A method and device for predicting the heat supply of a gas power plant

Abstract

The application discloses a kind of gas power plant heat supply quantity prediction method, it is related to electric power engineering technical field, wherein method includes: obtaining the multidimensional influence factor data of gas power plant heat supply quantity, the multidimensional influence factor data includes the meteorological factor data of current heat supply area, user side heat demand data, gas power plant unit operating parameter data and regional social and economic activity data;The multidimensional influence factor data is preprocessed, and the multidimensional influence factor data after preprocessing is obtained;The multidimensional influence factor data after preprocessing is input into the fusion model of preestablished, utilizes the time sequence correlation characteristic of improvement LSTM network layer extraction, and is extracted through GBDT feature extraction layer nonlinear correlation characteristic;Time sequence correlation characteristic and nonlinear correlation characteristic are dimensionally matched and matrix splicing in feature fusion layer, generate fusion feature matrix, and the prediction result of gas power plant heat supply quantity is output through fully connected output layer, complete prediction.The application can improve the prediction accuracy of gas power plant heat supply quantity.

Description

Technical Field

[0001] This invention relates to the field of power engineering technology, and in particular to a method and apparatus for predicting the heat supply of a gas-fired power plant. Background Technology

[0002] In northern China during the winter heating season and for industrial heating applications, gas-fired power plants are the core heating providers. Accurate forecasting of their heat supply is crucial for the safe, efficient, and economical operation of the heating system. By forecasting heat supply, gas-fired power plants can rationally allocate unit operating loads, optimize heat source distribution schemes, reduce energy losses in the heating network, and ensure the stability of heating supply for users.

[0003] Currently, the forecasting of heating supply from gas-fired power plants mainly relies on traditional statistical methods and basic machine learning methods. Among them, traditional statistical methods are typically represented by time series autoregressive models. These methods can only capture the single time-series variation pattern of heating supply and cannot effectively integrate multi-dimensional influencing factors such as meteorological conditions, user heating demand, and unit operating status, making them difficult to adapt to complex and ever-changing actual heating scenarios.

[0004] Basic machine learning methods, such as ordinary neural network models, have revealed obvious shortcomings in practical applications: on the one hand, they are not good at feature extraction from long-term time-series data, making it difficult to uncover the long-term correlation patterns hidden in the data; on the other hand, they are prone to the gradient vanishing problem, which limits the training effect of the model and ultimately makes it difficult to meet the high standards required for actual production in terms of prediction accuracy.

[0005] Therefore, there is an urgent need to propose a heat supply prediction method that can effectively integrate multi-dimensional influencing factors and take into account both time series and nonlinear feature extraction, so as to improve the accuracy and reliability of heat supply prediction for gas-fired power plants. Summary of the Invention

[0006] The main objective of this invention is to provide a method for predicting the heat supply of a gas-fired power plant.

[0007] Another object of the present invention is to provide a predictive device for the heat supply of a gas-fired power plant.

[0008] The third objective of this invention is to provide an electronic device.

[0009] A fourth objective of this invention is to provide a non-transitory computer-readable storage medium.

[0010] To achieve the above objectives, a first aspect of the present invention provides a method for predicting the heat supply of a gas-fired power plant, comprising:

[0011] S1, acquire multi-dimensional influencing factor data of heating supply from gas-fired power plants, including meteorological data of the current heating area, user-side heating demand data, gas-fired power plant unit operating parameter data, and regional socio-economic activity data; S2, preprocess the multidimensional influencing factor data to obtain preprocessed multidimensional influencing factor data; S3, input the preprocessed multidimensional influencing factor data into the preset fusion model, extract the temporal correlation features using the improved LSTM network layer, and extract the nonlinear correlation features through the GBDT feature extraction layer; S4. The temporal correlation features and the nonlinear correlation features are matched in dimension and concatenated in the feature fusion layer to generate a fused feature matrix. The prediction result of the gas-fired power plant's heat supply is then output through the fully connected output layer to complete the prediction.

[0012] Optionally, the multidimensional influencing factor data includes meteorological factor data of the current heating area and user-side heating demand data, wherein the meteorological factor data covers temperature, humidity, wind speed, precipitation, and sunshine duration, and the user-side heating demand data covers residential heating area and industrial user heating load.

[0013] Optionally, the multidimensional influencing factor data includes gas-fired power plant unit operating parameter data and regional socio-economic activity data. The gas-fired power plant unit operating parameter data covers steam pressure, steam temperature, steam extraction volume, and unit power generation load, while the regional socio-economic activity data covers industrial output value data, resident population data, and holiday arrangement data.

[0014] Optional, also includes: The outlier removal process is performed on the multidimensional influencing factor data to obtain multidimensional influencing factor data with outlier removal completed; The multidimensional influencing factor data after outlier removal is imputed to obtain multidimensional influencing factor data with missing values ​​filled. Based on the multidimensional influencing factor data filled with the missing values, preprocessed data is determined.

[0015] Optionally, preprocessing the multidimensional influencing factor data to obtain preprocessed multidimensional influencing factor data further includes: The multidimensional influencing factor data with missing values ​​filled in is subjected to data standardization to obtain standardized multidimensional influencing factor data. The standardized multidimensional influencing factor data is arranged into a time series to obtain the preprocessed multidimensional influencing factor data.

[0016] Optionally, the improved LSTM network layer includes at least one attention mechanism sublayer and at least one residual connection type LSTM hidden layer. The attention mechanism sublayer is used to dynamically allocate feature weights, and the residual connection type LSTM hidden layer is used to mitigate gradient vanishing.

[0017] Optionally, the GBDT feature extraction layer includes at least one multi-round decision tree training sub-layer and at least one feature output sub-layer. The multi-round decision tree training sub-layer is used to extract the nonlinear correlation features, and the feature output sub-layer is used to align the data dimensions.

[0018] To achieve the above objectives, a second aspect of the present invention provides a predictive device for the heating supply of a gas-fired power plant, comprising: The data acquisition module is used to acquire multi-dimensional influencing factor data of the heating supply of gas-fired power plants. The multi-dimensional influencing factor data includes meteorological factor data of the current heating area, user-side heating demand data, gas-fired power plant unit operating parameter data, and regional socio-economic activity data. The data preprocessing module is used to preprocess the multidimensional influencing factor data to obtain preprocessed multidimensional influencing factor data; The feature extraction module is used to input the preprocessed multidimensional influencing factor data into a preset fusion model, extract temporal correlation features using an improved LSTM network layer, and extract nonlinear correlation features through a GBDT feature extraction layer. The feature fusion and prediction module is used to perform dimensional matching and matrix concatenation of the temporal correlation features and the nonlinear correlation features in the feature fusion layer to generate a fused feature matrix, and output the prediction result of the gas-fired power plant's heating supply through the fully connected output layer to complete the prediction.

[0019] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0020] To achieve the above objectives, a third aspect of this application provides an electronic device, including a processor and a memory; wherein the processor runs a program corresponding to the executable program code stored in the memory, for implementing a method for predicting the heat supply of a gas-fired power plant as described in the first aspect embodiment.

[0021] To achieve the above objectives, a fourth aspect of this application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for predicting the heat supply of a gas-fired power plant as described in the first aspect embodiment.

[0022] The embodiments of this invention have the following beneficial effects: By comprehensively collecting multi-source data and fusing features for prediction, this invention effectively overcomes the limitations of traditional prediction methods that rely on single feature extraction, significantly improving the accuracy of heating supply prediction for gas-fired power plants. The highly accurate prediction results can directly guide heating dispatch optimization, helping power plants to rationally allocate unit operating loads and formulate scientific heat source allocation plans. Simultaneously, this invention can reduce energy losses in the heating network, achieve refined energy consumption control, and lower power plant operating costs. In northern winter heating seasons and industrial production heating scenarios, it can effectively ensure the stability of user-side heating, possessing both significant technological advancement value and practical benefits. Attached Figure Description

[0023] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 A flowchart illustrating a method for predicting the heat supply of a gas-fired power plant, provided as an embodiment of the present invention; Figure 2 This is a structural diagram of a gas-fired power plant heat supply prediction device provided in an embodiment of the present invention. Detailed Implementation

[0024] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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 scope of protection of the present invention.

[0026] A method and apparatus for predicting the heat supply of a gas-fired power plant according to an embodiment of the present invention are described below with reference to the accompanying drawings.

[0027] Example 1 This invention provides a method for predicting the heat supply of a gas-fired power plant. Figure 1 This is a schematic flowchart illustrating a method for predicting the heat supply of a gas-fired power plant according to an embodiment of the present invention. Figure 1 As shown, the method includes the following steps: Step S1: Obtain multi-dimensional influencing factor data on the heating supply of the gas-fired power plant. The multi-dimensional influencing factor data includes meteorological factor data of the current heating area, user-side heating demand data, gas-fired power plant unit operating parameter data, and regional socio-economic activity data.

[0028] In this embodiment, meteorological data includes temperature, humidity, wind speed, precipitation, and sunshine duration, used to characterize the basic heating demand that directly affects residential and industrial users. Temperature data is directly related to the start-up and shutdown frequency and operating power of user heating equipment. Residential heating demand increases exponentially in low-temperature environments, and temperature-controlled workshops in industrial production also increase their heating supply due to lower ambient temperatures. Humidity data indirectly changes actual user heating demand by affecting perceived temperature and the heat transfer efficiency of building envelopes. The coupling effect of multiple meteorological factors can be quantified through a perceived temperature equivalent model.

[0029] In the formula, For perceived equivalent temperature, Here, RH is the actual ambient temperature, v is the relative humidity, and v is the ambient wind speed. These are the temperature and humidity correction factors and the wind speed correction factor, respectively.

[0030] High humidity accelerates heat loss, further increasing the overall heating load. Wind speed data reflects the intensity of cold air infiltration; strong winds exacerbate heat exchange between the inside and outside of buildings, leading to increased heating demand from users. Rainfall and sunshine duration affect heating demand from the perspective of environmental heat balance; in cloudy and rainy weather, natural heat sources are insufficient, and users' active heating demand will increase significantly.

[0031] User-side heating demand data includes residential heating area and industrial user heating load, used to characterize the actual demand for end-user heating. Residential heating area is the core quantitative indicator of civil heating demand, and its value directly determines the basic volume of civil heating. Different house types and building insulation performance correspond to different heat consumption per unit area. Industrial user heating load covers multiple dimensions such as industrial production process heating and factory area heating. The production processes of industries such as chemical and manufacturing have extremely high requirements for heating stability and temperature accuracy. Fluctuations in their heating load directly affect production efficiency and product quality. The combination of the two can comprehensively cover the diversified demand scenarios of end-user heating.

[0032] Operating parameters of gas-fired power plant units include steam pressure, steam temperature, extraction steam rate, and unit power generation load. These parameters characterize the constraints on the power plant's own heating capacity. Steam pressure and steam temperature are key indicators of heating steam quality; insufficient steam pressure or substandard temperature directly reduces heating efficiency and fails to meet user-side heating temperature requirements. Extraction steam rate is a core parameter for the unit's external energy supply; its upper limit determines the power plant's maximum heating capacity, and real-time adjustment of extraction steam rate must match the unit's operating status. There is a significant coupling relationship between unit power generation load and heating load. In combined heat and power (CHP) mode, changes in unit power generation load directly affect the supply capacity of extraction steam for heating. These parameters collectively constitute the boundary conditions for the power plant's heating capacity.

[0033] Regional socioeconomic activity data, including industrial output data, resident population data, and holiday schedule data, can help determine the patterns of heat consumption fluctuations under specific scenarios. Industrial output data reflects the activity level of regional industrial production; growth in industrial output is usually accompanied by increased operating rates of industrial enterprises, which in turn drives up industrial heat load. Resident population data defines the user base for residential heating; areas with higher population density correspond to larger-scale centralized heating demand. Holiday schedule data is used to capture the characteristics of heat consumption fluctuations during specific periods. For example, during the Spring Festival holiday, urban residents returning to their hometowns lead to a decrease in residential heating demand, while the resumption of production in industrial parks causes periodic fluctuations in industrial heat load. This type of data can effectively improve the adaptability of heat supply prediction models to specific scenarios.

[0034] Step S2: Preprocess the multidimensional influencing factor data to obtain preprocessed multidimensional influencing factor data.

[0035] In this embodiment, the preprocessing operations sequentially include outlier removal, missing value imputation, data standardization, and time series sorting. The specific steps are as follows: First, outlier removal processing is performed on the multidimensional influencing factor data to obtain multidimensional influencing factor data with outlier removal completed. Data deviations such as sudden temperature changes caused by meteorological sensor failures and abnormal steam extraction caused by unit parameter acquisition errors are screened and removed to avoid abnormal data interfering with subsequent feature extraction.

[0036] Secondly, the multidimensional influencing factor data after outlier removal is imputed to obtain multidimensional influencing factor data with missing values. For missing data caused by data transmission interruption or temporary equipment failure, this application uses a weighted interpolation method based on adjacent time periods to complete the data. The weights are reasonably allocated according to the correlation between the data before and after the missing time period. The interpolation calculation expression is as follows:

[0037]

[0038] In the formula, The filler values ​​for missing data at time t. These are the valid sampled data from adjacent time periods before and after the missing time. Weights are assigned to the corresponding time periods, and the sum of the weights is 1, in order to ensure the integrity and continuity of the overall dataset.

[0039] Subsequently, the multidimensional influencing factor data with missing values ​​filled in was subjected to data standardization to obtain standardized multidimensional influencing factor data. Due to significant differences in the units of measurement among the various influencing factors collected (e.g., ambient temperature in "°C", unit steam extraction in "t / h", and user heating area in "㎡"), direct input into the model would cause an imbalance in feature weights. Therefore, this application adopts the z-score standardization method to normalize and transform multidimensional data such as meteorological, unit operation, and user heating consumption. The standardized calculation formula is as follows:

[0040] In the formula, This is the original sampling data of the i-th dimension influencing factors. For standardized data, The mean of a single-dimensional dataset. The standard deviation of a single-dimensional dataset can be transformed to unify different types of data to the same order of magnitude.

[0041] Finally, the standardized multidimensional influencing factor data is processed by time series arrangement to obtain preprocessed multidimensional influencing factor data. In this embodiment, the data of each dimension is integrated into a sequence dataset indexed by time step according to the time stamp order of data collection. At the same time, a reasonable time window length is set according to the time scale requirements of heat supply prediction, so as to provide structured input data for subsequent improvement of LSTM network layer to extract time series correlation features.

[0042] Step S3: Input the preprocessed multidimensional influencing factor data into a preset fusion model, extract temporal correlation features using an improved LSTM network layer, and extract nonlinear correlation features using a GBDT feature extraction layer.

[0043] In this embodiment, the preset fusion model adopts a hierarchical feature extraction and fusion architecture, including an input layer, an improved LSTM network layer, a GBDT feature extraction layer, a feature fusion layer, and a fully connected output layer connected in sequence. The input layer serves as the data entry point, responsible for receiving and classifying the preprocessed multidimensional influencing factor data. It can map meteorological factor data, user-side heating demand data, unit operating parameter data, and socio-economic activity data to corresponding feature channels according to the data type, laying the data foundation for subsequent hierarchical feature extraction.

[0044] In this embodiment, the improved LSTM network layer includes at least one attention mechanism sublayer and at least one residual connection type LSTM hidden layer. The attention mechanism sublayer dynamically assigns feature weights to the temporal influencing factor data through an attention weight calculation algorithm. It first calculates the attention score at each time step and then performs normalization processing. The calculation formula is as follows:

[0045] In the formula, Let the attention score be at time step t. The attention weights are normalized. This represents the temporal feature vector output by the LSTM hidden layer. , The model can learn the weight matrix. This is the bias term, and T is the total time step of the timing series.

[0046] This mechanism can automatically identify key periods such as extreme low temperatures and peak heating seasons, increase the weight of highly correlated data, weaken the interference of invalid data, and accurately capture the long-term variation pattern of heating supply.

[0047] The residual-connected LSTM hidden layer adds residual connection paths between network layers, superimposing and fusing the output features of the previous hidden layer with the output features of the current layer. The residual output expression is as follows:

[0048] In the formula, For the first The LSTM layer outputs a feature vector. The output feature vector of the lth layer after the introduction of residual connection can effectively alleviate the gradient vanishing problem in deep network training and ensure that long-term heating features are intact and not lost during network transmission.

[0049] The system dynamically assigns feature weights to the input time-series influencing factors, automatically identifying data from periods that significantly impact heating supply, such as extreme low temperatures and peak heating seasons. By increasing the weight of highly correlated data, its influence on prediction results is strengthened, while the interference from low-correlation, irrelevant data is weakened, allowing the model to more accurately capture the time-series changes in heating supply over long periods. The residual connection type LSTM hidden layer adds residual connection paths between network layers, superimposing the output features of previous hidden layers with the output features of the current layer. This effectively alleviates the gradient vanishing problem during deep network training, ensures the integrity of long-term heating data features during network transmission, and avoids the loss of deep features.

[0050] In this embodiment, the GBDT feature extraction layer includes at least one multi-round decision tree training sub-layer and at least one feature output sub-layer. The multi-round decision tree training sub-layer is based on a gradient boosting iterative strategy to deeply extract the nonlinear correlation features between various influencing factors and heat supply through ensemble learning of multiple decision trees. The iterative update formula is as follows:

[0051] In the formula, This is the output of the GBDT model after the m-th iteration. This is the result of the previous iteration model. Let be the fitted value of the m-th decision tree for the residuals. The model learning rate is used. Through multiple iterations, the decision tree splitting nodes and feature selection are continuously optimized to accurately adapt to complex nonlinear scenarios such as low temperatures, strong winds, and holiday load fluctuations. Key influencing factors are automatically selected and assigned corresponding weights. The feature output sub-layer integrates and reduces the dimensionality of scattered nonlinear features using a dimension alignment algorithm, unifying the feature vector dimension and format, and achieving efficient matching with improved LSTM time-series features.

[0052] Step S4: Perform dimensional matching and matrix concatenation of the temporal correlation features and the nonlinear correlation features in the feature fusion layer to generate a fused feature matrix, and output the prediction result of the gas-fired power plant's heating supply through the fully connected output layer to complete the prediction.

[0053] In this embodiment, the feature fusion layer first performs dimensionality calibration mapping on the two types of features, and completes dimensionality alignment (upgrading or downgrading) through a weight matrix. The dimensionality mapping relationship is as follows:

[0054] In the formula, These are the temporal feature matrix output by LSTM and the nonlinear feature matrix output by GBDT, respectively. The dimension-mapped weight matrix, To ensure consistency in the structure of the two types of feature data, the same dimension feature matrix is ​​calibrated.

[0055] After dimensional alignment, the two types of features are fused using a matrix horizontal concatenation method to construct a fused feature matrix:

[0056] In the formula, For the final fused feature matrix, It represents the horizontal concatenation operation of the matrix, which retains both the temporal variation pattern and nonlinear correlation characteristics after fusion, thus overcoming the limitations of a single feature dimension.

[0057] This fusion feature matrix comprehensively integrates the dynamic temporal variation characteristics and complex nonlinear mapping characteristics of factors influencing heating supply, providing high-dimensional, high-information feature inputs for prediction calculations. This facilitates the model's full exploration of the hidden heating supply impact patterns within the data. The fully connected output layer is configured with multiple levels of hidden layers and output neurons. Through weight matrix operations and activation function mapping, the fusion feature matrix is ​​transformed into predicted heating supply values ​​for gas-fired power plants and output. This application effectively overcomes the shortcomings of traditional prediction methods' single feature extraction by comprehensively acquiring multi-source data, using LSTM and GBDT hierarchical feature extraction, and feature fusion prediction. It significantly improves the accuracy of heating supply prediction for gas-fired power plants, providing reliable data support for heating dispatching. This helps gas-fired power plants achieve precise unit load allocation, scientific optimization of heat source distribution schemes, and reduced energy losses in heating networks, ensuring stable and reliable heating in northern winter heating and industrial production scenarios.

[0058] Example 2 This invention provides a device for predicting the heat supply of a gas-fired power plant. Figure 2 This is a schematic flowchart of a predictive device for the heat supply of a gas-fired power plant, provided in an embodiment of the present invention. Figure 2 As shown, the device includes: The data acquisition module 100 is used to acquire multi-dimensional influencing factor data of the heating supply of the gas-fired power plant. The multi-dimensional influencing factor data includes meteorological factor data of the current heating area, user-side heating demand data, gas-fired power plant unit operating parameter data, and regional socio-economic activity data. The data preprocessing module 200 is used to preprocess the multidimensional influencing factor data to obtain preprocessed multidimensional influencing factor data; The feature extraction module 300 is used to input the preprocessed multidimensional influencing factor data into a preset fusion model, extract temporal correlation features using an improved LSTM network layer, and extract nonlinear correlation features through a GBDT feature extraction layer. The feature fusion and prediction module 400 is used to perform dimensional matching and matrix concatenation of the time-series correlation features and the nonlinear correlation features in the feature fusion layer to generate a fused feature matrix, and output the prediction result of the heating supply of the gas-fired power plant through the fully connected output layer to complete the prediction.

[0059] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0060] Example 3 To implement the methods of the above embodiments, the present invention also provides an electronic device, which includes a memory and a processor; wherein the processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the various steps of the methods described above.

[0061] Example 4 To implement the above embodiments, this application also proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in the foregoing embodiments.

[0062] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0063] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0064] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A method for predicting the heat supply of a gas-fired power plant, characterized in that, include: S1, acquire multi-dimensional influencing factor data of heating supply from gas-fired power plants, including meteorological data of the current heating area, user-side heating demand data, gas-fired power plant unit operating parameter data, and regional socio-economic activity data; S2, preprocess the multidimensional influencing factor data to obtain preprocessed multidimensional influencing factor data; S3, input the preprocessed multidimensional influencing factor data into the preset fusion model, extract the temporal correlation features using the improved LSTM network layer, and extract the nonlinear correlation features through the GBDT feature extraction layer; S4. The temporal correlation features and the nonlinear correlation features are matched in dimension and concatenated in the feature fusion layer to generate a fused feature matrix. The prediction result of the gas-fired power plant's heat supply is then output through the fully connected output layer to complete the prediction.

2. The method according to claim 1, characterized in that, The multidimensional influencing factor data includes meteorological factor data of the current heating area and user-side heating demand data. The meteorological factor data covers temperature, humidity, wind speed, precipitation, and sunshine duration, while the user-side heating demand data covers residential heating area and industrial user heating load.

3. The method according to claim 2, characterized in that, The multidimensional influencing factor data includes operating parameter data of gas-fired power plant units and regional socio-economic activity data. The operating parameter data of gas-fired power plant units covers steam pressure, steam temperature, steam extraction volume, and unit power generation load, while the regional socio-economic activity data covers industrial output value data, resident population data, and holiday arrangement data.

4. The method according to claim 3, characterized in that, Also includes: The outlier removal process is performed on the multidimensional influencing factor data to obtain multidimensional influencing factor data with outlier removal completed; The multidimensional influencing factor data after outlier removal is imputed to obtain multidimensional influencing factor data with missing values ​​filled. Based on the multidimensional influencing factor data filled with the missing values, preprocessed data is determined.

5. The method according to claim 4, characterized in that, The multidimensional influencing factor data is preprocessed to obtain preprocessed multidimensional influencing factor data, which further includes: The multidimensional influencing factor data with missing values ​​filled in is subjected to data standardization to obtain standardized multidimensional influencing factor data. The standardized multidimensional influencing factor data is arranged into a time series to obtain the preprocessed multidimensional influencing factor data.

6. The method according to claim 5, characterized in that, The improved LSTM network layer includes at least one attention mechanism sublayer and at least one residual connection type LSTM hidden layer. The attention mechanism sublayer is used to dynamically allocate feature weights, and the residual connection type LSTM hidden layer is used to alleviate gradient vanishing.

7. The method according to claim 6, characterized in that, The GBDT feature extraction layer includes at least one multi-round decision tree training sub-layer and at least one feature output sub-layer. The multi-round decision tree training sub-layer is used to extract the nonlinear correlation features, and the feature output sub-layer is used to align the data dimensions.

8. A device for predicting the heat supply of a gas-fired power plant, characterized in that, include: The data acquisition module is used to acquire multi-dimensional influencing factor data of the heating supply of gas-fired power plants. The multi-dimensional influencing factor data includes meteorological factor data of the current heating area, user-side heating demand data, gas-fired power plant unit operating parameter data, and regional socio-economic activity data. The data preprocessing module is used to preprocess the multidimensional influencing factor data to obtain preprocessed multidimensional influencing factor data; The feature extraction module is used to input the preprocessed multidimensional influencing factor data into a preset fusion model, extract temporal correlation features using an improved LSTM network layer, and extract nonlinear correlation features through a GBDT feature extraction layer. The feature fusion and prediction module is used to perform dimensional matching and matrix concatenation of the temporal correlation features and the nonlinear correlation features in the feature fusion layer to generate a fused feature matrix, and output the prediction result of the gas-fired power plant's heating supply through the fully connected output layer to complete the prediction.

9. An electronic device, characterized in that, Including processor and memory; The processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the method as described in any one of claims 1-7.

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

Images