Thermal power generating unit operation optimization method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, and provides a thermal power generating unit operation optimization method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining multi-source input data which at least comprises thermal power generating unit operation data and power market data; performing coal consumption prediction based on the operation data of the thermal power generating unit to obtain a first prediction parameter representing the coal consumption; determining a second prediction parameter representing the power price signal based on the power market data and the first prediction parameter; based on the first prediction parameter, the second prediction parameter and the multi-source input data, output and declaration parameters used for guiding the thermal power generating unit to optimize the operation state are generated. According to the method, the technical path of multi-source data fusion and coupling of internal coal consumption and external price signals for step-by-step prediction and comprehensive optimization is adopted, accurate prediction and dynamic optimization of the operation parameters of the thermal power generating unit are achieved, and the method has the advantages of being high in prediction precision and flexible in operation control.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, apparatus, electronic device, and storage medium for optimizing the operation of thermal power units. Background Technology

[0002] In production and operation, thermal power plants need to develop internal operating plans for their thermal power units to guide production operations such as unit start-up, shutdown, and output adjustment. An accurate and reasonable operating plan is crucial for ensuring the economical and efficient operation of the units. As the power grid environment becomes increasingly complex, higher demands are being placed on the accuracy and flexibility of these operating plans.

[0003] However, existing technologies have shortcomings in formulating unit operation plans. First, when predicting unit output parameters, the input information relied upon is relatively limited, usually confined to the unit's own historical data, failing to fully integrate dynamic external information from the grid side, resulting in insufficient accuracy of the prediction results. Second, the generated operation plans are mostly static plans, lacking a mechanism for real-time dynamic adjustment based on actual operating conditions, causing a lag in the unit's production control response and making it difficult to consistently maintain optimal operating conditions. Summary of the Invention

[0004] This invention provides a method, apparatus, electronic device, and storage medium for optimizing the operation of thermal power units, in order to solve the defects of existing technologies in formulating unit operation plans, which result in insufficient accuracy of prediction results and suboptimal operating status due to the limited input information.

[0005] This invention provides a method for optimizing the operation of thermal power units, comprising: Acquire multi-source input data, which includes at least thermal power unit operation data and electricity market data; Based on the operating data of the thermal power units, coal consumption is predicted to obtain a first prediction parameter characterizing coal consumption. Based on the electricity market data and the first prediction parameter, a second prediction parameter characterizing the electricity price signal is determined; Based on the first prediction parameter, the second prediction parameter, and the multi-source input data, output and reporting parameters are generated to guide the thermal power unit in optimizing its operating status.

[0006] According to the method for optimizing the operation of thermal power units provided by the present invention, the multi-source input data further includes external environmental data, and the step of predicting coal consumption based on the thermal power unit operation data to obtain a first prediction parameter characterizing coal consumption includes: Based on the thermal power unit operating data and the external environment data, a sequence of driving variables related to coal consumption is determined. Based on the coal consumption sequence and the driving variable sequence in the thermal power unit operation data, a grey relational analysis is performed, and at least one strongly correlated variable is selected from the driving variable sequence according to the analysis results. The coal consumption sequence and the strongly correlated variables are input into the first prediction model to obtain the first prediction parameters. The first prediction model is trained based on the historical coal consumption sequence and the strongly correlated variables.

[0007] According to a method for optimizing the operation of thermal power units provided by the present invention, the step of inputting the coal consumption sequence and the strongly correlated variables into a first prediction model to obtain the first prediction parameters includes: The coal consumption sequence and the strongly correlated variables are input into the first prediction model to predict the coal consumption for the next period and obtain an initial prediction value. Based on the pre-constructed state transition probability matrix, the residual correction value for the next period is determined, and the first prediction parameter is determined based on the initial prediction value and the residual correction value.

[0008] According to the method for optimizing the operation of thermal power units provided by the present invention, the steps for constructing the state transition probability matrix include: A residual sequence is generated based on the difference between the actual historical coal consumption and the predicted historical coal consumption. The value range of the residual sequence is divided into multiple preset state intervals, and the transition probability from one state interval to the next state interval in the residual sequence is calculated to construct the state transition probability matrix.

[0009] According to a method for optimizing the operation of thermal power units provided by the present invention, determining a second prediction parameter characterizing the electricity price signal based on the electricity market data and the first prediction parameter includes: The electricity market data and the first prediction parameter are input into the second prediction model to obtain the second prediction parameter output by the second prediction model; The second prediction model is a neural network model obtained by adding an attention mechanism layer to a long short-term memory network. The attention mechanism layer is used to assign different attention weights to the historical time-series features in the electricity market data input to the second prediction model.

[0010] According to a method for optimizing the operation of a thermal power unit provided by the present invention, the step of generating output and reporting parameters for guiding the optimized operation of the thermal power unit based on the first prediction parameter, the second prediction parameter, and the multi-source input data includes: Based on the upper-level optimization model, the first prediction parameter, the second prediction parameter, and the multi-source input data are applied to generate candidate output and declaration parameters. Based on the lower-level response model, simulation calculations are performed using the target constraints and the candidate output and application parameters to obtain the corresponding cleared power parameters and cleared electricity price parameters. The target constraints include new energy priority consumption constraints and transmission capacity constraints. The transmission capacity constraints are determined based on the power curtailment warning information in the multi-source input data. Based on the cleared power parameters and the cleared price parameters, the objective function value of the upper-level optimization model is calculated, and the objective function value is used to guide the upper-level optimization model to perform the next round of iteration until the optimal power output and declaration parameters are obtained.

[0011] According to the present invention, a method for optimizing the operation of thermal power units is provided. The objective function of the upper-level optimization model includes a first objective component and a second objective component. The first objective component is determined based on the cleared power parameters, the cleared power price parameters, and the first prediction parameters. The second objective component is determined based on the cleared power parameters. The weights of the first objective component and the second objective component are determined based on the environmental cost quantification parameters in the multi-source input data.

[0012] The present invention also provides a thermal power unit operation optimization device, comprising: A data acquisition unit is used to acquire multi-source input data, which includes at least thermal power unit operation data and electricity market data. The first prediction unit is used to predict coal consumption based on the operating data of the thermal power unit and obtain a first prediction parameter characterizing coal consumption. The second prediction unit is used to determine a second prediction parameter characterizing the electricity price signal based on the electricity market data and the first prediction parameter. The parameter generation unit is used to generate output and reporting parameters to guide the thermal power unit to optimize its operating status based on the first prediction parameter, the second prediction parameter and the multi-source input data.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the thermal power unit operation optimization method as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the thermal power unit operation optimization method as described above.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the thermal power unit operation optimization method as described above.

[0016] The method, apparatus, electronic equipment, and storage medium for optimizing the operation of thermal power units provided by this invention overcome the limitations of relying solely on single historical data for prediction by introducing multi-source input data, thus providing a more comprehensive and multi-dimensional technical information foundation for subsequent parameter prediction and decision-making. Specifically, this invention first focuses on the internal workings of the unit, accurately predicting coal consumption based on actual operating data. Then, this prediction result is used as a key input, combined with external electricity market data, to determine a second prediction parameter characterizing the electricity price signal. This step establishes a direct technical link between the unit's inherent economics and the external environment, making the prediction of price signals no longer an isolated, general prediction, but a more targeted prediction coupled with the generator unit's own characteristics, thereby improving the effectiveness of the prediction. Finally, when generating output and reporting parameters to guide unit operation, this invention comprehensively utilizes the aforementioned accurately predicted parameters characterizing coal consumption and price signals, as well as the original multi-source input data, to achieve global optimization decision-making, ensuring that the final generated operating parameters are not only economically reasonable but also dynamically adaptable to changes in the power grid environment. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the thermal power unit operation optimization method provided by the present invention; Figure 2 This is a schematic diagram of the process for determining the first prediction parameter provided by the present invention; Figure 3 This is a flowchart illustrating the collaborative prediction method for thermal power output and reporting parameters based on multi-source data and dynamic game theory provided by the present invention. Figure 4 This is a schematic diagram of the thermal power unit operation optimization device provided by the present invention; Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] This invention provides a method for optimizing the operation of thermal power units. This method achieves accurate prediction and dynamic optimization of the operating parameters of thermal power units by fusing multi-source heterogeneous data and constructing a technical path of step-by-step prediction and integrated optimization.

[0021] Figure 1 This is a flowchart illustrating the thermal power unit operation optimization method provided by the present invention, as shown below. Figure 1 As shown, the method includes: Step S10: Obtain multi-source input data, which includes at least thermal power unit operation data and electricity market data.

[0022] Specifically, multi-source input data refers to a comprehensive dataset encompassing both internal factors and external environments affecting the operation of thermal power units, and its sources are wide-ranging and heterogeneous. In particular, multi-source input data includes at least two core components: thermal power unit operation data and electricity market data.

[0023] Among these, the operating data of thermal power units reflects the operating status and physical characteristics of the units themselves. For example, this may include historical coal consumption data under different loads, unit operating efficiency, coal quality parameters, and the operating status and cost of environmental protection equipment. This data is the core basis for internal cost forecasting, especially coal consumption forecasting. In one possible implementation, this data can be collected in real time through the monitoring information system or distributed control system within the thermal power plant.

[0024] Electricity market data reflects the external market environment faced by generating units when participating in market transactions. For example, it can include historical electricity market prices, real-time market price fluctuation sequences, load demand forecast curves, and green electricity quotas. This data is crucial for predicting future electricity price trends and assessing generation revenue. This data can be obtained from the public data platforms of power dispatch centers or trading centers via APIs (Application Programming Interfaces) provided by industrial internet platforms.

[0025] To further improve the accuracy of forecasts and the comprehensiveness of decision-making, the multi-source input data can preferably include external environmental data and policy data. For example, external environmental data may include meteorological information (such as temperature, humidity, and wind speed over a future period), as meteorological conditions directly affect the load level of the power grid and the output of new energy sources (such as wind and solar power), thereby indirectly affecting the peak-shaving demand of thermal power and electricity prices. Policy data may include carbon emission quota prices, green electricity consumption policy requirements, and power curtailment warnings issued by the power grid; these data are important considerations for compliance constraints and risk control.

[0026] Step S20: Based on the operating data of the thermal power unit, coal consumption is predicted to obtain the first prediction parameter characterizing coal consumption.

[0027] Specifically, after acquiring comprehensive data, this step is used to predict the core operating cost of thermal power units, namely coal consumption. Its purpose is to accurately predict the unit's coal consumption under various operating conditions.

[0028] Specifically, the process of predicting coal consumption based on thermal power unit operating data can be achieved by establishing a prediction model. Considering that coal consumption data is typically small-sample data highly correlated with the characteristics of the unit itself, this embodiment of the invention can employ a grey model (GM). The core idea of ​​the grey model is to weaken its randomness and volatility by accumulating the original data sequence, thereby uncovering potential patterns and making it suitable for scenarios with limited data but requiring trend prediction. The model takes historical thermal power unit operating data, such as unit output and coal consumption for power supply, as input, and calculates and outputs predicted coal consumption values ​​for one or more future time periods.

[0029] This predicted value is the first prediction parameter. This parameter can be a specific numerical value (such as predicting coal consumption of XX tons for the next hour) or a time series (such as predicting the coal consumption series for each hour of the next 24 hours). This parameter quantitatively represents the main costs of the unit in future operation.

[0030] Step S30: Based on the electricity market data and the first prediction parameter, determine the second prediction parameter that characterizes the electricity price signal.

[0031] Specifically, after obtaining the first forecasting parameter characterizing coal consumption, this parameter can be applied, combined with electricity market data, to predict the price environment that thermal power units may face in the electricity market, which is key to determining their revenue.

[0032] Specifically, the process of determining the second prediction parameter, which characterizes the electricity price signal, based on electricity market data and the first prediction parameter can also be implemented using a specific prediction model. Since the electricity price signal is influenced by multiple factors such as market supply and demand, participant game theory, and fluctuations in new energy sources, it exhibits high nonlinearity and time-series correlation. Therefore, this embodiment of the invention can employ a Long Short-Term Memory (LSTM) model for prediction. The LSTM model is a special type of recurrent neural network capable of effectively capturing long-term dependencies in time-series data.

[0033] In this step, the model's input includes not only electricity market data such as historical electricity prices and load curves, but also the first prediction parameter representing coal consumption obtained in the previous step. Using the predicted coal consumption as one of the input features allows the model to perceive the changing trends in the costs of the thermal power units themselves. This coupling of intrinsic costs and extrinsic prices means that the prediction of the price signal parameter is no longer an isolated analysis of market behavior, but is closely related to the economics of the unit itself, thus generating more targeted and realistic prediction results, namely the second prediction parameter. This second prediction parameter can also be a time series, such as the time-of-use electricity price prediction for the next 24 hours.

[0034] Step S40: Based on the first prediction parameter, the second prediction parameter, and the multi-source input data, generate output and reporting parameters to guide the thermal power unit in optimizing its operating status.

[0035] Specifically, after obtaining the first and second prediction parameters, these two parameters can be combined with multi-source input data to generate relevant parameters for guiding the optimal operation of thermal power units, namely, output and reporting parameters. Here, the process of generating output and reporting parameters can be an optimization process aimed at achieving optimal unit operating economy or optimal overall unit operating indicators.

[0036] In this embodiment of the invention, a dynamic game optimization strategy can be employed, such as constructing a Stackelberg game model with thermal power plants as participants. In this model, the predicted coal consumption cost (reflected by the first prediction parameter) and electricity price (reflected by the second prediction parameter) from the previous step are used as the core basis for profit and loss calculation. Simultaneously, other information from multi-source input data (such as minimum technical output of the unit, ramp-up rate, environmental emission requirements, carbon price, and power curtailment warnings) are used as constraints on the model. The Nash equilibrium point of this game model is solved using an optimization algorithm (such as an improved quantum particle swarm optimization algorithm), ultimately yielding a set of optimal decision variables.

[0037] These decision variables constitute the optimal output and reporting parameters. Output parameters guide the power plant's internal production operations, such as the unit start-up and shutdown plan for the next 24 hours and the specific power generation for each time period. Reporting parameters are used for participating in electricity market transactions, such as reporting the electricity volume and price for each future time period to the trading center. For example, output parameters can guide operators to maintain a unit at its minimum technical output of 40% rated power from 2-5 AM the following day, and increase it to full rated power of 95% during the peak period from 9-11 AM. Reporting parameters could be for reporting 500 MWh of electricity at 9 AM, with a price of XX yuan / MWh.

[0038] The method provided in this invention overcomes the limitations of relying solely on single historical data for prediction by introducing multi-source input data, providing a more comprehensive and multi-dimensional technical information foundation for subsequent parameter prediction and decision-making. Specifically, this invention first focuses on the internal workings of the generating unit, accurately predicting coal consumption based on actual operating data. Then, this prediction result is used as a key input, combined with external electricity market data, to determine a second prediction parameter characterizing the electricity price signal. This step establishes a direct technical link between the unit's inherent economics and the external environment, making the price signal prediction no longer an isolated, general prediction, but a more targeted prediction coupled with the generator unit's own characteristics, thereby improving the effectiveness of the prediction. Finally, when generating the output and reporting parameters to guide unit operation, this invention comprehensively utilizes the aforementioned accurately predicted parameters characterizing coal consumption and price signals, as well as the original multi-source input data, achieving global optimization decision-making. This ensures that the final generated operating parameters are not only economically reasonable but also dynamically adaptable to changes in the power grid environment.

[0039] Based on any of the above embodiments, the multi-source input data further includes external environment data, and correspondingly, step S20 specifically includes: Step S21: Based on the operating data of the thermal power unit and the external environment data, determine the sequence of driving variables related to coal consumption.

[0040] Specifically, multi-source input data can also include external environmental data, which mainly refers to external natural condition parameters that can affect the energy conversion efficiency of thermal power units. For example, this may include ambient temperature, atmospheric pressure, ambient humidity, wind speed, and predicted regional renewable energy output. The reason for introducing this data is that the actual coal consumption of thermal power units depends not only on their own operating load but also on these external environmental data. Therefore, incorporating external environmental data allows the prediction model to more comprehensively reflect the real physical processes affecting coal consumption.

[0041] Specifically, the first step is to preliminarily screen out all potential influencing factors that may be related to changes in coal consumption from massive amounts of data, constructing a feature set to be analyzed, namely the driving power series. Here, the driving variable series refers to a series of data that characterizes the unit's operating status or external environment and changes over time. These variables are considered potential causes driving changes in coal consumption.

[0042] These driving variable sequences can be obtained from two sources: one part comes from thermal power unit operating data, such as unit load rate sequences, main steam pressure sequences, reheat steam temperature sequences, and feedwater temperature sequences; the other part comes from newly added external environmental data, such as ambient temperature and humidity sequences and atmospheric pressure sequences. By integrating these data, a multi-dimensional driving variable matrix can be formed, preparing for subsequent correlation analysis.

[0043] Step S22: Based on the coal consumption sequence and the driving variable sequence in the thermal power unit operation data, perform grey relational analysis, and select at least one strongly correlated variable from the driving variable sequence according to the analysis results.

[0044] Specifically, not all driving variables have a significant impact on coal consumption. To avoid model redundancy and noise interference, screening is necessary first. Grey relational analysis is used to calculate the correlation between each driving variable and the target feature sequence (i.e., the coal consumption sequence).

[0045] Here, grey relational analysis is a multi-factor statistical analysis method that judges the strength of the relationship between different sequences based on the similarity of their geometric shapes. Specifically, the historical coal consumption sequence is first used as a reference sequence, and the sequences of driving variables determined in the previous step (such as unit load rate sequence, ambient temperature sequence, etc.) are used as comparison sequences. By calculating the grey relational coefficient between each comparison sequence and the reference sequence, a quantitative correlation value is finally obtained. The magnitude of this correlation value intuitively reflects the similarity between the changing trend of the driving variable and the changing trend of coal consumption, i.e., the strength of the correlation.

[0046] After obtaining the correlation between each driving variable and coal consumption, screening is required. Strongly correlated variables can be defined by setting a predetermined threshold (e.g., correlation greater than 0.75). Driving variables with a correlation exceeding this threshold are considered major factors affecting coal consumption and are thus screened out. This method effectively eliminates noise interference from weakly correlated or uncorrelated variables.

[0047] Step S23: Input the coal consumption sequence and the strongly correlated variables into the first prediction model to obtain the first prediction parameters. The first prediction model is trained based on the historical coal consumption sequence and the strongly correlated variables.

[0048] Specifically, after selecting the strongly correlated variables, the historical coal consumption sequence and these strongly correlated variables can be input into the first prediction model to obtain the first prediction parameters characterizing coal consumption. Here, the first prediction model can be a grey model, or more specifically, a GM(1,N) model containing multiple variables.

[0049] Furthermore, when the grey model predicts coal consumption based on the input data, in order to cope with the random disturbances and abrupt changes in actual industrial data, this embodiment of the invention introduces an adaptive disturbance factor to dynamically correct the coal consumption prediction deviation. Specifically, when the data fluctuates significantly (i.e., the disturbance) at a certain point in time, this factor will be used to appropriately reduce the weight of the data point in the parameter fitting, so as to avoid the model being affected by noise.

[0050] Understandably, before using the first prediction model to predict coal consumption, it can be trained to obtain the prediction model. Specifically, during the model training phase, historical coal consumption sequences and the corresponding historical sequences of all selected strongly correlated variables can be used as input to train the GM(1,N) model, enabling it to learn and fit the quantitative relationship between the strongly correlated variables and coal consumption. "Training" means determining the parameters in the model using historical data to minimize the model's fitting error to historical data. During the prediction phase, the latest coal consumption data and the latest data of each strongly correlated variable are input into the trained model, which then outputs predicted coal consumption values ​​for one or more future time periods, thus obtaining the first prediction parameters.

[0051] The method provided in this invention uses grey relational analysis to scientifically screen input features, avoiding interference from irrelevant or weakly correlated variables on the model. This allows the coal consumption prediction model (i.e., the first prediction model) to focus more on core influencing factors, thereby significantly improving the accuracy and reliability of the prediction results (i.e., the first prediction parameters).

[0052] Based on any of the above embodiments Figure 2 This is a schematic diagram of the process for determining the first prediction parameter provided by the present invention, as shown below. Figure 2 As shown, step S23 specifically includes: Step S231: Input the coal consumption sequence and the strongly correlated variables into the first prediction model to predict the coal consumption for the next period and obtain an initial prediction value. Step S232: Based on the pre-constructed state transition probability matrix, determine the residual correction value for the next period, and determine the first prediction parameter according to the initial prediction value and the residual correction value.

[0053] It should be noted that, considering the error (i.e., residual) between the predicted value and the true value output by the first prediction model, these residual sequences are often not completely random, but rather exhibit certain state transition patterns. Therefore, this embodiment of the invention does not use the direct output of the first prediction model as the final result, but instead introduces a residual correction mechanism based on the Markov chain concept to dynamically compensate for the initial predicted value.

[0054] Specifically, historical coal consumption sequences and selected strongly correlated variables are used as raw input data. First, this data undergoes preprocessing, such as data cleaning and normalization. Then, based on the preprocessed data, a pre-trained prediction model (e.g., the GM(1,N) model) is used to calculate a preliminary prediction of coal consumption for the next period. The next period can be the next 24 hours or other preset time steps. The prediction result directly output by the model is the initial prediction value. This initial prediction value reflects the coal consumption trend judged by the model based on the main influencing factors, but it may still contain some prediction errors, which sometimes exhibit certain regularities.

[0055] To capture and utilize the patterns of prediction errors, this invention introduces Markov chain residual correction. Its core is to determine the residual correction value for the next cycle based on a pre-constructed state transition probability matrix. Here, it is necessary to first explain the state transition probability matrix and its construction process. The construction steps of the state transition probability matrix are completed offline before model application, and specifically include: A residual sequence is generated based on the difference between the actual historical coal consumption and the predicted historical coal consumption. The value range of the residual sequence is divided into multiple preset state intervals, and the transition probability from one state interval to the next state interval in the residual sequence is calculated to construct the state transition probability matrix.

[0056] Specifically, to construct the state transition probability matrix, we first utilize existing historical data to perform a comprehensive historical backtesting prediction using the first prediction model, obtaining a sequence of predicted historical coal consumption values. Then, we subtract this predicted value sequence from the historical actual value sequence at each time point (residual = actual value - predicted value), thus obtaining a residual sequence containing all historical prediction error information of the model. Next, we divide the value range of the residual sequence into multiple predefined state intervals. This is the process of discretizing continuous residual values ​​into a finite state space; for example, it can be divided into three state intervals. The number and width of the state intervals can be determined based on the distribution characteristics of the residuals, using methods such as equal-width partitioning, equal-frequency partitioning, or cluster analysis to ensure the rationality of the partitioning. Next, we calculate the transition probability from one state interval to the next in the residual sequence, which can be done by traversing the entire historical residual sequence. Finally, we arrange all the calculated transition probabilities in a row-column correspondence format to construct the state transition probability matrix.

[0057] After pre-constructing the aforementioned matrix, residual correction can be performed online. First, based on the actual residual value of the current period (or the previous period), the state interval to which it belongs is determined. Then, the row corresponding to that state interval is found in the matrix; this row gives all the possible states the residual may enter in the next period and their corresponding probabilities. To obtain a definite residual correction value, the expected value of the residual in the next period can be calculated. For example, the midpoint value of each state interval is taken as the representative value of that state, and then each representative value is multiplied by its corresponding transition probability and summed. The resulting expected value is the predicted residual for the next period, which is the required residual correction value.

[0058] Finally, a simple addition operation is performed based on the initial predicted value and the residual correction value to obtain the final first prediction parameter. By compensating the prediction error back to the initial predicted value, a corrected and more accurate final coal consumption prediction result is obtained.

[0059] The method provided in this invention constructs a state transition probability matrix that describes the evolution of prediction error, and uses this matrix to predict the error value at the next moment to correct the initial prediction result of the model. This effectively identifies and compensates for the systematic bias of the basic prediction model, and significantly improves the accuracy of coal consumption prediction.

[0060] Based on any of the above embodiments, step S30 specifically includes: The electricity market data and the first prediction parameter are input into the second prediction model to obtain the second prediction parameter output by the second prediction model; The second prediction model is a neural network model obtained by adding an attention mechanism layer to a long short-term memory network. The attention mechanism layer is used to assign different attention weights to the historical time-series features in the electricity market data input to the second prediction model.

[0061] It should be noted that in the actual operation of power generation companies, electricity market price signals (such as day-ahead clearing prices) are highly nonlinear, highly volatile, and have complex time-series dependencies. Simple time series forecasting models often struggle to capture their inherent patterns. In particular, the importance of data from different historical points in time for predicting prices at a specific future point in time varies. To address this, this embodiment of the invention introduces an attention mechanism in the second forecasting model.

[0062] Specifically, the second prediction model is a neural network model obtained by adding an attention mechanism layer to the Long Short-Term Memory (LSTM) network. The LSTM network, as the basic model, is itself a special type of recurrent neural network. Through its internal gate structure, it can effectively learn and remember long-term dependencies in time series data, making it suitable for processing data with obvious time correlations, such as electricity prices and loads.

[0063] Preferably, this embodiment of the invention adds an attention mechanism layer on top of the LSTM network structure. This attention mechanism layer assigns different attention weights to historical time-series features in the input electricity market data. Its working principle can be understood as simulating human attention mechanisms: when the model needs to predict electricity prices at a future point in time, this attention mechanism layer automatically and selectively focuses on the parts of the input historical sequence most relevant to the current prediction task and assigns these parts higher weights; while assigning lower weights to parts with low relevance. This dynamic allocation of weights enables the model to accurately capture the most crucial predictive information from noisy and lengthy historical data.

[0064] Based on any of the above embodiments, step S40 specifically includes: Step S41: Based on the upper-level optimization model, the first prediction parameter, the second prediction parameter, and the multi-source input data are applied to generate candidate output and declaration parameters.

[0065] It should be noted that, in order to generate the optimal output and reporting parameters, this embodiment of the invention constructs a two-layer optimization model that can simulate the actual game process and take into account multiple objectives, so as to ensure that the final generated operation and reporting strategy reaches the optimal under complex constraints.

[0066] Specifically, the upper-level optimization model represents the decision-making brain of thermal power plants. Its goal is to formulate optimal operation and reporting strategies while meeting its own operational constraints. In this step, the first prediction parameter (predicted coal consumption), the second prediction parameter (predicted price signal), and unit operating parameters (such as minimum technical output, ramp-up rate, etc.) from multi-source input data, external environmental data (impact on load, renewable energy output), and policy data (carbon price, power curtailment warning, etc.) are used to generate candidate output and reporting parameters through an optimization algorithm (such as an improved quantum particle swarm optimization algorithm).

[0067] Understandably, the core of solving this higher-level optimization model lies in its objective function. This objective function can include a first objective component and a second objective component, aiming to achieve a balance between economic and environmental benefits. The first objective component aims to maximize the firm's profit; it is calculated based on the cleared electricity quantity parameter, the cleared electricity price parameter, and the first prediction parameter. The second objective component aims to minimize the environmental impact of power generation, typically quantified as carbon emission costs. It can be determined based on the cleared electricity quantity parameter, as power generation (i.e., cleared electricity quantity) and carbon emissions are generally positively correlated.

[0068] The final objective function of the upper-level model can be expressed as a weighted sum of the first and second objective components. The weights of these two components are not fixed but can be dynamically adjusted based on environmental cost quantification parameters from the multi-source input data. For example, the environmental cost quantification parameter could be the carbon emission allowance price. When the carbon price increases, the second objective component automatically increases, making the model more inclined to sacrifice some power generation revenue in exchange for lower carbon emissions, thus adapting the overall strategy to changes in external environmental policies.

[0069] Step S42: Based on the lower-level response model, simulation calculations are performed using the target constraints and the candidate output and application parameters to obtain the corresponding cleared power parameters and cleared price parameters. The target constraints include new energy priority consumption constraints and transmission capacity constraints. The transmission capacity constraints are determined based on the power curtailment warning information in the multi-source input data.

[0070] Specifically, the role of the lower-level response model is to simulate the clearing process of the electricity market in order to evaluate the candidate strategies given by the upper-level model. In this step, simulation calculations are performed using target constraints and candidate output and declaration parameters.

[0071] Here, the target constraints are the rules that must be followed during the simulated market clearing process. These can include constraints on prioritizing renewable energy consumption and transmission capacity constraints. The renewable energy priority consumption constraint means that, in the simulation, the predicted power generation from renewable energy sources such as wind and solar power must be satisfied first, and thermal power units can only compete for the remaining load space. The specific value of the transmission capacity constraint is determined based on the power curtailment warning information in the multi-source input data.

[0072] The lower-level model receives the candidate application parameters from the upper-level model, combines them with the aforementioned constraints, simulates the market clearing algorithm, and finally obtains the corresponding cleared electricity volume parameters and cleared electricity price parameters. These two parameters represent the actual electricity volume that the thermal power unit can generate and the transaction price under the candidate strategy, respectively.

[0073] Step S43: Based on the cleared power parameters and the cleared price parameters, calculate the objective function value of the upper-level optimization model, and use the objective function value to guide the upper-level optimization model to perform the next round of iteration until the optimal power output and reporting parameters are obtained.

[0074] Specifically, the clearing results (cleared electricity volume and price parameters) calculated by the lower-level model are fed back to the upper-level model. Based on the cleared electricity volume and price parameters, the upper-level model calculates the objective function value of its optimization model; this value represents the overall score of the strategy in that round. Subsequently, the optimization algorithm uses the objective function value to guide the upper-level optimization model in the next iteration, generating a new set of candidate parameters that are more likely to achieve a high score. This process is repeated until the optimal output and reporting parameters are obtained, meaning the objective function value of the upper-level model converges and no longer shows significant improvement. The candidate parameters obtained at this point are the final decision result.

[0075] The method provided in this invention iteratively optimizes its output and reporting strategies through simulated feedback, while combining multi-objective optimization with a dynamic weight adjustment mechanism to enable flexible decision-making in response to policy changes such as carbon prices. Furthermore, by introducing realistic constraints such as grid curtailment warnings, the feasibility and security of the final solution are ensured.

[0076] Based on any of the above embodiments Figure 3 This is a flowchart illustrating the collaborative prediction method for thermal power output and reporting parameters based on multi-source data and dynamic game theory provided by this invention. Figure 3 As shown, the system architecture consists of three layers, from top to bottom: the data layer, the model layer, and the application layer, forming a closed loop of data acquisition, model prediction, decision optimization, and result output. The method includes the following steps: S1, multi-source heterogeneous data fusion.

[0077] First, the system collects multi-dimensional and heterogeneous data in real time through various interfaces such as the Industrial Internet Platform API. Data sources include thermal power unit operation data, such as unit output, coal consumption, efficiency, and environmental protection equipment status; electricity market data, such as historical electricity prices, spot market clearing prices, carbon quota prices, and green electricity trading prices; external environmental data, such as meteorological data (temperature, humidity, wind speed, etc.) based on satellite remote sensing and ground monitoring, and predicted values ​​of renewable energy output; and policy data, such as power curtailment warnings (e.g., warnings when transmission capacity margin falls below 15%) and carbon emission prices. Standardization processes eliminate dimensional differences, providing comprehensive and high-quality input for subsequent models.

[0078] S2, Construct a hybrid prediction model.

[0079] The model consists of two parts: a grey model and an LSTM model, which are fused using adaptive weights. For data with small sample characteristics, such as coal consumption, grey relational analysis is first used to screen strongly correlated variables (e.g., correlation > 0.8), and then an improved grey model is used for trend prediction. The model also uses Markov chains for residual correction to improve prediction accuracy.

[0080] For data series like electricity prices, which exhibit strong time-series characteristics and large data volumes, an LSTM model is employed for prediction. This model not only utilizes historical electricity price information but also incorporates coal consumption costs predicted by a grey model as one of its input features, achieving a correlation between cost and price prediction. Furthermore, an attention mechanism is added to the model, enabling it to automatically focus on historical time points that have the greatest impact on the current prediction, thus enhancing its ability to capture sudden price changes.

[0081] Furthermore, an adaptive weight allocation module can dynamically adjust the weights of the grey model and the LSTM model based on their root mean square error (RMSE) in recent predictions. The better-performing model will receive a higher weight, thus achieving complementary advantages and outputting a more accurate final prediction result.

[0082] S3 is an optimization of output and reporting strategies based on dynamic game theory.

[0083] After obtaining accurate coal consumption and electricity price forecasts, the system activates a dynamic game theory module to generate the optimal output and application strategy. During this generation process, it considers not only the predicted coal consumption and electricity price, but also unit efficiency, market data, external environmental data, and policy data from thermal power unit operation data to construct target constraints and ensure the feasibility and security of the final parameters.

[0084] The thermal power unit operation optimization device provided by the present invention is described below. The thermal power unit operation optimization device described below and the thermal power unit operation optimization method described above can be referred to in correspondence.

[0085] Based on any of the above embodiments Figure 4 This is a schematic diagram of the thermal power unit operation optimization device provided by the present invention, as shown below. Figure 4 As shown, the device includes: Data acquisition unit 410 is used to acquire multi-source input data, which includes at least thermal power unit operation data and electricity market data; The first prediction unit 420 is used to predict coal consumption based on the operating data of the thermal power unit and obtain a first prediction parameter characterizing coal consumption. The second prediction unit 430 is used to determine a second prediction parameter characterizing the electricity price signal based on the electricity market data and the first prediction parameter; The parameter generation unit 440 is used to generate output and reporting parameters to guide the thermal power unit to optimize its operating status based on the first prediction parameter, the second prediction parameter and the multi-source input data.

[0086] The apparatus provided in this invention overcomes the limitations of relying solely on single historical data for prediction by introducing multi-source input data, providing a more comprehensive and multi-dimensional technical information foundation for subsequent parameter prediction and decision-making. Specifically, this invention first focuses on the internal workings of the generating unit, accurately predicting coal consumption based on actual operating data. Then, this prediction result is used as a key input, combined with external electricity market data, to determine a second prediction parameter characterizing the electricity price signal. This step establishes a direct technical link between the unit's inherent economics and the external environment, making the prediction of price signals no longer an isolated, general prediction, but a more targeted prediction coupled with the generator unit's own characteristics, thereby improving the effectiveness of the prediction. Finally, when generating output and reporting parameters to guide unit operation, this invention comprehensively utilizes the aforementioned accurately predicted parameters characterizing coal consumption and price signals, as well as the original multi-source input data, to achieve global optimization decision-making, ensuring that the final generated operating parameters are not only economically reasonable but also dynamically adaptable to changes in the power grid environment.

[0087] Based on any of the above embodiments, the multi-source input data further includes external environment data, and the first prediction unit 420 specifically includes: The variable determination subunit is used to determine the sequence of driving variables related to coal consumption based on the thermal power unit operating data and the external environment data. The correlation analysis subunit is used to perform grey correlation analysis based on the coal consumption sequence and the driving variable sequence in the thermal power unit operation data, and to select at least one strongly correlated variable from the driving variable sequence according to the analysis results. The parameter prediction subunit is used to input the coal consumption sequence and the strongly correlated variables into the first prediction model to obtain the first prediction parameters. The first prediction model is trained based on the historical coal consumption sequence and the strongly correlated variables.

[0088] Based on any of the above embodiments, the parameter prediction subunit is specifically used for: The coal consumption sequence and the strongly correlated variables are input into the first prediction model to predict the coal consumption for the next period and obtain an initial prediction value. Based on the pre-constructed state transition probability matrix, the residual correction value for the next period is determined, and the first prediction parameter is determined based on the initial prediction value and the residual correction value.

[0089] Based on any of the above embodiments, the first prediction unit 420 further includes a matrix construction subunit, the matrix construction subunit being used for: A residual sequence is generated based on the difference between the actual historical coal consumption and the predicted historical coal consumption. The value range of the residual sequence is divided into multiple preset state intervals, and the transition probability from one state interval to the next state interval in the residual sequence is calculated to construct the state transition probability matrix.

[0090] Based on any of the above embodiments, the second prediction unit 430 is specifically used for: The electricity market data and the first prediction parameter are input into the second prediction model to obtain the second prediction parameter output by the second prediction model; The second prediction model is a neural network model obtained by adding an attention mechanism layer to a long short-term memory network. The attention mechanism layer is used to assign different attention weights to the historical time-series features in the electricity market data input to the second prediction model.

[0091] Based on any of the above embodiments, the parameter generation unit 440 is specifically used for: Based on the upper-level optimization model, the first prediction parameter, the second prediction parameter, and the multi-source input data are applied to generate candidate output and declaration parameters. Based on the lower-level response model, simulation calculations are performed using the target constraints and the candidate output and application parameters to obtain the corresponding cleared power parameters and cleared electricity price parameters. The target constraints include new energy priority consumption constraints and transmission capacity constraints. The transmission capacity constraints are determined based on the power curtailment warning information in the multi-source input data. Based on the cleared power parameters and the cleared price parameters, the objective function value of the upper-level optimization model is calculated, and the objective function value is used to guide the upper-level optimization model to perform the next round of iteration until the optimal power output and declaration parameters are obtained.

[0092] Based on any of the above embodiments, the objective function of the upper-level optimization model includes a first objective component and a second objective component. The first objective component is determined based on the cleared electricity parameter, the cleared electricity price parameter, and the first prediction parameter. The second objective component is determined based on the cleared electricity parameter. The weights of the first objective component and the second objective component are determined based on the environmental cost quantification parameter in the multi-source input data.

[0093] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a thermal power unit operation optimization method, which includes: acquiring multi-source input data, the multi-source input data including at least thermal power unit operation data and electricity market data; predicting coal consumption based on the thermal power unit operation data to obtain a first prediction parameter characterizing coal consumption; determining a second prediction parameter characterizing electricity price signals based on the electricity market data and the first prediction parameter; and generating output and reporting parameters to guide the optimized operation of the thermal power unit based on the first prediction parameter, the second prediction parameter, and the multi-source input data.

[0094] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to related technologies, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0095] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the thermal power unit operation optimization method provided by the above methods. The method includes: acquiring multi-source input data, which includes at least thermal power unit operation data and electricity market data; predicting coal consumption based on the thermal power unit operation data to obtain a first prediction parameter characterizing coal consumption; determining a second prediction parameter characterizing electricity price signals based on the electricity market data and the first prediction parameter; and generating output and reporting parameters for guiding the optimized operation state of the thermal power unit based on the first prediction parameter, the second prediction parameter, and the multi-source input data.

[0096] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the thermal power unit operation optimization method provided by the above methods. The method includes: acquiring multi-source input data, the multi-source input data including at least thermal power unit operation data and electricity market data; predicting coal consumption based on the thermal power unit operation data to obtain a first prediction parameter characterizing coal consumption; determining a second prediction parameter characterizing electricity price signals based on the electricity market data and the first prediction parameter; and generating output and reporting parameters for guiding the optimized operation state of the thermal power unit based on the first prediction parameter, the second prediction parameter, and the multi-source input data.

[0097] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0098] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing the operation of thermal power units, characterized in that, include: Acquire multi-source input data, which includes at least thermal power unit operation data and electricity market data; Based on the operating data of the thermal power units, coal consumption is predicted to obtain a first prediction parameter characterizing coal consumption. Based on the electricity market data and the first prediction parameter, a second prediction parameter characterizing the electricity price signal is determined; Based on the first prediction parameter, the second prediction parameter, and the multi-source input data, output and reporting parameters are generated to guide the thermal power unit in optimizing its operating status.

2. The thermal power unit operation optimization method according to claim 1, characterized in that, The multi-source input data also includes external environmental data. The prediction of coal consumption based on the thermal power unit operating data, to obtain a first prediction parameter characterizing coal consumption, includes: Based on the thermal power unit operating data and the external environment data, a sequence of driving variables related to coal consumption is determined. Based on the coal consumption sequence and the driving variable sequence in the thermal power unit operation data, a grey relational analysis is performed, and at least one strongly correlated variable is selected from the driving variable sequence according to the analysis results. The coal consumption sequence and the strongly correlated variables are input into the first prediction model to obtain the first prediction parameters. The first prediction model is trained based on the historical coal consumption sequence and the strongly correlated variables.

3. The thermal power unit operation optimization method according to claim 2, characterized in that, The step of inputting the coal consumption sequence and the strongly correlated variables into the first prediction model to obtain the first prediction parameters includes: The coal consumption sequence and the strongly correlated variables are input into the first prediction model to predict the coal consumption for the next period and obtain an initial prediction value. Based on the pre-constructed state transition probability matrix, the residual correction value for the next period is determined, and the first prediction parameter is determined based on the initial prediction value and the residual correction value.

4. The thermal power unit operation optimization method according to claim 3, characterized in that, The steps for constructing the state transition probability matrix include: A residual sequence is generated based on the difference between the actual historical coal consumption and the predicted historical coal consumption. The value range of the residual sequence is divided into multiple preset state intervals, and the transition probability from one state interval to the next state interval in the residual sequence is calculated to construct the state transition probability matrix.

5. The method for optimizing the operation of thermal power units according to claim 1, characterized in that, The step of determining a second forecasting parameter characterizing the electricity price signal based on the electricity market data and the first forecasting parameter includes: The electricity market data and the first prediction parameter are input into the second prediction model to obtain the second prediction parameter output by the second prediction model; The second prediction model is a neural network model obtained by adding an attention mechanism layer to a long short-term memory network. The attention mechanism layer is used to assign different attention weights to the historical time-series features in the electricity market data input to the second prediction model.

6. The method for optimizing the operation of thermal power units according to any one of claims 1 to 5, characterized in that, The step of generating output and reporting parameters to guide the optimized operation of the thermal power unit based on the first prediction parameter, the second prediction parameter, and the multi-source input data includes: Based on the upper-level optimization model, the first prediction parameter, the second prediction parameter, and the multi-source input data are applied to generate candidate output and declaration parameters. Based on the lower-level response model, simulation calculations are performed using the target constraints and the candidate output and application parameters to obtain the corresponding cleared power parameters and cleared electricity price parameters. The target constraints include new energy priority consumption constraints and transmission capacity constraints. The transmission capacity constraints are determined based on the power curtailment warning information in the multi-source input data. Based on the cleared power parameters and the cleared price parameters, the objective function value of the upper-level optimization model is calculated, and the objective function value is used to guide the upper-level optimization model to perform the next round of iteration until the optimal power output and declaration parameters are obtained.

7. The thermal power unit operation optimization method according to claim 6, characterized in that, The objective function of the upper-level optimization model includes a first objective component and a second objective component. The first objective component is determined based on the cleared electricity parameter, the cleared electricity price parameter, and the first prediction parameter. The second objective component is determined based on the cleared electricity parameter. The weights of the first objective component and the second objective component are determined based on the environmental cost quantification parameter in the multi-source input data.

8. A thermal power unit operation optimization device, characterized in that, include: A data acquisition unit is used to acquire multi-source input data, which includes at least thermal power unit operation data and electricity market data. The first prediction unit is used to predict coal consumption based on the operating data of the thermal power unit and obtain a first prediction parameter characterizing coal consumption. The second prediction unit is used to determine a second prediction parameter characterizing the electricity price signal based on the electricity market data and the first prediction parameter. The parameter generation unit is used to generate output and reporting parameters to guide the thermal power unit to optimize its operating status based on the first prediction parameter, the second prediction parameter and the multi-source input data.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the thermal power unit operation optimization method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the thermal power unit operation optimization method as described in any one of claims 1 to 7.

Images