A heat system energy efficiency index real-time optimization method, system, device and medium fusing target working condition library and prior knowledge

Abstract

The application discloses a kind of fusion target working condition library and prior knowledge heat system energy efficiency index instant optimization method, system, equipment and medium, comprising the following steps: determining the target parameter, relevant parameter, boundary parameter and characteristic variable required;Collect target parameter, relevant parameter, boundary parameter and characteristic variable in the historical operation data of certain year heat system unit in a year, and form target working condition library by construction, and filter optimal working condition sample;Form simulation data working condition library by construction;In the process of unit operation, real-time acquisition obtains boundary parameter, relevant parameter and target parameter, and determines the grid unit of current working condition, carries out differentiation optimization according to target working condition library and simulation data working condition library, and outputs optimal target parameter and optimal relevant parameter in the grid unit of current working condition;Target working condition library is dynamically updated.The application realizes the high-precision instant optimization of energy efficiency index and the accurate guidance of operating parameter.

Description

Technical Field

[0001] This invention belongs to the field of thermal system technology, and particularly relates to a method, system, equipment and medium for real-time optimization of thermal system energy efficiency indicators by integrating a target operating condition library and prior knowledge. Background Technology

[0002] In the energy and power sector, thermal systems (including thermal power generating units, industrial boilers, and steam turbine systems) serve as the core carriers of energy conversion and utilization, and their energy efficiency directly determines energy utilization efficiency and carbon emission intensity. Thermal systems are characterized by strong coupling of multiple parameters, dynamic and varied operating conditions, and complex energy conversion processes. Energy efficiency indicators such as power generation coal consumption rate, boiler efficiency, and steam turbine heat rate are directly affected by operating parameters such as main steam pressure and condenser vacuum, and are also closely related to boundary parameters such as unit load and ambient temperature. Furthermore, the dynamic coupling relationship of various parameters further increases the difficulty of energy efficiency optimization during load increases and decreases and variable operating conditions. However, existing energy efficiency optimization technologies for thermal systems still have significant limitations and are insufficient to meet actual industrial needs.

[0003] The optimization method, which constructs a system mechanism model based on thermodynamic laws and energy conservation equations and obtains the theoretical optimal operating condition by solving the model, can reflect the physical essence of energy conversion and has strong interpretability. However, it has two shortcomings: First, the thermodynamic system has a complex structure (involving subsystems such as boiler combustion, steam-water circulation, and turbine work). The construction of a complete mechanism model requires accurate understanding of the characteristic parameters and coupling relationships of each component, resulting in a long modeling cycle and high maintenance costs. Second, in actual operation, factors such as sensor errors and equipment aging can cause deviations between the model and the real system. Especially in dynamic and variable operating condition scenarios, the model calculation delay is large, making it difficult to achieve "real-time optimization" and providing insufficient real-time guidance for the optimization results.

[0004] Optimization methods that leverage data mining and machine learning algorithms to extract relationships between energy efficiency indicators and related parameters from historical operating data to determine the optimal operating range offer high flexibility without requiring complex mechanistic modeling. However, they suffer from two major drawbacks: First, they lack a systematic division of operating conditions. When historical data is sparsely distributed across operating conditions such as low loads and extreme ambient temperatures, "data silos" easily emerge, resulting in weak generalization ability of the optimization results. Second, they do not incorporate prior knowledge of the thermodynamic system. Relying solely on actual data makes them susceptible to noise interference. Under scenarios of drastic changes in operating conditions, such as sudden increases or decreases in load, the reliability of optimization significantly decreases, and results that contradict thermodynamic laws may even occur.

[0005] In summary, existing technologies for optimizing energy efficiency indicators in thermal systems generally suffer from poor real-time performance, weak generalization ability in areas with scarce samples, insufficient physical consistency, and a lack of dynamic adaptability. More critically, existing research largely focuses on optimizing a single energy efficiency indicator, failing to establish a closed-loop decision-making mechanism from precise optimization to guidance from operating parameters. Traditional optimization methods face fundamental bottlenecks: First, there is a strong coupling conflict between optimal energy efficiency indicators and the operability of operating parameters; relying on simple combinations of experience for selection cannot dynamically generate optimal parameters suitable for actual applications. Second, optimization based on a single data source or model struggles to consider both physical laws and actual operating characteristics, leading to deviations in optimization results under complex operating conditions. Third, due to the lack of a dynamically updated optimization mechanism, factors such as increased unit operating time and equipment aging can easily cause optimization results to deviate from reality. The dual constraints of insufficient model accuracy and defects in optimization methods result in deviations of over 15% between theoretically optimal energy efficiency and actual operation. Summary of the Invention

[0006] Purpose of the invention: The purpose of this invention is to provide a method, system, device and medium for real-time optimization of energy efficiency indicators of thermal systems that integrates a target operating condition library and prior knowledge, so as to achieve high-precision real-time optimization of energy efficiency indicators and accurate guidance of operating parameters, fundamentally solving the problem of the difficulty in balancing real-time performance and accuracy in energy efficiency optimization of thermal systems.

[0007] Technical Solution: To achieve the above objectives, this invention discloses a method for real-time optimization of energy efficiency indicators of a thermal system that integrates a target operating condition database and prior knowledge, comprising the following steps: S1. Determine the required target parameters, relevant parameters, boundary parameters, and characteristic variables; S2. Collect target parameters, related parameters, boundary parameters, and characteristic variables from the historical operating data of the thermal system units throughout a given year. Select 2 or 3 boundary parameters as coordinate system dimensions to construct a grid coordinate system covering the entire operating condition range. Use the sliding window method to sample the historical operating data to form several sample operating conditions. Map each sample operating condition to a grid cell in the grid coordinate system according to its boundary parameters to construct a target operating condition library and select the optimal operating condition sample in each grid cell. S3. Construct a modular simulation model of the thermal system, build a grid coordinate system consistent with the target operating condition library, and divide each grid cell into multiple grids evenly. The boundary parameter values ​​are generated at the intersection of the grids. Input the boundary parameter values ​​at the intersections into the modular simulation model. The modular simulation model solves the boundary parameter values ​​at each intersection to obtain the corresponding target parameters, thereby generating a simulation dataset that corresponds one-to-one with the target operating condition library and constructing a simulation data operating condition library. S4. During unit operation, boundary parameters, relevant parameters and target parameters are collected in real time, and the boundary parameters are mapped to the grid coordinate system of the target operating condition library to determine the grid cell of the current operating condition. The number of operating condition samples in the grid cell of the current operating condition is determined. Differential optimization is performed based on the target operating condition library and the simulation data operating condition library, and the optimal target parameter and optimal relevant parameter in the grid cell of the current operating condition are output. S5. Dynamically update the target operating condition library.

[0008] Optionally, in step S1, the target parameter refers to the core indicator used to evaluate the energy consumption of the thermal system or the energy efficiency of the equipment; the relevant parameter refers to the working fluid parameter that can be manually adjusted to control the operating state of the unit; the boundary parameter refers to the parameter that is affected by external objective conditions and cannot be directly controlled by the thermal system; the characteristic variable refers to the parameter that can characterize the steady-state operation of the system, and the characteristic variables are selected as load, main steam pressure, main steam temperature, reheat steam temperature and feedwater flow rate.

[0009] Optionally, step S2 specifically includes the following steps: S21. Collect target parameters, related parameters, boundary parameters, and characteristic variables from the historical operating data of the thermal system units throughout a year. Select 2 or 3 boundary parameters as coordinate system dimensions. Based on all boundary parameter data throughout the year, determine the extreme values ​​of the boundary parameters. The scale range of each coordinate axis is the extreme value interval. Set a fixed step size for each coordinate system dimension. Divide the coordinate axes with the fixed step size to form a grid coordinate system covering the entire operating condition range. Each grid cell corresponds to a discretized operating condition. S22. The sliding window method is used to extract historical operating data of the thermal system units throughout the year, generating an overlapping set of data windows. Each data window contains the target parameter, relevant parameters, boundary parameters, and feature variables for a continuous time period. The target parameter, relevant parameters, boundary parameters, and feature variables of each sampling point in each data window constitute each sample operating condition. Each sample operating condition is mapped to the corresponding grid cell according to the value of its boundary parameter, and the statistical characteristics of each grid cell are output. The statistical characteristics include cell coordinates, number of samples, maximum value of the target parameter, minimum value of the target parameter, average value of the target parameter, and scarcity factor, where the scarcity factor refers to the preset sample number threshold in the grid cell. S23. After cleaning and filling all sample working condition data in the grid cell, the optimal working condition sample is screened for the sample working conditions in each grid cell. Specifically, the steady-state screening is first performed on the sample working conditions in the grid cell. That is, when the difference between the extreme values ​​of the characteristic variables of the sample is lower than the preset threshold, the corresponding sample belongs to the steady-state sample. Then, the optimization criteria of the target parameter are determined based on empirical knowledge. The single sample data with the best performance of the target parameter is selected from the samples that have passed the steady-state screening as the optimal working condition sample of the grid cell. S24. Output the target working condition library and the optimal working condition sample for each grid cell.

[0010] Optionally, step S4 specifically includes the following steps: S41. Compare the number of working condition samples in the grid cell of the current working condition in the target working condition library with the preset scarcity factor. When the number of working condition samples is not less than the scarcity factor, the grid cell of the current working condition is a normal cell. When the number of working condition samples is less than the scarcity factor, the grid cell of the current working condition is a scarce cell. S42. For normal cells, extract the historical best value of the target parameter within the grid cell of the current working condition in the target working condition library, and compare the historical best value of the target parameter with the real-time value of the target parameter of the current working condition; if the real-time value of the target parameter is better, update the real-time value in the target working condition library to the best value, and output the real-time value of the target parameter; otherwise, directly output the corresponding historical best value of the target parameter in the target working condition library. S43. For scarce cells, take the scarce cells in the target working condition library as the center and gradually expand the neighborhood outward until the total number of working condition samples of all grid cells in the expanded neighborhood is not less than the scarcity factor, and denote the working condition samples of all grid cells in the expanded neighborhood as the local sample set D1. From the simulation data condition library, extract the simulation dataset D2 corresponding to the extended neighborhood. Using simulation dataset D2 as the training set, construct the basic prediction model using the support vector machine (SVM) algorithm. The input of the basic prediction model is the boundary parameter values, and the output of the basic prediction model is the predicted value of the target parameter. A modified network model is added to the basic prediction model, and the deviation between the predicted value of the target parameter of the basic prediction model and the true value of the local sample set D1 is learned and corrected using the local sample set D1 as training data, forming a final local model that integrates simulation data and real data; the current real-time boundary parameters are input into the final local model, and the output is the optimal value of the predicted target parameter of the scarce unit. S44. For normal units, extract the relevant parameter values ​​corresponding to the optimal working condition sample from the target working condition library as the optimal relevant parameters; S45. For scarce cells, take the scarce cells in the target working condition library as the center and gradually expand the neighborhood outward until the total number of working condition samples of all grid cells in the expanded neighborhood is not less than the scarcity factor, and denote the working condition samples of all grid cells in the expanded neighborhood as the local sample set D1. From the simulation data condition library, the simulation dataset D2 corresponding to the extended neighborhood is extracted. Using the simulation dataset D2 as the training set, the basic prediction model is constructed using the support vector machine (SVM) algorithm. The input of the basic prediction model is the boundary parameter values, and the output of the basic prediction model is the predicted values ​​of the relevant parameters. A modified network model is added to the basic prediction model, and the deviation between the predicted values ​​of the relevant parameters of the basic prediction model and the true values ​​of the local sample set D1 is learned and corrected using the local sample set D1 as training data, forming a final local model that integrates simulation data and real data. The current real-time boundary parameters are input into the final local model, and the output is the optimal value of the predicted relevant parameters of the scarce unit.

[0011] The present invention discloses a real-time optimization system for energy efficiency indicators of a thermal system that integrates a target operating condition database and prior knowledge, comprising: The parameter selection module is used to determine the required target parameters, relevant parameters, boundary parameters, and feature variables; The target operating condition library construction module is used to collect target parameters, related parameters, boundary parameters, and feature variables from the historical operating data of the thermal system units throughout a year. It selects two or three boundary parameters as coordinate system dimensions to construct a grid coordinate system covering the entire operating condition range. It uses the sliding window method to sample the historical operating data to form several sample operating conditions. Each sample operating condition is mapped to a grid cell in the grid coordinate system according to its boundary parameters to construct the target operating condition library and select the optimal operating condition sample in each grid cell. The simulation data condition library construction module is used to build a modular simulation model of the thermal system, construct a grid coordinate system consistent with the target condition library, and divide each grid cell into multiple grids evenly. Boundary parameter values ​​are generated at the intersection points of the grids. The boundary parameter values ​​at the intersection points are input into the modular simulation model. The modular simulation model solves for the boundary parameter values ​​at each intersection point to obtain the corresponding target parameters, thereby generating a simulation dataset that corresponds one-to-one with the target condition library and constructing the simulation data condition library. The parameter optimization module is used to collect boundary parameters, relevant parameters, and target parameters in real time during unit operation. It maps the boundary parameters to the grid coordinate system of the target operating condition library to determine the grid cell of the current operating condition, determines the number of operating condition samples in the grid cell of the current operating condition, performs differentiated optimization based on the target operating condition library and the simulation data operating condition library, and outputs the optimal target parameter and optimal relevant parameter in the grid cell of the current operating condition.

[0012] Optionally, in the parameter selection module, the target parameter refers to the core indicator used to evaluate the energy consumption of the thermal system or the energy efficiency of the equipment; the relevant parameter refers to the working fluid parameter that can be manually adjusted to control the operating state of the unit; the boundary parameter refers to the parameter that is affected by external objective conditions and cannot be directly controlled by the thermal system; the characteristic variable refers to the parameter that can characterize the steady-state operation of the system, and the characteristic variables are selected as load, main steam pressure, main steam temperature, reheat steam temperature and feedwater flow rate.

[0013] Optionally, the target operating condition library construction module collects target parameters, related parameters, boundary parameters, and feature variables from the historical operating data of the thermal system units throughout a certain year. It selects two or three boundary parameters as coordinate system dimensions, determines the extreme values ​​of the boundary parameters based on all boundary parameter data throughout the year, and sets the scale range of each coordinate axis as the extreme value interval. It sets a fixed step size for each coordinate system dimension and divides the coordinate axes with the fixed step size to form a grid coordinate system covering the entire operating condition range. Each grid cell corresponds to a discretized operating condition. A sliding window method is used to extract historical operating data of the thermal system units throughout the year, generating an overlapping set of data windows. Each data window contains target parameters, relevant parameters, boundary parameters, and characteristic variables for a continuous time period. The target parameters, relevant parameters, boundary parameters, and characteristic variables of each sampling point within each data window constitute each sample operating condition. Each sample operating condition is mapped to a corresponding grid cell according to the values ​​of its boundary parameters, and the statistical characteristics of each grid cell are output. The statistical characteristics include cell coordinates, number of samples, maximum value of target parameter, minimum value of target parameter, average value of target parameter, and scarcity factor, where the scarcity factor refers to the preset sample number threshold within the grid cell. After cleaning and filling all sample working condition data within the grid cell, the optimal working condition sample is screened for each grid cell. Specifically, the steady-state screening is first performed on the sample working conditions within the grid cell. That is, when the difference between the extreme values ​​of the characteristic variables of the sample is lower than the preset threshold, the corresponding sample belongs to the steady-state sample. Then, the optimization criteria of the target parameter are determined based on empirical knowledge. The single sample data with the best performance of the target parameter is selected from the samples that have passed the steady-state screening as the optimal working condition sample of the grid cell. Output the target working condition library and the optimal working condition sample for each grid cell.

[0014] Optionally, the parameter optimization module compares the number of working condition samples in the grid cell of the current working condition in the target working condition library with a preset scarcity factor. When the number of working condition samples is not less than the scarcity factor, the grid cell of the current working condition is a normal cell. When the number of working condition samples is less than the scarcity factor, the grid cell of the current working condition is a scarce cell. For normal cells, extract the historical best value of the target parameter within the grid cell of the current working condition from the target working condition library, and compare the historical best value of the target parameter with the real-time value of the target parameter of the current working condition; if the real-time value of the target parameter is better, update the real-time value in the target working condition library to the best value, and output the real-time value of the target parameter; otherwise, directly output the corresponding historical best value of the target parameter in the target working condition library. For scarce cells, take the scarce cells in the target working condition library as the center and gradually expand the neighborhood outward until the total number of working condition samples of all grid cells in the expanded neighborhood is not less than the scarcity factor, and denote the working condition samples of all grid cells in the expanded neighborhood as the local sample set D1. From the simulation data condition library, extract the simulation dataset D2 corresponding to the extended neighborhood. Using simulation dataset D2 as the training set, construct the basic prediction model using the support vector machine (SVM) algorithm. The input of the basic prediction model is the boundary parameter values, and the output of the basic prediction model is the predicted value of the target parameter. A modified network model is added to the basic prediction model, and the deviation between the predicted value of the target parameter of the basic prediction model and the true value of the local sample set D1 is learned and corrected using the local sample set D1 as training data, forming a final local model that integrates simulation data and real data; the current real-time boundary parameters are input into the final local model, and the output is the optimal value of the predicted target parameter of the scarce unit. For normal units, the relevant parameter values ​​corresponding to the optimal working condition sample are extracted from the target working condition library as the optimal relevant parameters. For scarce cells, take the scarce cells in the target working condition library as the center and gradually expand the neighborhood outward until the total number of working condition samples of all grid cells in the expanded neighborhood is not less than the scarcity factor, and denote the working condition samples of all grid cells in the expanded neighborhood as the local sample set D1. From the simulation data condition library, the simulation dataset D2 corresponding to the extended neighborhood is extracted. Using the simulation dataset D2 as the training set, the basic prediction model is constructed using the support vector machine (SVM) algorithm. The input of the basic prediction model is the boundary parameter values, and the output of the basic prediction model is the predicted values ​​of the relevant parameters. A modified network model is added to the basic prediction model, and the deviation between the predicted values ​​of the relevant parameters of the basic prediction model and the true values ​​of the local sample set D1 is learned and corrected using the local sample set D1 as training data, forming a final local model that integrates simulation data and real data. The current real-time boundary parameters are input into the final local model, and the output is the optimal value of the predicted relevant parameters of the scarce unit.

[0015] The electronic device of the present invention includes a processor and a storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method described above.

[0016] The present invention discloses a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the steps of the method described above.

[0017] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: (1) This invention provides an instant optimization method that integrates target operating condition library and prior knowledge and has dynamic adaptability. By systematically constructing operating condition units, integrating historical data and simulation data, and dynamically updating the database, it realizes high-precision instant optimization of energy efficiency indicators and accurate guidance of operating parameters, fundamentally solving the problem that it is difficult to balance the real-time performance and accuracy of energy efficiency optimization in thermal systems, and providing reliable support for the efficient operation of thermal systems. (2) The present invention aims to solve the problems in the prior art such as inaccurate identification of operating conditions caused by continuous changes in the operating conditions of thermal power units, modeling blind spots caused by uneven distribution of historical samples, difficulty in obtaining the optimal value of scarce operating conditions, and non-executable optimization results. The present invention constructs a target operating condition library covering the entire operating space, combines prior knowledge of mechanism simulation to complete the data of scarce operating conditions, and uses a local high-precision prediction model that integrates simulation and real data to realize real-time prediction of energy efficiency indicators, judgment of optimal values, and matching output of controllable related parameters of the system under any operating conditions. Thus, it provides the unit with an implementable basis for operation and adjustment, and improves the energy efficiency and operating economy of the thermal system under different operating conditions. (3) This invention realizes the discretization and structured expression of continuous working conditions space, which significantly improves the accuracy of working condition identification. Specifically, it constructs a two-dimensional or multi-dimensional grid coordinate system based on boundary parameters, maps complex continuous working conditions into computable discrete working condition units, enables the system to accurately locate the current operating condition, and realizes rapid matching of working condition status, solving the problem that working conditions are difficult to quantify and difficult to express uniformly in traditional methods. (4) This invention comprehensively compensates for insufficient historical data coverage and improves the reliability of modeling under scarce operating conditions. After identifying scarce sample units, this invention introduces a simulation operating condition library that corresponds one-to-one with the structure of the operating condition library. The scarce area is supplemented by a high-fidelity mechanism simulation model, so that the coverage of the operating space changes from "partially sparse" to "fully available". This mechanism effectively solves the problem of modeling blind spots caused by uneven distribution of historical data. (5) This invention improves prediction accuracy and generalization ability through the fusion of "simulation + real data" local modeling; this invention proposes a local modeling method of "data expansion - simulation assistance - model correction - optimal prediction", which uses simulation data to train a trend-based basic model and then uses real samples to correct the prediction deviation, so that the model has both the comprehensiveness of simulation and the accuracy of real data; this prediction model significantly improves the prediction accuracy of target parameters under scarce working conditions and avoids the limitations of pure simulation or pure data-driven methods; (6) This invention realizes differentiated optimization between normal units and scarce units, and improves the reliability of optimization results. For normal units with sufficient samples, this invention directly outputs the historical best sample to achieve efficient online optimization. For scarce units, the optimal value is predicted by a local high-precision model and compared with the historical best value to ensure that the output result is selected from the "historical reliable optimal value" and the "model predicted optimal value", so that the optimization process has stability and consistency under different working conditions. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the target working condition library in this invention; Figure 2 This is a schematic diagram of the target parameter optimization in this invention; Figure 3 This is a schematic diagram illustrating the optimization of relevant parameters in this invention. Detailed Implementation

[0019] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0020] Example 1: A method for real-time optimization of energy efficiency indicators of a thermal system that integrates a target operating condition database and prior knowledge according to the present invention includes the following steps: S1. Determine the required target parameters, relevant parameters, boundary parameters, and characteristic variables; The target parameters refer to the core indicators used to evaluate the energy consumption or equipment efficiency of the thermal system, such as turbine heat rate, power generation coal consumption rate, and boiler efficiency. Relevant parameters refer to working fluid parameters that can be manually adjusted to control the unit's operating state; the adjustment level of these parameters directly affects the economics of the thermal system, such as main steam temperature and pressure, reheat temperature and pressure, and flue gas temperature. Boundary parameters refer to parameters that are affected by external objective conditions and cannot be directly controlled by the thermal system, such as ambient temperature, coal quality characteristics, unit load, and load factor. Characteristic variables refer to parameters that characterize the steady-state operation of the system; the characteristic variables selected are load, main steam pressure, main steam temperature, reheat steam temperature, and feedwater flow rate. In this example, referring to the ASME PTC 6-2004 standard for turbine performance testing issued by the American Society of Mechanical Engineers in 2004, five characteristic variables were selected for steady-state screening: load, main steam pressure, main steam temperature, reheat steam temperature, and feedwater flow rate. When the difference between the extreme values ​​of the characteristic variables is lower than a preset threshold, the data in this example is considered to be steady-state data.

[0021] S2. Collect target parameters, related parameters, boundary parameters, and characteristic variables from the historical operating data of the thermal system units for a given year (2024). Select 2 or 3 boundary parameters as coordinate system dimensions to construct a grid coordinate system covering the entire operating condition range. Use the sliding window method to sample the historical operating data to form several sample operating conditions. Map each sample operating condition to a grid cell in the grid coordinate system according to its boundary parameters to construct a target operating condition library and select the optimal operating condition sample in each grid cell. Step S2 specifically includes the following steps: S21. Collect target parameters, related parameters, boundary parameters, and characteristic variables from the historical operating data of the thermal system units throughout a given year. Select 2 or 3 boundary parameters as coordinate system dimensions. Based on all boundary parameter data for the year, determine the extreme values ​​of the boundary parameters. The scale range of each coordinate axis is the extreme value interval. Set a fixed step size for each coordinate system dimension. Divide the coordinate axes with the fixed step size to form a grid coordinate system covering the entire operating condition range. Each grid cell corresponds to a discretized operating condition.

[0022] like Figure 1 As shown, taking a two-dimensional operating space as an example, boundary parameter P1 (e.g., unit load, unit: MW) and boundary parameter P2 (e.g., ambient temperature, unit: °C) are selected as the dimensions of the coordinate system. A fixed step size is set for each dimension. For example, a step size ∆1 (e.g., 10MW) is set for the P1 axis, and a step size ∆2 (e.g., 2°C) is set for the P2 axis. The coordinate axes are divided with a fixed step size. The scale range of the P1 axis starts from the lowest stable unit load P1. min Up to rated unit load P1 max The P2 axis scale ranges from the lowest ambient temperature P2min Up to the highest ambient temperature P2 max A two-dimensional grid coordinate system is ultimately formed by dividing the data with a fixed step size. Each grid cell in the two-dimensional grid coordinate system corresponds to a discretized operating condition.

[0023] S22. The sliding window method is used to extract historical operating data of the thermal system units throughout the year, generating a set of overlapping data windows. Each data window contains the target parameter, relevant parameters, boundary parameters, and characteristic variables for a continuous time period. The target parameter, relevant parameters, boundary parameters, and characteristic variables of each sampling point in each data window constitute each sample operating condition. Each sample operating condition is mapped to the corresponding grid cell according to the value of its boundary parameter, and the statistical characteristics of each grid cell are output. The statistical characteristics include cell coordinates, number of samples, maximum value of target parameter, minimum value of target parameter, average value of target parameter, and scarcity factor, where the scarcity factor refers to the preset sample number threshold in the grid cell.

[0024] Specifically, for each boundary parameter, the operating condition number of the historical operating data is determined according to the interval number in which its value falls. For example, if the unit load falls into grid number 3 and the ambient temperature falls into grid number 10, then the operating condition number corresponding to the operating condition grid cell is B-3-10, where B refers to the boundary parameter. After all historical operating data are mapped, the number of samples in different operating condition grid cells reflects the data density of that grid cell.

[0025] To segment a continuous data stream into statistically significant discrete segments, this invention employs a sliding window method for data extraction. The sliding window method defines two core parameters: Time window width: Defines the time span contained in a single data sample, for example, set to 5 minutes, to ensure that the data within the window can fully reflect a relatively stable operating state.

[0026] Sampling step size: Defines the distance the window slides on the time axis each time, for example, set to 1 minute, to ensure that the generated data window sequence can effectively cover the dynamic changes of historical data.

[0027] The processing flow is as follows: The window slides along the time axis from the starting point of the historical data, with a set sampling step size. Each slide captures a full-width portion of the data, forming a "data window". This operation is repeated until all historical data has been traversed, ultimately generating a batch of overlapping data window sets.

[0028] S23. After cleaning and filling all sample working condition data in the grid cell, the optimal working condition sample is screened for the sample working conditions in each grid cell. Specifically, the steady-state screening is first performed on the sample working conditions in the grid cell. That is, when the difference between the extreme values ​​of the characteristic variables of the sample is lower than the preset threshold, the corresponding sample belongs to the steady-state sample. Then, the optimization criteria of the target parameter is determined according to empirical knowledge (that is, the maximum value is good or the minimum value is good). The single sample data with the best performance of the target parameter is selected from the samples that have passed the steady-state screening as the optimal working condition sample of the grid cell. S24. Output the target working condition library and the optimal working condition sample for each grid cell.

[0029] S3. A modular simulation model of the thermal system is constructed using the EBSILON platform. The modular simulation model includes core physical subsystems such as boilers and steam turbines. A grid coordinate system consistent with the target operating condition library is constructed. Each grid cell is further divided into multiple cells. Boundary parameter values ​​are generated at the intersections of the cells. The boundary parameter values ​​at the intersections are input into the modular simulation model. The modular simulation model solves for the boundary parameter values ​​at each intersection to obtain the corresponding target parameters, thereby generating a simulation dataset that corresponds one-to-one with the target operating condition library, thus constructing a simulation data operating condition library. The simulation dataset and the target operating condition library strictly correspond in terms of the number of cells, cell location, and cell number, realizing the alignment between the simulation space and the actual operating space, and providing data supplementation for cells with scarce samples in the target operating condition library.

[0030] The target operating condition library constructed in this invention not only achieves complete discretization of the unit's historical operating space, but also establishes a structured information archive for each operating condition unit, including sample density, steady-state state, optimal sample, and simulation completion data. By maintaining a one-to-one correspondence with the simulation operating condition library, the target operating condition library provides a unified operating condition coordinate system and high-quality input data for subsequent neighborhood expansion, local modeling, and online optimization, achieving alignment and fusion between the operating space and the simulation space.

[0031] S4. During unit operation, boundary parameters, relevant parameters and target parameters are collected in real time, and the boundary parameters are mapped to the grid coordinate system of the target operating condition library to determine the grid cell of the current operating condition. The number of operating condition samples in the grid cell of the current operating condition is determined. Differential optimization is performed based on the target operating condition library and the simulation data operating condition library, and the optimal target parameter and optimal relevant parameter in the grid cell of the current operating condition are output. like Figure 2 and Figure 3 As shown, step S4 specifically includes the following steps: S41. Compare the number of working condition samples in the grid cell of the current working condition in the target working condition library with the preset scarcity factor. When the number of working condition samples is not less than the scarcity factor, the grid cell of the current working condition is a normal cell. When the number of working condition samples is less than the scarcity factor, the grid cell of the current working condition is a scarce cell. S42. For normal cells, extract the historical best value of the target parameter within the grid cell of the current working condition in the target working condition library, and compare the historical best value of the target parameter with the real-time value of the target parameter of the current working condition; if the real-time value of the target parameter is better, update the real-time value in the target working condition library to the best value, and output the real-time value of the target parameter; otherwise, directly output the corresponding historical best value of the target parameter in the target working condition library. S43. For scarce cells, take the scarce cells in the target working condition library as the center and gradually expand the neighborhood outward until the total number of working condition samples of all grid cells in the expanded neighborhood is not less than the scarcity factor, and denote the working condition samples of all grid cells in the expanded neighborhood as the local sample set D1. From the simulation data condition library, extract the simulation dataset D2 corresponding to the extended neighborhood. Using simulation dataset D2 as the training set, construct the basic prediction model using the support vector machine (SVM) algorithm. The input of the basic prediction model is the boundary parameter values, and the output of the basic prediction model is the predicted value of the target parameter. A modified network model (such as a BP neural network) is added to the basic prediction model, and the deviation between the predicted value of the target parameter of the basic prediction model and the true value of the local sample set D1 is learned and corrected using the local sample set D1 as training data, forming a final local model that integrates simulation data and real data; the current real-time boundary parameters are input into the final local model, and the output is the optimal value of the predicted target parameter of the scarce unit. S44. For normal units, extract the relevant parameter values ​​corresponding to the optimal working condition sample from the target working condition library as the optimal relevant parameters; S45. For scarce cells, take the scarce cells in the target working condition library as the center and gradually expand the neighborhood outward until the total number of working condition samples of all grid cells in the expanded neighborhood is not less than the scarcity factor, and denote the working condition samples of all grid cells in the expanded neighborhood as the local sample set D1. From the simulation data condition library, the simulation dataset D2 corresponding to the extended neighborhood is extracted. Using the simulation dataset D2 as the training set, the basic prediction model is constructed using the support vector machine (SVM) algorithm. The input of the basic prediction model is the boundary parameter values, and the output of the basic prediction model is the predicted values ​​of the relevant parameters. A modified network model is added to the basic prediction model, and the deviation between the predicted values ​​of the relevant parameters of the basic prediction model and the true values ​​of the local sample set D1 is learned and corrected using the local sample set D1 as training data, forming a final local model that integrates simulation data and real data. The current real-time boundary parameters are input into the final local model, and the output is the optimal value of the predicted relevant parameters of the scarce unit.

[0032] S5. Dynamically update the target operating condition database. To ensure the database continuously adapts to changes in unit operating characteristics, a dynamic update mechanism is implemented. This mechanism continuously collects new operating data, performs steady-state and anomaly checks, and maps and integrates it into the corresponding operating condition unit based on its boundary parameters. When a scarce unit's sample size meets the criteria for a "normal unit" due to the continuous influx of new data, the system automatically triggers a re-screening and information update of the optimal operating condition sample for that unit. Furthermore, the system periodically performs consistency checks on the entire database to remove invalid historical data that has not been updated for a long time or deviates from the current unit characteristics, ensuring the long-term effectiveness of the database.

[0033] Example 2: A real-time optimization system for energy efficiency indicators of a thermal system that integrates a target operating condition database and prior knowledge, comprising: The parameter selection module is used to determine the required target parameters, relevant parameters, boundary parameters, and characteristic variables.

[0034] In the parameter selection module, the target parameter refers to the core indicator used to evaluate the energy consumption of the thermal system or the energy efficiency of the equipment. The relevant parameter refers to the working fluid parameter that can be manually adjusted to control the operating status of the unit. The boundary parameter refers to the parameter that is affected by external objective conditions and cannot be directly controlled by the thermal system. The characteristic variable refers to the parameter that can characterize the steady-state operation of the system. The characteristic variables are selected as load, main steam pressure, main steam temperature, reheat steam temperature and feedwater flow rate.

[0035] The target operating condition library construction module is used to collect target parameters, related parameters, boundary parameters, and feature variables from the historical operating data of the thermal system units throughout a given year. It selects two or three boundary parameters as coordinate system dimensions to construct a grid coordinate system covering the entire operating condition range. The sliding window method is used to sample the historical operating data to form several sample operating conditions. Each sample operating condition is mapped to a grid cell in the grid coordinate system according to its boundary parameters to construct the target operating condition library and select the optimal operating condition sample in each grid cell.

[0036] The target operating condition library construction module collects target parameters, related parameters, boundary parameters, and characteristic variables from the historical operating data of the thermal system units throughout a certain year. It selects two or three boundary parameters as coordinate system dimensions, determines the extreme values ​​of the boundary parameters based on all boundary parameter data throughout the year, and sets the scale range of each coordinate axis as the extreme value interval. It sets a fixed step size for each coordinate system dimension and divides the coordinate axis with the fixed step size to form a grid coordinate system covering the entire operating condition range. Each grid cell corresponds to a discretized operating condition. A sliding window method is used to extract historical operating data of the thermal system units throughout the year, generating an overlapping set of data windows. Each data window contains target parameters, relevant parameters, boundary parameters, and characteristic variables for a continuous time period. The target parameters, relevant parameters, boundary parameters, and characteristic variables of each sampling point within each data window constitute each sample operating condition. Each sample operating condition is mapped to a corresponding grid cell according to the values ​​of its boundary parameters, and the statistical characteristics of each grid cell are output. The statistical characteristics include cell coordinates, number of samples, maximum value of target parameter, minimum value of target parameter, average value of target parameter, and scarcity factor, where the scarcity factor refers to the preset sample number threshold within the grid cell. After cleaning and filling all sample working condition data within the grid cell, the optimal working condition sample is screened for each grid cell. Specifically, the steady-state screening is first performed on the sample working conditions within the grid cell. That is, when the difference between the extreme values ​​of the characteristic variables of the sample is lower than the preset threshold, the corresponding sample belongs to the steady-state sample. Then, the optimization criteria of the target parameter are determined based on empirical knowledge. The single sample data with the best performance of the target parameter is selected from the samples that have passed the steady-state screening as the optimal working condition sample of the grid cell. Output the target working condition library and the optimal working condition sample for each grid cell.

[0037] The simulation data condition library construction module is used to build a modular simulation model of the thermal system, construct a grid coordinate system consistent with the target condition library, and divide each grid cell into multiple grids evenly. Boundary parameter values ​​are generated at the intersection points of the grids. The boundary parameter values ​​at the intersection points are input into the modular simulation model. The modular simulation model solves for the boundary parameter values ​​at each intersection point to obtain the corresponding target parameters, thereby generating a simulation dataset that corresponds one-to-one with the target condition library and constructing the simulation data condition library. The parameter optimization module is used to collect boundary parameters, relevant parameters, and target parameters in real time during unit operation. It maps the boundary parameters to the grid coordinate system of the target operating condition library to determine the grid cell of the current operating condition, determines the number of operating condition samples in the grid cell of the current operating condition, performs differentiated optimization based on the target operating condition library and the simulation data operating condition library, and outputs the optimal target parameter and optimal relevant parameter in the grid cell of the current operating condition.

[0038] In the parameter optimization module, the number of working condition samples in the grid cell of the current working condition in the target working condition library is compared with the preset scarcity factor. When the number of working condition samples is not less than the scarcity factor, the grid cell of the current working condition is a normal cell. When the number of working condition samples is less than the scarcity factor, the grid cell of the current working condition is a scarce cell. For normal cells, extract the historical best value of the target parameter within the grid cell of the current working condition from the target working condition library, and compare the historical best value of the target parameter with the real-time value of the target parameter of the current working condition; if the real-time value of the target parameter is better, update the real-time value in the target working condition library to the best value, and output the real-time value of the target parameter; otherwise, directly output the corresponding historical best value of the target parameter in the target working condition library. For scarce cells, take the scarce cells in the target working condition library as the center and gradually expand the neighborhood outward until the total number of working condition samples of all grid cells in the expanded neighborhood is not less than the scarcity factor, and denote the working condition samples of all grid cells in the expanded neighborhood as the local sample set D1. From the simulation data condition library, extract the simulation dataset D2 corresponding to the extended neighborhood. Using simulation dataset D2 as the training set, construct the basic prediction model using the support vector machine (SVM) algorithm. The input of the basic prediction model is the boundary parameter values, and the output of the basic prediction model is the predicted value of the target parameter. A modified network model is added to the basic prediction model, and the deviation between the predicted value of the target parameter of the basic prediction model and the true value of the local sample set D1 is learned and corrected using the local sample set D1 as training data, forming a final local model that integrates simulation data and real data; the current real-time boundary parameters are input into the final local model, and the output is the optimal value of the predicted target parameter of the scarce unit. For normal units, the relevant parameter values ​​corresponding to the optimal working condition sample are extracted from the target working condition library as the optimal relevant parameters. For scarce cells, take the scarce cells in the target working condition library as the center and gradually expand the neighborhood outward until the total number of working condition samples of all grid cells in the expanded neighborhood is not less than the scarcity factor, and denote the working condition samples of all grid cells in the expanded neighborhood as the local sample set D1. From the simulation data condition library, the simulation dataset D2 corresponding to the extended neighborhood is extracted. Using the simulation dataset D2 as the training set, the basic prediction model is constructed using the support vector machine (SVM) algorithm. The input of the basic prediction model is the boundary parameter values, and the output of the basic prediction model is the predicted values ​​of the relevant parameters. A modified network model is added to the basic prediction model, and the deviation between the predicted values ​​of the relevant parameters of the basic prediction model and the true values ​​of the local sample set D1 is learned and corrected using the local sample set D1 as training data, forming a final local model that integrates simulation data and real data. The current real-time boundary parameters are input into the final local model, and the output is the optimal value of the predicted relevant parameters of the scarce unit.

[0039] Example 3: An electronic device includes a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the method described above.

[0040] Example 4: A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

Claims

1. A method for real-time optimization of energy efficiency indicators of a thermal system that integrates a target operating condition database and prior knowledge, characterized in that, Includes the following steps: S1. Determine the required target parameters, relevant parameters, boundary parameters, and characteristic variables; S2. Collect target parameters, related parameters, boundary parameters, and characteristic variables from the historical operating data of the thermal system units throughout a given year. Select 2 or 3 boundary parameters as coordinate system dimensions to construct a grid coordinate system covering the entire operating condition range. Use the sliding window method to sample the historical operating data to form several sample operating conditions. Map each sample operating condition to a grid cell in the grid coordinate system according to its boundary parameters to construct a target operating condition library and select the optimal operating condition sample in each grid cell. S3. Construct a modular simulation model of the thermal system, build a grid coordinate system consistent with the target operating condition library, and divide each grid cell into multiple grids evenly. The boundary parameter values ​​are generated at the intersection of the grids. Input the boundary parameter values ​​at the intersections into the modular simulation model. The modular simulation model solves the boundary parameter values ​​at each intersection to obtain the corresponding target parameters, thereby generating a simulation dataset that corresponds one-to-one with the target operating condition library and constructing a simulation data operating condition library. S4. During unit operation, boundary parameters, relevant parameters and target parameters are collected in real time, and the boundary parameters are mapped to the grid coordinate system of the target operating condition library to determine the grid cell of the current operating condition. The number of operating condition samples in the grid cell of the current operating condition is determined. Differential optimization is performed based on the target operating condition library and the simulation data operating condition library, and the optimal target parameter and optimal relevant parameter in the grid cell of the current operating condition are output. S5. Dynamically update the target operating condition library.

2. The method for real-time optimization of energy efficiency indicators of a thermal system by integrating a target operating condition database and prior knowledge as described in claim 1, characterized in that, In step S1, the target parameter refers to the core indicator used to evaluate the energy consumption of the thermal system or the energy efficiency of the equipment; the relevant parameter refers to the working fluid parameter that can be manually adjusted to control the operating state of the unit; the boundary parameter refers to the parameter that is affected by external objective conditions and cannot be directly controlled by the thermal system; the characteristic variable refers to the parameter that can characterize the steady-state operation of the system. The characteristic variables are selected as load, main steam pressure, main steam temperature, reheat steam temperature and feedwater flow rate.

3. The method for real-time optimization of energy efficiency indicators of a thermal system by integrating a target operating condition database and prior knowledge as described in claim 1, characterized in that... Step S2 specifically includes the following steps: S21. Collect target parameters, related parameters, boundary parameters, and characteristic variables from the historical operating data of the thermal system units throughout a year. Select 2 or 3 boundary parameters as coordinate system dimensions. Based on all boundary parameter data throughout the year, determine the extreme values ​​of the boundary parameters. The scale range of each coordinate axis is the extreme value interval. Set a fixed step size for each coordinate system dimension. Divide the coordinate axes with the fixed step size to form a grid coordinate system covering the entire operating condition range. Each grid cell corresponds to a discretized operating condition. S22. The sliding window method is used to extract historical operating data of the thermal system units throughout the year, generating a set of overlapping data windows. Each data window contains the target parameter, relevant parameters, boundary parameters and characteristic variables for a continuous time period. The target parameter, relevant parameters, boundary parameters and characteristic variables of each sampling point in each data window constitute each sample operating condition. Each sample working condition is mapped to the corresponding grid cell according to the value of its boundary parameter, and the statistical characteristics of each grid cell are output. The statistical characteristics include cell coordinates, number of samples, maximum value of target parameter, minimum value of target parameter, average value of target parameter and scarcity factor, where scarcity factor refers to the preset sample number threshold in the grid cell. S23. After cleaning and filling all sample working condition data in the grid cell, the optimal working condition sample is screened for the sample working conditions in each grid cell. Specifically, the steady-state screening is first performed on the sample working conditions in the grid cell. That is, when the difference between the extreme values ​​of the characteristic variables of the sample is lower than the preset threshold, the corresponding sample belongs to the steady-state sample. Then, the optimization criteria of the target parameter are determined based on empirical knowledge. The single sample data with the best performance of the target parameter is selected from the samples that have passed the steady-state screening as the optimal working condition sample of the grid cell. S24. Output the target working condition library and the optimal working condition sample for each grid cell.

4. The method for real-time optimization of energy efficiency indicators of a thermal system by integrating a target operating condition database and prior knowledge as described in claim 1, characterized in that... Step S4 specifically includes the following steps: S41. Compare the number of working condition samples in the grid cell of the current working condition in the target working condition library with the preset scarcity factor. When the number of working condition samples is not less than the scarcity factor, the grid cell of the current working condition is a normal cell. When the number of working condition samples is less than the scarcity factor, the grid cell of the current working condition is a scarce cell. S42. For normal cells, extract the historical optimal values ​​of the target parameters within the grid cells of the current working condition in the target working condition library, and compare the historical optimal values ​​of the target parameters with the real-time values ​​of the target parameters of the current working condition. If the real-time value of the target parameter is better, then update the real-time value in the target operating condition library to the optimal value and output the real-time value of the target parameter; otherwise, directly output the historical optimal value of the target parameter in the target operating condition library. S43. For scarce cells, take the scarce cells in the target working condition library as the center and gradually expand the neighborhood outward until the total number of working condition samples of all grid cells in the expanded neighborhood is not less than the scarcity factor, and denote the working condition samples of all grid cells in the expanded neighborhood as the local sample set D1. From the simulation data condition library, extract the simulation dataset D2 corresponding to the extended neighborhood. Using simulation dataset D2 as the training set, construct the basic prediction model using the support vector machine (SVM) algorithm. The input of the basic prediction model is the boundary parameter values, and the output of the basic prediction model is the predicted value of the target parameter. A modified network model is added to the basic prediction model, and the deviation between the predicted value of the target parameter of the basic prediction model and the true value of the local sample set D1 is learned and corrected using the local sample set D1 as training data, forming a final local model that integrates simulation data and real data; the current real-time boundary parameters are input into the final local model, and the output is the optimal value of the predicted target parameter of the scarce unit. S44. For normal units, extract the relevant parameter values ​​corresponding to the optimal working condition sample from the target working condition library as the optimal relevant parameters; S45. For scarce cells, take the scarce cells in the target working condition library as the center and gradually expand the neighborhood outward until the total number of working condition samples of all grid cells in the expanded neighborhood is not less than the scarcity factor, and denote the working condition samples of all grid cells in the expanded neighborhood as the local sample set D1. From the simulation data condition library, the simulation dataset D2 corresponding to the extended neighborhood is extracted. Using the simulation dataset D2 as the training set, the basic prediction model is constructed using the support vector machine (SVM) algorithm. The input of the basic prediction model is the boundary parameter values, and the output of the basic prediction model is the predicted values ​​of the relevant parameters. A modified network model is added to the basic prediction model, and the deviation between the predicted values ​​of the relevant parameters of the basic prediction model and the true values ​​of the local sample set D1 is learned and corrected using the local sample set D1 as training data, forming a final local model that integrates simulation data and real data. The current real-time boundary parameters are input into the final local model, and the output is the optimal value of the predicted relevant parameters of the scarce unit.

5. A system for real-time optimization of energy efficiency indicators for thermal systems that integrates a target operating condition database and prior knowledge, characterized in that, include: The parameter selection module is used to determine the required target parameters, relevant parameters, boundary parameters, and feature variables; The target operating condition library construction module is used to collect target parameters, related parameters, boundary parameters, and feature variables from the historical operating data of the thermal system units throughout a year. It selects two or three boundary parameters as coordinate system dimensions to construct a grid coordinate system covering the entire operating condition range. It uses the sliding window method to sample the historical operating data to form several sample operating conditions. Each sample operating condition is mapped to a grid cell in the grid coordinate system according to its boundary parameters to construct the target operating condition library and select the optimal operating condition sample in each grid cell. The simulation data condition library construction module is used to build a modular simulation model of the thermal system, construct a grid coordinate system consistent with the target condition library, and divide each grid cell into multiple grids evenly. Boundary parameter values ​​are generated at the intersection points of the grids. The boundary parameter values ​​at the intersection points are input into the modular simulation model. The modular simulation model solves for the boundary parameter values ​​at each intersection point to obtain the corresponding target parameters, thereby generating a simulation dataset that corresponds one-to-one with the target condition library and constructing the simulation data condition library. The parameter optimization module is used to collect boundary parameters, relevant parameters, and target parameters in real time during unit operation. It maps the boundary parameters to the grid coordinate system of the target operating condition library to determine the grid cell of the current operating condition, determines the number of operating condition samples in the grid cell of the current operating condition, performs differentiated optimization based on the target operating condition library and the simulation data operating condition library, and outputs the optimal target parameter and optimal relevant parameter in the grid cell of the current operating condition.

6. The real-time optimization system for energy efficiency indicators of a thermal system that integrates a target operating condition database and prior knowledge, as described in claim 5, is characterized in that: In the parameter selection module, the target parameter refers to the core indicator used to evaluate the energy consumption of the thermal system or the energy efficiency of the equipment; the relevant parameter refers to the working fluid parameter that can be manually adjusted to control the operating state of the unit; the boundary parameter refers to the parameter that is affected by external objective conditions and cannot be directly controlled by the thermal system; the characteristic variable refers to the parameter that can characterize the steady-state operation of the system. The characteristic variables are selected as load, main steam pressure, main steam temperature, reheat steam temperature and feedwater flow rate.

7. The real-time optimization system for energy efficiency indicators of a thermal system that integrates a target operating condition database and prior knowledge, as described in claim 1, is characterized in that: The target operating condition library construction module collects target parameters, related parameters, boundary parameters, and feature variables from the historical operating data of the thermal system units throughout a certain year. It selects two or three boundary parameters as coordinate system dimensions, determines the extreme values ​​of the boundary parameters based on all boundary parameter data throughout the year, and sets the scale range of each coordinate axis as the extreme value interval. It sets a fixed step size for each coordinate system dimension and divides the coordinate axes with the fixed step size to form a grid coordinate system covering the entire operating condition range. Each grid cell corresponds to a discretized operating condition. The sliding window method is used to extract historical operating data of the thermal system units throughout the year, generating an overlapping set of data windows. Each data window contains the target parameter, relevant parameters, boundary parameters and characteristic variables for a continuous time period. The target parameter, relevant parameters, boundary parameters and characteristic variables of each sampling point in each window constitute each sample operating condition. Each sample working condition is mapped to the corresponding grid cell according to the value of its boundary parameter, and the statistical characteristics of each grid cell are output. The statistical characteristics include cell coordinates, number of samples, maximum value of target parameter, minimum value of target parameter, average value of target parameter and scarcity factor, where scarcity factor refers to the preset sample number threshold in the grid cell. After cleaning and filling all sample working condition data within the grid cell, the optimal working condition sample is screened for each grid cell. Specifically, the steady-state screening is first performed on the sample working conditions within the grid cell. That is, when the difference between the extreme values ​​of the characteristic variables of the sample is lower than the preset threshold, the corresponding sample belongs to the steady-state sample. Then, the optimization criteria of the target parameter are determined based on empirical knowledge. The single sample data with the best performance of the target parameter is selected from the samples that have passed the steady-state screening as the optimal working condition sample of the grid cell. Output the target working condition library and the optimal working condition sample for each grid cell.

8. The real-time optimization system for energy efficiency indicators of a thermal system that integrates a target operating condition database and prior knowledge, as described in claim 1, is characterized in that: In the parameter optimization module, the number of working condition samples in the grid cell of the current working condition in the target working condition library is compared with a preset scarcity factor. When the number of working condition samples is not less than the scarcity factor, the grid cell of the current working condition is a normal cell. When the number of working condition samples is less than the scarcity factor, the grid cell of the current working condition is a scarce cell. For normal cells, extract the historical optimal values ​​of target parameters within the grid cells of the current working condition from the target working condition library, and compare the historical optimal values ​​of target parameters with the real-time values ​​of target parameters for the current working condition. If the real-time value of the target parameter is better, then update the real-time value in the target operating condition library to the optimal value and output the real-time value of the target parameter; otherwise, directly output the historical optimal value of the target parameter in the target operating condition library. For scarce cells, take the scarce cells in the target working condition library as the center and gradually expand the neighborhood outward until the total number of working condition samples of all grid cells in the expanded neighborhood is not less than the scarcity factor, and denote the working condition samples of all grid cells in the expanded neighborhood as the local sample set D1. From the simulation data condition library, extract the simulation dataset D2 corresponding to the extended neighborhood. Using simulation dataset D2 as the training set, construct the basic prediction model using the support vector machine (SVM) algorithm. The input of the basic prediction model is the boundary parameter values, and the output of the basic prediction model is the predicted value of the target parameter. A modified network model is added to the basic prediction model, and the deviation between the predicted value of the target parameter of the basic prediction model and the true value of the local sample set D1 is learned and corrected using the local sample set D1 as training data, forming a final local model that integrates simulation data and real data; the current real-time boundary parameters are input into the final local model, and the output is the optimal value of the predicted target parameter of the scarce unit. For normal units, the relevant parameter values ​​corresponding to the optimal working condition sample are extracted from the target working condition library as the optimal relevant parameters. For scarce cells, take the scarce cells in the target working condition library as the center and gradually expand the neighborhood outward until the total number of working condition samples of all grid cells in the expanded neighborhood is not less than the scarcity factor, and denote the working condition samples of all grid cells in the expanded neighborhood as the local sample set D1. From the simulation data condition library, the simulation dataset D2 corresponding to the extended neighborhood is extracted. Using the simulation dataset D2 as the training set, the basic prediction model is constructed using the support vector machine (SVM) algorithm. The input of the basic prediction model is the boundary parameter values, and the output of the basic prediction model is the predicted values ​​of the relevant parameters. A modified network model is added to the basic prediction model, and the deviation between the predicted values ​​of the relevant parameters of the basic prediction model and the true values ​​of the local sample set D1 is learned and corrected using the local sample set D1 as training data, forming a final local model that integrates simulation data and real data. The current real-time boundary parameters are input into the final local model, and the output is the optimal value of the predicted relevant parameters of the scarce unit.

9. An electronic device, characterized in that, Including processor and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1 to 4.

Images