power unit in the spot market can be effectively improved, and the operation risk can be controlled.

Description

Technical Field

[0001] This invention relates to the field of electricity spot market trading technology, specifically to a method for optimizing thermal power spot market quotations based on multi-dimensional data fusion. Background Technology

[0002] Currently, there are numerous research findings both domestically and internationally regarding electricity spot market price forecasting and pricing strategies; existing technologies mainly focus on categories such as electricity price forecasting and trading decisions.

[0003] Among them, the electricity price prediction category includes, for example, Chinese patent with publication number CN120355445A, entitled "Method for Predicting Spot Electricity Prices Based on Machine Learning and Adaptive Mechanisms"; which predicts segmented electricity prices by constructing a composite feature factor library; and patent with publication number CN119294606A, entitled "Multi-Dimensional Load Electricity Price Prediction Method", which uses the Informer network for load electricity price prediction.

[0004] Transaction decision optimization methods, such as the Chinese patent with publication number CN117910644A, entitled "Transaction Decision Optimization Method Based on Data Value Estimation", make pricing decisions by constructing a dual-settlement market transaction decision optimization model.

[0005] In addition, some companies have proposed plant-level coal-fired power unit load optimization allocation models to determine the optimal load allocation scheme with the goal of maximizing power generation profits.

[0006] However, existing methods mostly focus on improving the accuracy of electricity price forecasts, but the predicted electricity prices cannot be directly converted into executable pricing strategies. The core question that thermal power units need to answer in actual operation is "how much price and how much quantity to quote", not just "what will the electricity price be". There are complex information updates and decision-making linkages between the day-ahead, intraday, and real-time markets in the spot market. Existing methods mostly optimize for a single time scale and lack the ability to make collaborative decisions across time scales. Summary of the Invention

[0007] To address this issue, the present invention provides a method for optimizing spot market quotations for thermal power based on multi-dimensional data fusion, thereby resolving the problems in the prior art.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] The method for optimizing spot market quotations for thermal power based on multi-dimensional data fusion includes the following steps:

[0010] Step 1: Collect multi-source data related to the operation of thermal power units and market transactions, and clean, normalize and time-align the multi-source data to form a standardized time-series dataset;

[0011] Step 2: Based on the unit distributed control system operation data and fuel data in the multi-source data, construct a dynamic marginal cost model that reflects the nonlinear cost characteristics of thermal power units in the full load range;

[0012] The dynamic marginal cost model divides the unit output range into low load zone, economic load zone and high load zone, and quantifies the marginal cost composition of each load zone.

[0013] Step 3: Based on the operating data of the unit's distributed control system and the dynamic marginal cost model, extract the feature vector of the unit's current operating state;

[0014] Step 4: Based on the historical electricity price data, new energy output forecasts, and system load forecasts from the multi-source data, construct an electricity price probability prediction model that considers the uncertainty of new energy output, and generate multiple electricity price scenarios and their corresponding probabilities.

[0015] Step 5: Input the feature vector and the electricity price probability scenario into the pricing optimization model to solve for the optimal segmented pricing curve that maximizes expected revenue;

[0016] The pricing optimization model parameterizes the pricing curve into multiple segments, each segment being defined by the upper limit of the output range and the pricing price, with the optimization objective being to maximize expected revenue, while also setting constraints.

[0017] Step 6: Adopt a two-stage rolling optimization architecture, specifically as follows:

[0018] During the daytime market phase, execute steps 4 and 5 to generate the next day's segmented price curve and the next day's production plan;

[0019] During the intraday-real-time rolling correction phase, step 5 or steps 3 and 5 are re-executed based on the trigger conditions to update the segmented pricing curve and output plan.

[0020] Step 7: Submit the segmented price curve to the power trading platform, collect actual data after market clearing, calculate the prediction deviation and revenue deviation, and use the deviation data of prediction deviation and revenue deviation as feedback signals to update the parameters of the dynamic marginal cost model and the electricity price probability prediction model.

[0021] Furthermore: the multi-source data in step 1 includes: operating data of the distributed control system of thermal power units, electricity market clearing data, and environmental and fuel data;

[0022] The operating data of the distributed control system of the thermal power unit includes the unit's real-time output, main steam pressure, main steam temperature, reheat steam temperature, turbine vibration, boiler furnace pressure, coal feed rate, feedwater flow rate, flue gas temperature at the denitrification system inlet, and nitrogen oxide concentration.

[0023] The electricity market clearing data includes day-ahead clearing prices, real-time clearing prices, system load forecasts, renewable energy output forecasts, tie-line plans, and reserve capacity requirements.

[0024] The environmental and fuel data include ambient temperature, coal calorific value, coal sulfur content, carbon quota trading price, and fuel oil price.

[0025] Furthermore: In the dynamic marginal cost model of step 2; the unit output at time t is The overall marginal cost at that time Represented as:

[0026] ;

[0027] in, Benchmark fuel cost; This is the allocated value for the unit's start-up and shutdown costs; These are start / stop state variables; This is the ramp-up cost coefficient; Costs related to fatigue life loss.

[0028] Furthermore: the fatigue life loss cost The following method is used for quantification:

[0029] ;

[0030] in, Costs for replacing or repairing critical components; The number of fatigue cycles allowed for the component; This refers to the temperature change caused by this peak-shaving process; For reference temperature change range; This is the load factor correction coefficient.

[0031] Furthermore, the feature vector of the current operating status of the unit in step 3 includes: the current output value and its load range, the output change rate between the current time and the previous time, the cumulative fatigue life loss value of key components, the temperature deviation value between the inlet flue gas temperature of the denitrification system and the catalyst activity window, and the boiler combustion stability index.

[0032] Furthermore, step 4, which generates the electricity price probability scenario, specifically includes:

[0033] Statistical analysis was conducted on the prediction errors of new energy power output and load, and an error probability distribution model was established.

[0034] Monte Carlo sampling is performed based on the aforementioned error probability distribution model to generate multiple sets of error scenarios for new energy output and load.

[0035] The sampled error scenarios are input into the electricity price response model and mapped to the corresponding electricity price scenarios, forming the electricity price probability density function.

[0036] Furthermore: In step 5, the price quote curve is divided into K segments, each segment consisting of an upper limit of the output range. and quoted price Two parameters are defined. ;

[0037] The output range satisfies the following relationship:

[0038] ;

[0039] The price curve satisfies the monotonically increasing constraint:

[0040] ;

[0041] The optimization objective function is expressed as:

[0042] ;

[0043] in, Let be the probability of the s-th electricity price scenario; Let be the price quoted in the k-th price segment during time period t; This refers to the cleared electricity volume of the k-th price segment in time period t under scenario s; It is a dynamic marginal cost function; and These are start / stop state variables.

[0044] Furthermore, the constraints in step 5 include upper and lower limits of output, ramp rate constraints, non-economic zone avoidance constraints, minimum continuous operating time constraints, minimum continuous downtime constraints, and price monotonicity constraints.

[0045] Furthermore: the intraday-real-time rolling correction phase in step 6 includes an intraday rolling correction phase and a real-time market phase;

[0046] The intraday rolling correction phase is triggered when the forecast deviation of new energy output or system load exceeds the preset threshold, or when the unit experiences an unplanned shutdown or output limitation, and step 5 is re-executed with the next 4-6 hours as the optimization window.

[0047] Before the real-time market clearing, the real-time market stage performs partial adjustments in steps 3 and 5 based on ultra-short-term forecast data and the actual operating status of the units, and outputs the real-time market bid price and bid output.

[0048] Furthermore: the prediction deviation and revenue deviation in step 7 include the deviation between the actual clearing price and the predicted price, the deviation between the unit's winning bid volume and the declared volume, and the deviation between the actual operating cost and the cost calculated by the dynamic marginal cost model; the deviation data is used to iteratively optimize the parameters of the dynamic marginal cost model and the electricity price probability prediction model online.

[0049] This invention has the following advantages: By establishing a dynamic marginal cost model for the entire load range, this invention quantifies fuel costs, ramp-up losses, and fatigue life losses, thereby improving the accuracy of cost accounting; through the generation of electricity price probability scenarios and a segmented pricing strategy based on state dependence, this invention incorporates the real-time operating status of the unit and market uncertainties into a unified optimization framework, enabling the pricing curve to have adaptive adjustment capabilities; and by adopting a two-stage rolling optimization architecture, this invention effectively connects multi-timescale decision-making, enhancing the pricing strategy's responsiveness to prediction deviations and unit state fluctuations.

[0050] Other features and advantages of the present invention will be set forth in the following description. Attached Figure Description

[0051] To more intuitively illustrate the prior art and this application, exemplary drawings are provided below. It should be understood that the specific shapes and structures shown in the drawings should not generally be regarded as limiting conditions for implementing this application; for example, based on the technical concept disclosed in this application and the exemplary drawings, those skilled in the art are able to easily make conventional adjustments or further optimizations to the addition / reduction / classification, specific shapes, positional relationships, connection methods, size ratios, etc. of certain units (components).

[0052] Figure 1 A flowchart illustrating the implementation of the thermal power spot market pricing optimization method based on multi-dimensional data fusion, as provided in this application embodiment. Detailed Implementation

[0053] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these embodiments are merely for further explanation of the present invention and should not be construed as limiting the scope of protection of the present invention. Those skilled in the art can make some non-essential improvements and adjustments to the present invention based on the above-described content.

[0054] Please see Figure 1 The method for optimizing thermal power spot market quotations based on multi-dimensional data fusion includes the following steps:

[0055] Step 1: Multi-source data acquisition and preprocessing;

[0056] Collect multi-source data related to the operation of thermal power units and market transactions; clean, normalize and time-align the collected multi-source data to form a standardized time-series dataset.

[0057] Multi-source data includes DCS operation data of thermal power units, electricity market clearing data, and environmental and fuel data;

[0058] Among them, the DCS operation data of thermal power units includes real-time unit output, main steam pressure, main steam temperature, reheat steam temperature, turbine vibration, boiler furnace pressure, coal feed rate, feedwater flow rate, flue gas temperature at the denitrification system inlet, and nitrogen oxide concentration, etc.

[0059] Electricity market clearing data includes day-ahead clearing prices, real-time clearing prices, system load forecasts, renewable energy (wind and solar) output forecasts, tie-line plans, and reserve capacity requirements.

[0060] Environmental and fuel data, including ambient temperature, calorific value of coal, sulfur content of coal, carbon quota trading price, and fuel oil price.

[0061] Step 2: Construct a dynamic marginal cost model for the unit across its full load range;

[0062] Based on the DCS operation data and fuel data of the unit collected in step 1, a dynamic marginal cost model is constructed to reflect the nonlinear cost characteristics of the thermal power unit in the full load range. The dynamic marginal cost model divides the unit output range into three typical sections: low load range, economic load range and high load range, and quantifies the marginal cost composition of each load range.

[0063] Specifically, the unit output at time t is The overall marginal cost at that time Represented as:

[0064] ;

[0065] in, Using the benchmark fuel cost, a quadratic function form is obtained by fitting historical operating data of the unit; This is the allocated value for the unit's start-up and shutdown costs; This is a start / stop status variable. It is set to 1 when the unit changes from a stop state to a running state, and 0 otherwise. This is the ramp-up cost coefficient, which reflects the additional coal consumption caused by changes in thermal stress and boiler combustion adjustments during the unit's load change process. The fatigue life loss cost is used to quantify the low-cycle fatigue life loss of thick-walled components such as turbine rotors and boiler drums caused by the accumulation of alternating thermal stress under deep peak shaving and frequent load changes.

[0066] Fatigue life loss cost The following method is used for quantification:

[0067] ;

[0068] in, Costs for replacing or repairing critical components (steam turbine rotor or boiler drum); The number of fatigue cycles allowed for the component; This refers to the temperature change caused by this peak-shaving process; For reference temperature change range; This is the load rate correction factor, reflecting the differentiated impact of different load ranges on fatigue damage.

[0069] The dynamic marginal cost model described above is used to obtain the dynamic marginal cost of the unit at any output point, thus forming a cost curve for the entire load range.

[0070] Step 3: Extract the operating status characteristics of the unit;

[0071] Based on the real-time DCS data collected in step 1 and the dynamic marginal cost model constructed in step 2, the feature vector of the current operating status of the unit is extracted; this feature vector is used to set dynamic constraints in the subsequent pricing optimization step.

[0072] The characteristic vector of the unit's current operating status includes: the current output value and its load range (low load zone, economic load zone, or high load zone); the output change rate between the current moment and the previous moment, used to characterize the unit's ramp-up status; the cumulative fatigue life loss value of key components; the temperature deviation between the inlet flue gas temperature of the denitrification system and the catalyst activity window; and boiler combustion stability indicators, including furnace pressure fluctuation variance and coal feed fluctuation variance.

[0073] Step 4: Generate electricity price probability scenarios;

[0074] Based on the historical electricity price data, renewable energy output forecasts, and system load forecasts collected in step 1, an electricity price probability prediction model considering the uncertainty of renewable energy output is constructed; the specific content of the generated electricity price probability scenario includes:

[0075] Error distribution modeling involves statistical analysis of the prediction errors for new energy power output and load, establishing an error probability distribution model. The error distribution model is fitted using a non-parametric kernel density estimation method to avoid prior assumptions about the shape of the error distribution.

[0076] Monte Carlo sampling is performed based on an error probability distribution model to generate N sets of error scenarios for new energy output and load. Each set of scenarios contains error values ​​for all time points within the prediction period.

[0077] Electricity price response mapping involves inputting the sampled error scenarios into the electricity price response model, mapping them to the corresponding electricity price scenarios. The electricity price response model uses the quantile regression forest method, taking renewable energy output, system load, fuel price, reserve capacity, and transmission congestion status as input features, and directly outputs the predicted values ​​of each quantile of electricity price for each time period, forming the electricity price probability density function.

[0078] Through the above steps, we obtain S electricity price scenarios and their corresponding probabilities. This provides quantitative input for uncertainty in subsequent pricing optimization.

[0079] Step 5: Optimize the segmented pricing strategy based on state dependencies;

[0080] Input the unit state feature vector obtained in step 3 and the set of electricity price probability scenarios obtained in step 4 into the bidding optimization model to solve for the optimal segmented bidding curve that maximizes the expected revenue.

[0081] The specific details of optimizing the segmented pricing strategy based on state dependencies include:

[0082] A. Price Curve Parametricization: Divide the price curve into K segments, each segment defined by the upper limit of the output range. and quoted price Two parameters are defined. ;

[0083] The output range satisfies the following relationship:

[0084] ;

[0085] The price curve satisfies the monotonically increasing constraint:

[0086] ;

[0087] B. Setting the Optimization Objective Function: With maximizing expected return as the optimization objective, the objective function is expressed as follows:

[0088] ;

[0089] in, Let be the probability of the s-th electricity price scenario; Let be the price quoted in the k-th price segment during time period t; The clearing volume of the k-th bidding segment in time period t under scenario s is determined by both the bidding curve and the market clearing mechanism; The dynamic marginal cost function defined in step 2; and These are start-up and shutdown status variables, representing the unit's start-up and shutdown actions, respectively;

[0090] C. Set constraints; the optimization process must satisfy the following constraints: in, Minimize the technical output of the generator unit; Rated output; This is the unit's maximum ramp rate; and These are the lower and upper limits of the non-economic operating range (such as the vibration zone, denitrification temperature window, etc.); These are the start-up and shutdown status variables for the generating unit; Minimum continuous running time; Minimum continuous downtime;

[0091] D. Solution and Output: An adaptive particle swarm optimization algorithm and a mixed integer linear programming nested solution strategy are used to solve the above-mentioned price optimization model, and the optimal segmented price curve and the declared output plan for each time period are output.

[0092] Step 6: Two-stage rolling optimization and correction

[0093] To adapt to the different decision-making needs of the day-ahead market and the real-time market, this invention adopts a two-stage rolling optimization architecture, which specifically includes a day-ahead market stage (first stage) and an intraday-real-time rolling correction stage (second stage).

[0094] During the current market phase, the next 24 hours are used as the optimization window, with the hour as the time granularity. Steps 4 and 5 are executed to generate the next day's 96-point segmented price curve and the next day's power generation plan, which are then submitted to the power trading center.

[0095] The intraday-real-time rolling correction phase includes the intraday rolling correction phase and the real-time market phase.

[0096] During the intraday rolling correction phase, the intraday rolling correction will be initiated when the following triggering conditions occur. After the correction is triggered, step 5 will be re-executed with an optimization window of 4-6 hours and a time granularity of 15 minutes to update the segmented price curves and output plans for subsequent time periods.

[0097] The specific triggering conditions are: the deviation of the new energy output forecast exceeds the preset threshold (for example, exceeding 15% of the forecast value); the deviation of the system load forecast exceeds the preset threshold; and the unit experiences an unplanned shutdown or output limitation.

[0098] Real-time market phase: 15 minutes before the real-time market clearing, based on the latest ultra-short-term forecast data (1-4 hours ahead) and the actual operating status of the units, perform partial adjustments in steps 3 and 5, and output the real-time market bid price and bid output.

[0099] Step 7: Execution and feedback of pricing strategy;

[0100] The segmented pricing curve output in step 6 is sent to the power trading platform through an automated interface; after real-time market clearing, the actual clearing price, the winning bid volume of the generating unit, and the actual operating data of the generating unit are collected and compared with the predicted values ​​during optimization in step 5 to calculate the prediction deviation and revenue deviation.

[0101] The deviation data of prediction deviation and revenue deviation are used as feedback signals to update the parameters of the dynamic cost model in step 2 and the parameters of the electricity price response model in step 4, so as to realize the online iterative optimization of the model.

[0102] Through the above steps, the present invention realizes end-to-end closed-loop control from multi-source data acquisition, unit status identification, electricity price scenario generation to bidding strategy optimization, which can effectively improve the profitability of thermal power units in the spot market and control operational risks.

[0103] 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, and improvements 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 optimizing thermal power spot market pricing based on multi-dimensional data fusion, characterized in that, Includes the following steps: Step 1: Collect multi-source data related to the operation of thermal power units and market transactions, and clean, normalize and time-align the multi-source data to form a standardized time-series dataset; Step 2: Based on the unit distributed control system operation data and fuel data in the multi-source data, construct a dynamic marginal cost model that reflects the nonlinear cost characteristics of thermal power units in the full load range; The dynamic marginal cost model divides the unit output range into low load zone, economic load zone and high load zone, and quantifies the marginal cost composition of each load zone. Step 3: Based on the operating data of the unit's distributed control system and the dynamic marginal cost model, extract the feature vector of the unit's current operating state; Step 4: Based on the historical electricity price data, new energy output forecasts, and system load forecasts from the multi-source data, construct an electricity price probability prediction model that considers the uncertainty of new energy output, and generate multiple electricity price scenarios and their corresponding probabilities. Step 5: Input the feature vector and the electricity price probability scenario into the pricing optimization model to solve for the optimal segmented pricing curve that maximizes expected revenue; The pricing optimization model parameterizes the pricing curve into multiple segments, each segment being defined by the upper limit of the output range and the pricing price, with the optimization objective being to maximize expected revenue, while also setting constraints. Step 6: Adopt a two-stage rolling optimization architecture, specifically as follows: During the daytime market phase, execute steps 4 and 5 to generate the next day's segmented price curve and the next day's production plan; During the intraday-real-time rolling correction phase, step 5 or steps 3 and 5 are re-executed based on the trigger conditions to update the segmented pricing curve and output plan. Step 7: Submit the segmented price curve to the power trading platform, collect actual data after market clearing, calculate the prediction deviation and revenue deviation, and use the deviation data of prediction deviation and revenue deviation as feedback signals to update the parameters of the dynamic marginal cost model and the electricity price probability prediction model.

2. The method for optimizing thermal power spot market quotations based on multi-dimensional data fusion according to claim 1, characterized in that, The multi-source data in step 1 includes: operating data of the distributed control system of thermal power units, electricity market clearing data, and environmental and fuel data; The operating data of the distributed control system of the thermal power unit includes the unit's real-time output, main steam pressure, main steam temperature, reheat steam temperature, turbine vibration, boiler furnace pressure, coal feed rate, feedwater flow rate, flue gas temperature at the denitrification system inlet, and nitrogen oxide concentration. The electricity market clearing data includes day-ahead clearing prices, real-time clearing prices, system load forecasts, renewable energy output forecasts, tie-line plans, and reserve capacity requirements. The environmental and fuel data include ambient temperature, coal calorific value, coal sulfur content, carbon quota trading price, and fuel oil price.

3. The method for optimizing thermal power spot market pricing based on multi-dimensional data fusion according to claim 1, characterized in that, In the dynamic marginal cost model of step 2, the unit output at time t is... The overall marginal cost at that time Represented as: ； in, Benchmark fuel cost; This is the allocated value for the unit's start-up and shutdown costs; These are start / stop state variables; This is the ramp-up cost coefficient; Costs related to fatigue life loss.

4. The method for optimizing thermal power spot market pricing based on multi-dimensional data fusion according to claim 3, characterized in that, The fatigue life loss cost The following method is used for quantification: ; in, Costs for replacing or repairing critical components; The number of fatigue cycles allowed for the component; This refers to the temperature change caused by this peak-shaving process; For reference temperature change range; This is the load factor correction factor.

5. The method for optimizing thermal power spot market quotations based on multi-dimensional data fusion according to claim 1, characterized in that, The feature vector of the current operating status of the unit in step 3 includes: the current output value and its load range, the output change rate between the current time and the previous time, the cumulative fatigue life loss value of key components, the temperature deviation value between the inlet flue gas temperature of the denitrification system and the catalyst activity window, and the boiler combustion stability index.

6. The method for optimizing thermal power spot market quotations based on multi-dimensional data fusion according to claim 1, characterized in that, The specific steps in step 4 of generating the electricity price probability scenario include: Statistical analysis was conducted on the prediction errors of new energy power output and load, and an error probability distribution model was established. Monte Carlo sampling is performed based on the aforementioned error probability distribution model to generate multiple sets of error scenarios for new energy output and load. The sampled error scenarios are input into the electricity price response model and mapped to the corresponding electricity price scenarios, forming the electricity price probability density function.

7. The method for optimizing thermal power spot market quotations based on multi-dimensional data fusion according to claim 1, characterized in that, In step 5, the price curve is divided into K segments, each segment consisting of the upper limit of the output range. and quoted price Two parameters are defined. ; The output range satisfies the following relationship: ； The price curve satisfies the monotonically increasing constraint: ； The optimization objective function is expressed as: ； in, Let be the probability of the s-th electricity price scenario; Let be the price quoted in the k-th price segment during time period t; This refers to the cleared electricity volume of the k-th price segment in time period t under scenario s; It is a dynamic marginal cost function; and These are start / stop state variables.

8. The method for optimizing thermal power spot market quotations based on multi-dimensional data fusion according to claim 7, characterized in that, The constraints in step 5 include upper and lower limits of output, ramp rate constraints, non-economic zone avoidance constraints, minimum continuous operating time constraints, minimum continuous downtime constraints, and price monotonicity constraints.

9. The method for optimizing thermal power spot market quotations based on multi-dimensional data fusion according to claim 1, characterized in that, The intraday-real-time rolling correction phase in step 6 includes an intraday rolling correction phase and a real-time market phase. The intraday rolling correction phase is triggered when the forecast deviation of new energy output or system load exceeds the preset threshold, or when the unit experiences an unplanned shutdown or output limitation, and step 5 is re-executed with the next 4-6 hours as the optimization window. Before the real-time market clearing, the real-time market stage performs partial adjustments in steps 3 and 5 based on ultra-short-term forecast data and the actual operating status of the units, and outputs the real-time market bid price and bid output.

10. The method for optimizing thermal power spot market quotations based on multi-dimensional data fusion according to claim 1, characterized in that, The prediction and revenue deviations in step 7 include the deviation between the actual clearing price and the predicted price, the deviation between the winning bid volume and the declared volume, and the deviation between the actual operating cost and the cost calculated by the dynamic marginal cost model. The deviation data is used to iteratively optimize the parameters of the dynamic marginal cost model and the electricity price probability prediction model online.

Images