Power plant operation intelligent decision-making method, system, equipment and medium

Abstract

The invention discloses an intelligent decision-making method, system, equipment and medium for power plant operation, and relates to the technical field of intelligent decision-making for power plant operation, and the method comprises the steps: collecting historical operation data, real-time state data and power plant multi-source heterogeneous data, carrying out the feature extraction and training through the historical operation data, and constructing a time sequence prediction model in combination with a deep learning structure; inputting the historical operation data and the real-time state data into a time sequence prediction model to generate a target load prediction result; fusing the target load prediction result with the multi-source heterogeneous data of the power plant, and constructing a structured input data set; based on the structured input data set, constructing a multi-target constraint coal blending combustion model, and executing coal type combination and distribution calculation; according to a coal type combination and distribution calculation result, executing a strategy simulation and scheduling task generation process to form strategy parameters; and reversely correcting data acquisition configuration and prediction model parameter setting based on strategy parameters. According to the method, the hysteresis problem of a simple regression model is solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent decision-making technology for power plant operation, specifically to an intelligent decision-making method, system, equipment, and medium for power plant operation. Background Technology

[0002] Due to the strong uncertainty of external factors such as fuel costs, load fluctuations, and electricity pricing mechanisms, traditional operating models that rely on experience or static dispatch tables are unable to cope with real-time changes. There are a large number of heterogeneous subsystems within power plants, including boiler control systems, auxiliary equipment management systems, coal yard management systems, and information dispatch platforms. The data structures of these subsystems are inconsistent and the interaction efficiency is low, which restricts the ability to coordinate and optimize the overall operating strategy.

[0003] In traditional technologies, load forecasting typically employs methods based on statistical regression or simple time series analysis, which struggles to capture the deep dynamic correlation between operating conditions and external influencing factors. Furthermore, coal blending strategy design is often based on historical experience and static parameters, lacking a regulatory mechanism linked to real-time electricity prices and equipment status, resulting in low fuel utilization efficiency and inadequate output planning. Simultaneously, dispatching strategy formulation is mostly a one-way process, lacking the ability to reverse-engineer the data acquisition system and forecasting models based on strategy execution results, thus failing to achieve closed-loop optimization.

[0004] Existing technologies struggle to efficiently integrate and utilize multi-source heterogeneous data from power plants, collaboratively integrate real-time load forecasting with coal blending models, and dynamically optimize data acquisition configurations and model parameters based on strategy simulation results.

[0005] Therefore, there is an urgent need for a smart decision-making method for power plants that can integrate real-time data perception and strategy optimization to achieve coal type allocation, task scheduling and parameter adaptive adjustment driven by multiple data sources, thereby improving the overall operating economy and responsiveness of power plants. Summary of the Invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, the technical problem solved by this invention is: existing power plant operation strategy formulation methods suffer from insufficient load forecasting accuracy, static and fixed coal distribution scheduling, lack of closed-loop feedback in strategy adjustment, and the problem of how to integrate real-time operation data with multi-source heterogeneous information to construct an intelligent scheduling method with data-driven capabilities, forecasting accuracy, and adaptive optimization capabilities.

[0008] To address the aforementioned technical problems, this invention provides the following technical solution: a smart decision-making method for power plant operation, comprising collecting historical operating data, real-time status data, and multi-source heterogeneous data of the power plant, extracting and training features from historical operating data, and constructing a time-series prediction model by combining a deep learning structure.

[0009] Historical operating data and real-time status data are input into the time-series forecasting model to generate target load forecasting results.

[0010] The target load forecast results are fused with multi-source heterogeneous data from power plants to construct a structured input data set.

[0011] Based on a structured input dataset, a multi-objective constrained coal blending model is constructed, and coal type combination and allocation calculations are performed.

[0012] Based on the coal type combination and allocation calculation results, the strategy simulation and scheduling task generation process is executed to form strategy parameters.

[0013] Based on the strategy parameters, the data acquisition configuration and prediction model parameter settings are corrected in reverse.

[0014] As a preferred embodiment of the intelligent decision-making method for power plant operation described in this invention, the collection of historical operating data, real-time status data, and multi-source heterogeneous data of the power plant includes: acquiring historical operating data by setting up acquisition interfaces for the boiler main control system and auxiliary machine monitoring system; acquiring real-time status data by setting up real-time monitoring nodes; extracting multi-source heterogeneous data of the power plant from the power plant management information system; and uniformly encoding the three types of data and writing them into the data cache area.

[0015] Historical operating data includes load changes, fuel usage, and equipment output records.

[0016] Real-time status data includes power generation, thermal load, steam flow rate, and flue gas temperature.

[0017] The power plant's multi-source heterogeneous data includes dispatch data, electricity price information, fuel inventory status, equipment maintenance plans, electricity prices, coal yard status, and equipment maintenance plans.

[0018] As a preferred embodiment of the intelligent decision-making method for power plant operation described in this invention, the generation of target load prediction results includes processing historical operating data by sliding slicing according to time windows, and selecting time series feature fields as input vectors based on data characteristics.

[0019] The input vector is concatenated with real-time state data and then fed into the temporal modeling unit in the deep learning structure.

[0020] By using a multi-layered nested time recursive network, an output sequence reflecting future load change trends is generated.

[0021] The output sequence is bound to time stamps to form the target load prediction result.

[0022] As a preferred embodiment of the intelligent decision-making method for power plant operation described in this invention, the construction of the structured input data set includes aligning the load time series in the target load prediction results with the multi-source heterogeneous data of the power plant according to the time dimension, and performing field mapping, dimension expansion and normalization processing on the data.

[0023] Construct a unified data structure format to generate a set of structured input data containing predicted load, electricity price level, equipment status, and coal yard inventory information at each time point.

[0024] The structured input data set is arranged in sequential order and serves as the unified input source for the multi-objective constrained coal blending and combustion model.

[0025] As a preferred embodiment of the intelligent decision-making method for power plant operation described in this invention, the construction of the multi-objective constrained coal blending model includes using the predicted load, electricity price level, equipment status, and coal yard inventory information from the structured input data set as input to establish scheduling constraints.

[0026] A feasibility constraint matrix is ​​established based on the type of coal to be blended, and the corresponding coal blending ratio variables are defined.

[0027] By constructing an objective function, the fuel allocation problem is transformed into a multi-objective optimization problem. The structured dataset is input into the multi-objective constrained coal blending model to obtain the allocation ratio of each type of coal.

[0028] The objective function includes the calorific value balance constraint, the cost weighting function, and the output feasibility function.

[0029] As a preferred embodiment of the intelligent decision-making method for power plant operation described in this invention, the calculation of coal type combination and allocation includes calculating the actual allocation quantity of each coal type by combining the allocation ratio of each coal type with the current coal yard inventory data.

[0030] Establish a coal allocation plan and generate a sequence of co-firing instructions for the boilers based on the allocation results.

[0031] The instruction sequence serves as input data for policy simulation and is used for configuration calculations in the policy generation process.

[0032] As a preferred embodiment of the intelligent decision-making method for power plant operation described in this invention, the formation of strategy parameters includes inputting the coal type combination and allocation calculation results into the strategy generation process, and configuring a corresponding scheduling task execution table based on equipment constraints, electricity price constraints and fuel blending parameters.

[0033] Extract control parameters based on the task items generated in the scheduling execution table.

[0034] Control parameters include equipment start / stop signals, fuel adjustment commands, and load distribution sequences.

[0035] The control parameters are encoded as policy parameters, and the policy parameters are written into the policy buffer as the final scheduling input.

[0036] The reverse correction of data acquisition configuration and prediction model parameter settings includes the introduction of a two-layer interactive judgment mechanism after the strategy parameters are generated, which enables the setting of parameters for data acquisition configuration and prediction model.

[0037] The two-layer interactive judgment mechanism includes collecting the actual scheduling results and calculating the execution residual with the policy target output, setting a residual threshold, and triggering a second-level judgment process if the execution residual value is greater than the set residual value.

[0038] After the residual is triggered, a judgment matrix is ​​constructed based on the source of error for further detailed judgment.

[0039] If the load response delay is greater than the delay response threshold, it is inferred that the prediction model is lagging, and the model parameter set is adjusted.

[0040] If the equipment start-up and shutdown response deviates abnormally, it is determined that the control signal accuracy is too low, and the control parameter structure and task frequency are adjusted.

[0041] If the failure rate of coal type allocation increases, it indicates that the granularity of the collected data is insufficient. The sampling cycle should be adjusted and the number of data nodes increased.

[0042] As a preferred embodiment of the intelligent decision-making system for power plant operation described in this invention, it includes: a data perception and modeling construction module, a load forecasting generation module, a structured data construction module, a multi-objective coal blending optimization module, a scheduling strategy generation module, and a parameter reverse correction module.

[0043] The data perception and modeling module is used to collect historical operating data, real-time status data, and multi-source heterogeneous data of the power plant. It performs feature extraction and training through historical operating data and builds a time-series prediction model by combining deep learning structures.

[0044] The load forecasting generation module is used to input historical operating data and real-time status data into the time series forecasting model to generate target load forecasting results.

[0045] The structured data construction module is used to fuse the target load prediction results with multi-source heterogeneous data from power plants to construct a structured input data set.

[0046] The multi-objective coal blending optimization module is used to construct a multi-objective constrained coal blending and combustion model based on a structured input data set, and to perform coal type combination and allocation calculations.

[0047] The scheduling strategy generation module is used to execute the strategy simulation and scheduling task generation process based on the coal type combination and allocation calculation results to form strategy parameters.

[0048] The parameter reverse correction module is used to reverse correct the data acquisition configuration and prediction model parameter settings based on the strategy parameters.

[0049] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement a method for intelligent decision-making in power plant operation.

[0050] A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of an intelligent decision-making method for power plant operation.

[0051] The beneficial effects of the present invention are as follows: The intelligent decision-making method for power plant operation provided by the present invention collects historical operation data and real-time status data and combines them with a deep learning structure to build a time series prediction model. The present invention effectively overcomes the lag and instability problems of traditional empirical methods or simple regression models in load forecasting, thereby significantly improving the accuracy and timeliness of load forecasting.

[0052] By constructing a structured input data set that integrates target load forecasting results with multi-source heterogeneous data from power plants, this invention solves the problems of information fragmentation and insufficient basic input dimensions for scheduling in traditional data-driven strategies, and achieves the unity and completeness of scheduling input data.

[0053] By establishing a multi-objective constrained coal blending model that includes calorific value limits, cost functions, and equipment operating boundaries, compared with traditional blending strategies based on a single cost or calorific value index, this invention can simultaneously take into account multiple operating constraints, achieving the optimization of fuel utilization efficiency and the simultaneous guarantee of operational safety.

[0054] By generating a blending instruction sequence based on the allocation ratio and driving the strategy generation process to execute the scheduling task configuration, this invention improves upon the traditional one-way strategy generation and lack of a real-time response mechanism, and realizes dynamic linkage and real-time update capability from strategy generation to scheduling configuration.

[0055] By writing the control parameters in the scheduling execution table into the policy buffer and using them as the final scheduling input, this invention overcomes the problem that traditional policy results cannot directly drive scheduling execution, thus enabling a direct parameter mapping relationship between policy output and scheduling control.

[0056] By reversing the data acquisition logic and prediction model parameters based on strategy parameters, this invention introduces a closed-loop feedback optimization mechanism of data-model-scheduling, which differs from the existing static execution method after model deployment. This mechanism enables the invention to have adaptive and continuous optimization capabilities during long-term operation, and achieves better results in terms of prediction accuracy, resource allocation efficiency, and scheduling response flexibility. Attached Figure Description

[0057] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 The first embodiment of the present invention provides an overall flowchart of a smart decision-making method for power plant operation. Detailed Implementation

[0059] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0060] Example 1, referring to Figure 1 As an embodiment of the present invention, a smart decision-making method for power plant operation is provided, comprising:

[0061] S1: Collect historical operation data, real-time status data, and multi-source heterogeneous data from the power plant. Extract features and train the model using historical operation data, and build a time-series prediction model using a deep learning structure.

[0062] Historical operating data is acquired by setting up acquisition interfaces for the boiler main control system and auxiliary machine monitoring system, and real-time status data is acquired by setting up real-time monitoring nodes. Multi-source heterogeneous data of the power plant is extracted from the power plant management information system. The three types of data are uniformly encoded and written into the data cache area.

[0063] Historical operating data includes load changes, fuel usage, and equipment output records.

[0064] Real-time status data includes power generation, thermal load, steam flow rate, and flue gas temperature.

[0065] The power plant's multi-source heterogeneous data includes dispatch data, electricity price information, fuel inventory status, equipment maintenance plans, electricity prices, coal yard status, and equipment maintenance plans.

[0066] S2: Input historical operating data and real-time status data into the time series forecasting model to generate target load forecasting results.

[0067] Historical operational data is processed by sliding slices according to time windows, and time series feature fields are selected as input vectors based on data characteristics.

[0068] A preferred approach to obtaining the input vector composed of historical feature sequences is as follows:

[0069]

[0070] in, H represents the input vector composed of historical feature sequences. t-k This represents the running data at time tk, where T1 represents the sliding window length, t represents the current prediction time, and k represents the time interval step index.

[0071] The input vector is concatenated with real-time state data and then fed into the temporal modeling unit in the deep learning structure.

[0072] A preferred approach to concatenating the input vector with real-time state data is as follows:

[0073]

[0074] in, R represents the input vector of the final time series prediction model. t This represents the real-time state vector at the current moment, and Concat(·) represents the vector concatenation operation.

[0075] By using a multi-layered nested time recursive network, an output sequence reflecting future load change trends is generated.

[0076] A preferred approach to generating an output sequence that reflects future load change trends is as follows:

[0077]

[0078] in, This represents the predicted load at time tk. T1 represents the prediction function of the LSTM structure, and T2 represents the total number of future step sizes to be predicted.

[0079] The output sequence is bound to time stamps to form the target load prediction result.

[0080] Furthermore, to address the temporal dependencies and nonlinear relationships between different operating variables, a multi-layered deep learning prediction model is introduced for training.

[0081] A time recursive network is used to minimize the deviation between the predicted output and the actual load by setting a loss function.

[0082] A preferred approach to the deviation between the predicted output and the actual load is:

[0083]

[0084] in, This represents the predicted loss value. L represents the predicted load. t+k This indicates the actual load.

[0085] The model can automatically learn the trends of each feature dimension over time and their internal coupling relationships.

[0086] S3: Integrate the target load forecast results with multi-source heterogeneous data from power plants to construct a structured input data set.

[0087] The load time series in the target load forecast results are aligned with the multi-source heterogeneous data of the power plant according to the time dimension, and the data are processed by field mapping, dimension expansion and normalization.

[0088] Construct a unified data structure format to generate a set of structured input data containing predicted load, electricity price level, equipment status, and coal yard inventory information at each time point.

[0089] A preferred approach for constructing a unified data structure format is:

[0090]

[0091] Among them, S t Represents a structured input vector. E represents the predicted load. t Indicates the electricity price, D t C represents the device state vector. t This represents coal yard inventory information, and Norm(·) represents the normalization function.

[0092] The structured input data set is arranged in sequential order and serves as the unified input source for the multi-objective constrained coal blending and combustion model.

[0093] S4: Based on the structured input data set, construct a multi-objective constrained coal blending and combustion model, and perform coal type combination and allocation calculations.

[0094] Using the predicted load, electricity price level, equipment status, and coal yard inventory information from the structured input dataset as input, scheduling constraints are established.

[0095] A feasibility constraint matrix is ​​established based on the type of coal to be blended, and the corresponding coal blending ratio variables are defined.

[0096] One preferred approach to defining the corresponding coal blending ratio variable is as follows:

[0097]

[0098] Where, α t This represents the coal blending ratio vector at time t. Let represent the proportion of the i-th type of coal at time t, where i represents the number of coal types.

[0099] By constructing an objective function, the fuel allocation problem is transformed into a multi-objective optimization problem. The structured dataset is input into the multi-objective constrained coal blending model to obtain the allocation ratio of each type of coal.

[0100] A preferred approach to transforming the fuel allocation problem into a multi-objective optimization problem by constructing an objective function is as follows:

[0101]

[0102] Among them, C burn P represents the unit cost of blending and firing. (i) This represents the unit price of the i-th type of coal.

[0103] An optimal scheme for obtaining the allocation ratio of each type of coal is as follows:

[0104]

[0105] Among them, J total Q represents the overall optimization objective value, λ1, λ2, and λ3 represent weighting coefficients, and Q represents the weighting coefficients. target R represents the target calorific value. avail This represents the resource balance factor.

[0106] The objective function includes the calorific value balance constraint, the cost weighting function, and the output feasibility function.

[0107] A preferred approach to constructing the objective function is as follows:

[0108]

[0109] Among them, Q (i) Let Q represent the calorific value of the i-th type of coal. min This represents the minimum calorific value constraint, N represents the total number of currently selectable coal types, and i represents the index item number.

[0110] Furthermore, the allocation ratio of each type of coal is combined with the current coal yard inventory data to calculate the actual quantity of coal to be allocated.

[0111] A preferred method for calculating the actual allocation quantity of coal is as follows:

[0112]

[0113] in, M represents the actual blending quality of the i-th type of coal. total This indicates the total allocation amount.

[0114] Establish a coal allocation plan and generate a sequence of co-firing instructions for the boilers based on the allocation results.

[0115] The instruction sequence serves as input data for policy simulation and is used for configuration calculations in the policy generation process.

[0116] S5: Based on the coal type combination and allocation calculation results, execute the strategy simulation and scheduling task generation process to form strategy parameters.

[0117] The coal type combination and allocation calculation results are input into the strategy generation process, and the corresponding scheduling task execution table is configured based on equipment constraints, electricity price constraints and fuel blending parameters.

[0118] Extract control parameters based on the task items generated in the scheduling execution table.

[0119] Control parameters include equipment start / stop signals, fuel adjustment commands, and load distribution sequences.

[0120] The control parameters are encoded as policy parameters, and the policy parameters are written into the policy buffer as the final scheduling input.

[0121] A preferred approach to encoding control parameters as policy parameters is as follows:

[0122]

[0123] Among them, U t M represents the strategy control parameters. t D represents the model parameter state vector at time t. t This represents the vector of perturbation parameters at time t. Let t represent the load adjustment parameter vector at time t, and Encode([·]) represent the strategy parameter encoding function.

[0124] The function expression for the strategy parameter encoding function is specifically represented as follows:

[0125]

[0126] Among them, b eW represents the policy encoding bias vector. e This represents the policy encoding weight matrix.

[0127] S6: Based on strategy parameters, reverse the data acquisition configuration and prediction model parameter settings.

[0128] In the process of performing reverse correction based on policy parameters, it is necessary to clarify the causal relationship between policy execution deviation and data configuration.

[0129] After the strategy parameters are generated, a two-layer interactive judgment mechanism is introduced to realize differentiated correction path control of data acquisition configuration and prediction model parameters, ensuring the stability and adaptability of the closed-loop operation of the scheduling chain.

[0130] Furthermore, the actual scheduling results are first collected and the execution residual is calculated with the strategy target output. A residual threshold is set. If the execution residual value is greater than the set residual value, a secondary judgment process is triggered.

[0131] A preferred approach is to first collect the actual scheduling results and calculate the execution residuals using the strategy target output:

[0132]

[0133] in, Indicates the actual load after scheduling. Indicates the actual amount of coal used, ∈ t This represents the residual signal at the current time t.

[0134] A preferred approach is to set a residual threshold, and trigger a secondary judgment process if the executed residual value exceeds the set residual value.

[0135]

[0136] Where, δ t Indicates the correction trigger flag, ∈ th Represents the residual trigger threshold, ∈ t This indicates the execution residual value.

[0137] After the residual is triggered, a judgment matrix is ​​constructed based on the source of error for further detailed judgment.

[0138] A preferred approach for segmenting and judging based on the source of error is:

[0139]

[0140] Among them, J tLet γ represent the judgment matrix, where Delay(L) represents the load response delay, ActErr(U) represents the equipment deviation, FailRate(M) represents the dispatch failure rate, γ1 represents the upper limit of delay tolerance, γ2 represents the upper limit of action error allowable, and γ3 represents the failure rate threshold.

[0141] If the load response delay is greater than the delay response threshold, it is inferred that the prediction model is lagging. Adjust the values ​​of some parameters in the model parameter set to improve the model's ability to fit the actual load or scheduling response.

[0142] If the equipment start-up and shutdown response deviates abnormally, it is determined that the control signal accuracy is too low, and the control parameter structure and task frequency are adjusted.

[0143] A preferred approach for adjusting the control parameter structure and task frequency is as follows:

[0144] E t =EncErr(∈ t ,δ t J t )

[0145] Among them, E t This represents the error signal written to the buffer, and EncErr(·) represents the error encoding function.

[0146] The specific expression of the error coding function is as follows:

[0147] EncErr(∈ t ,δ t J t ) = tanh(W ∈ ·∈ t +W δ ·δ t +W J ·J t +b e )

[0148] Where tanh represents the hyperbolic tangent function, W ∈ Represents the weights assigned to the residual signal, W δ This indicates that the trigger flag allocation weight has been corrected, W. J This indicates the weights assigned to the judgment matrix.

[0149] If the failure rate of coal type allocation increases, it indicates that the granularity of the collected data is insufficient. The sampling cycle should be adjusted and the number of data nodes increased.

[0150] A preferred solution for adjusting the sampling period and increasing the number of data nodes is:

[0151] S t+1 =S t +φ(E t )

[0152] Among them, S t+1 S represents the modified sampling strategy. t φ(·) represents the current sampling strategy, and φ(·) represents the sampling strategy update function.

[0153] φ(Et)=η s ·E t

[0154] Where, η s This indicates the sampling strategy update step size, which is set manually.

[0155] Example 2, an embodiment of the present invention, provides an intelligent decision-making system for power plant operation, including a data perception and modeling construction module, a load forecasting generation module, a structured data construction module, a multi-objective coal blending optimization module, a scheduling strategy generation module, and a parameter reverse correction module.

[0156] The data perception and modeling module is used to collect historical operating data, real-time status data, and multi-source heterogeneous data from the power plant. It extracts and trains features from historical operating data and builds a time-series prediction model by combining it with a deep learning structure.

[0157] The load forecasting generation module is used to input historical operating data and real-time status data into the time series forecasting model to generate target load forecasting results.

[0158] The structured data construction module is used to fuse target load forecast results with multi-source heterogeneous data from power plants to construct a structured input data set.

[0159] The multi-objective coal blending optimization module is used to construct a multi-objective constrained coal blending and combustion model based on a structured input data set, and to perform coal type combination and allocation calculations.

[0160] The scheduling strategy generation module is used to execute the strategy simulation and scheduling task generation process based on the coal type combination and allocation calculation results, and form strategy parameters.

[0161] The parameter reverse correction module is used to reverse correct the data acquisition configuration and prediction model parameter settings based on the strategy parameters.

Claims

1. A smart decision-making method for power plant operation, characterized in that, include: Collect historical operation data, real-time status data, and multi-source heterogeneous data from the power plant. Extract features and train the model using historical operation data, and build a time-series prediction model by combining it with a deep learning structure. Input historical operating data and real-time status data into the time series forecasting model to generate target load forecasting results; The target load forecast results are fused with multi-source heterogeneous data from power plants to construct a structured input data set; Based on a structured input dataset, a multi-objective constrained coal blending and combustion model is constructed, and coal type combination and allocation calculations are performed. Based on the coal type combination and allocation calculation results, execute the strategy simulation and scheduling task generation process to form strategy parameters; Based on the strategy parameters, the data acquisition configuration and prediction model parameter settings are corrected in reverse.

2. The intelligent decision-making method for power plant operation as described in claim 1, characterized in that: The collected historical operating data, real-time status data, and multi-source heterogeneous data of the power plant include... Historical operating data is acquired by setting up acquisition interfaces for the boiler main control system and auxiliary machine monitoring system, and real-time status data is acquired by setting up real-time monitoring nodes. Multi-source heterogeneous data of the power plant is extracted from the power plant management information system. The three types of data are uniformly encoded and written into the data cache area. Historical operating data includes load changes, fuel usage, and equipment output records; Real-time status data includes power generation, thermal load, steam flow rate, and flue gas temperature; The power plant's multi-source heterogeneous data includes dispatch data, electricity price information, fuel inventory status, equipment maintenance plans, electricity prices, coal yard status, and equipment maintenance plans.

3. The intelligent decision-making method for power plant operation as described in claim 2, characterized in that: The generated target load forecast results include, Historical operational data is processed by sliding slices according to time windows, and time series feature fields are selected as input vectors based on data characteristics; The input vector is concatenated with real-time state data and then fed into the temporal modeling unit in the deep learning structure. By processing through a multi-layered nested time recursive network, an output sequence reflecting future load change trends is generated. The output sequence is bound to time stamps to form the target load prediction result.

4. The intelligent decision-making method for power plant operation as described in claim 3, characterized in that: The construction of the structured input data set includes, Align the load time series in the target load forecast results with the multi-source heterogeneous data of the power plant by time dimension, and perform field mapping, dimension expansion and normalization on the data; Construct a unified data structure format to generate a set of structured input data containing predicted load, electricity price level, equipment status, and coal yard inventory information for each moment; The structured input data set is arranged in sequential order and serves as the unified input source for the multi-objective constrained coal blending and combustion model.

5. The intelligent decision-making method for power plant operation as described in claim 4, characterized in that: The construction of the multi-objective constrained coal blending and combustion model includes... Using the predicted load, electricity price level, equipment status, and coal yard inventory information from the structured input dataset as input, scheduling constraints are established. Establish a feasibility constraint matrix based on the types of coal to be blended, and define the corresponding coal blending ratio variables; The fuel allocation problem is transformed into a multi-objective optimization problem by constructing an objective function. The structured dataset is input into the multi-objective constrained coal blending model to obtain the allocation ratio of each type of coal. The objective function includes the calorific value balance constraint, the cost weighting function, and the output feasibility function.

6. The intelligent decision-making method for power plant operation as described in claim 5, characterized in that: The calculation of coal type combination and allocation includes, The allocation ratio of each type of coal is combined with the current coal yard inventory data to calculate the actual quantity of coal to be allocated. Establish a coal type allocation plan and generate a sequence of co-firing instructions for the boilers based on the allocation results; The instruction sequence serves as input data for policy simulation and is used for configuration calculations in the policy generation process.

7. The intelligent decision-making method for power plant operation as described in claim 6, characterized in that: The formation strategy parameters include, Input the coal type combination and allocation calculation results into the strategy generation process, and configure the corresponding scheduling task execution table based on equipment constraints, electricity price constraints and fuel blending parameters; Extract control parameters based on the task items generated in the scheduling execution table; Control parameters include equipment start / stop signals, fuel adjustment commands, and load distribution sequences; The control parameters are encoded as policy parameters, and the policy parameters are written into the policy buffer as the final scheduling input. The reverse correction data acquisition configuration and prediction model parameter settings include, A two-layer interactive judgment mechanism was introduced after the strategy parameters were generated to enable the configuration of data collection and the setting of parameters for the prediction model. The two-layer interactive judgment mechanism includes: collecting the actual scheduling results and calculating the execution residual with the policy target output; setting a residual threshold; and triggering a second-level judgment process if the execution residual value is greater than the set residual value. After the residual is triggered, a detailed judgment is made based on the source of error, and a judgment matrix is ​​constructed. If the load response delay is greater than the delay response threshold, it is inferred that the prediction model is lagging, and the model parameter set is adjusted. If the equipment start-up and shutdown response deviation is abnormal, it is determined that the control signal accuracy is too low, and the control parameter structure and task frequency are adjusted. If the failure rate of coal type allocation increases, it indicates that the granularity of the collected data is insufficient. The sampling cycle should be adjusted and the number of data nodes increased.

8. A system employing the intelligent decision-making method for power plant operation as described in any one of claims 1 to 7, characterized in that: It includes a data perception and modeling module, a load forecasting generation module, a structured data construction module, a multi-objective coal blending optimization module, a scheduling strategy generation module, and a parameter reverse correction module; The data perception and modeling module is used to collect historical operating data, real-time status data and multi-source heterogeneous data of the power plant. It performs feature extraction and training through historical operating data and builds a time-series prediction model by combining deep learning structures. The load forecasting generation module is used to input historical operating data and real-time status data into the time series forecasting model to generate target load forecasting results. The structured data construction module is used to fuse the target load forecast results with multi-source heterogeneous data from power plants to construct a structured input data set. The multi-objective coal blending optimization module is used to construct a multi-objective constrained coal blending and combustion model based on a structured input data set, and to perform coal type combination and allocation calculations. The scheduling strategy generation module is used to execute the strategy simulation and scheduling task generation process based on the coal type combination and allocation calculation results to form strategy parameters. The parameter reverse correction module is used to reverse correct the data acquisition configuration and prediction model parameter settings based on the strategy parameters.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent decision-making method for power plant operation as described in any one of claims 1 to 7.

10. A 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 steps of the intelligent decision-making method for power plant operation as described in any one of claims 1 to 7.

Images