Load prediction adjustment method and device for combined cycle power generation system

Abstract

The embodiment of the application provides a load prediction adjustment method and device of a combined cycle power generation system, belongs to the technical field of combined cycle power generation, and solves the problems that the self-adaptation of an LSTM neural network model to unit data is poor and gradient dispersion cannot be avoided when a unit data is processed by an existing method, the method comprises the following steps: obtaining time series working parameters of the combined cycle power generation system, pre-building a load prediction model, outputting a load curve of the combined cycle power generation system in a next period, and realizing distributed adjustment calculation of the combined cycle power generation system based on load output constraints and load adjustment constraints of the combined cycle power generation system; the load prediction model is built by combining a group intelligence algorithm and a back propagation neural network, the load prediction model has the advantages of the group intelligence algorithm and the back propagation neural network in processing time series working parameters, can accurately output the load curve of the combined cycle power generation system in the next period, and provides a basis for distributed adjustment of the combined cycle power generation system.

Description

TECHNICAL FIELD

[0001] The embodiment of the present application relates to the technical field of combined cycle power generation, in particular to a load prediction adjustment method and device of a combined cycle power generation system. BACKGROUND

[0002] The combined cycle power generation system (CCPP) combines the working principle of gas turbines and steam turbines, improves the overall thermal efficiency through the series connection of two different thermodynamic cycles, and usually includes a gas turbine whose exhaust gas is used to generate steam to drive a steam turbine, thereby realizing the cascade utilization of energy. The gas turbine is the most critical part of the combined cycle system, and its design and manufacturing level directly affects the performance and efficiency of the entire system. At the same time, accurate load prediction of power generation equipment in the combined cycle power generation system helps to reduce unnecessary energy waste, such as reducing fuel input during low demand periods, thereby reducing overall energy consumption.

[0003] Chinese patent CN113107626B discloses a combined cycle generator set load prediction method based on multivariate LSTM, which includes the following steps: data acquisition; data preprocessing; dividing one of the unit data into a training set and a test set, and setting the other unit data as a validation set; building a multivariate input LSTM neural network model; substituting the validation set data into the model for prediction, using combined cycle power as the prediction target of unit load; evaluating the model according to the loss value in model training and the root mean square error in prediction; generating a true value-predicted value curve; however, when the existing method LSTM neural network model processes unit data, the LSTM neural network model has poor self-adaptation to unit data, cannot avoid gradient dispersion, and thus cannot retain longer effective memory information; in view of the above problems, we propose a load prediction adjustment method and system for a combined cycle power generation system. SUMMARY

[0004] The embodiment of the present application aims to at least solve one of the technical problems existing in the prior art, and provides a load prediction adjustment method and device for a combined cycle power generation system.

[0005] In a first aspect, the embodiment of the present application provides a load prediction adjustment method for a combined cycle power generation system, which includes:

[0006] Obtaining time series working parameters of a combined cycle power generation system; wherein the combined cycle power generation system includes a gas turbine unit, a steam turbine, a waste heat boiler and a generator; normalizing the time series working parameters to obtain a normalized data set;

[0007] Obtaining the normalized dataset, pre-constructing a load prediction model combined with a swarm intelligence algorithm and a back propagation neural network, taking the back propagation neural network as an initial model, iteratively training the load prediction model by using the normalized dataset, and outputting the trained load prediction model;

[0008] Collecting real-time working parameters within a day, normalizing the real-time working parameters within the day to obtain a real-time dataset, taking the real-time dataset as an input, executing the load prediction model, and outputting a combined cycle power generation system load curve in a next period.

[0009] In some possible embodiments, the method further comprises:

[0010] Obtaining a combined cycle power generation system load curve, defining combined cycle power generation system load output constraints and load regulation constraints in combination with the time-series working parameters, and realizing combined cycle power generation system distributed regulation calculation based on the combined cycle power generation system load output constraints and load regulation constraints.

[0011] In some possible embodiments, the method of normalizing the time-series working parameters specifically comprises:

[0012] Loading the time-series working parameters, completing missing values of the time-series working parameters, and eliminating abnormal values;

[0013] Performing normalization processing on the time-series working parameters, and the normalization processing formula is as follows:

[0014]

[0015] Wherein, x(t) is a normalized value after normalization processing, t is an input time-series working parameter, t max is a maximum value of the time-series working parameter, and t min is a minimum value of the time-series working parameter.

[0016] Label mixed enhancement is performed on the normalized value by using a hybrid enhancement method, and the label mixed enhancement formula is as follows:

[0017] mix-batch x(t) = α × batch x(n) + (1-α) batch x(t-n) (2)

[0018] mix-batch y(t) = α × batch y(n) + (1-α) batch y(t-n) (3)

[0019] Wherein, batch x(n)To process a batch of data, batch x(t-n) For another batch of data to process, batch y(n) A batch of data is used to create augmented labels. y(t-n) For another batch of data, the labels are enhanced by mixing, and α is the mixing enhancement coefficient, which satisfies the following relationship;

[0020]

[0021] Where t is the number of data points, x(t) represents the mean variance of the normalized values, and λ is the mixing enhancement constant;

[0022] The normalized values ​​after hybrid enhancement are vectorized to obtain the normalized dataset.

[0023] In some possible embodiments, the method of iteratively training the load prediction model using a normalized dataset specifically includes:

[0024] Using a backpropagation neural network as the initial model, the initial model consists of an input layer, a hidden layer, and an activation layer. A gating structure based on a swarm intelligence algorithm is introduced into the input layer of the initial model.

[0025] Freeze the hidden layers of the initial model. Introduce forget gate, input gate, and output gate into the hidden layers of the initial model. The hidden layers extract candidate values ​​for hidden features from the training samples. The formula for calculating the candidate values ​​is as follows:

[0026]

[0027] in, To extract candidate values ​​for the hidden features of the training samples, tanh(·) is the activation function, U C Let b be the weight matrix. c Represents the bias term;

[0028] The graph topology of the combined cycle power generation system is constructed by mining the correlation between the normalized dataset and the load curve of the combined cycle power generation system through graph topology representation, and an adaptive matrix is ​​introduced into the graph topology of the combined cycle power generation system to realize the dynamic change of the graph topology;

[0029] Obtain the normalized dataset, divide the normalized dataset into training and testing sets, and initialize the hyperparameters of the initial model;

[0030] Using the training set as input, the initial model is executed, and the internal parameters of the model are iteratively updated and learned to obtain a converged initial model.

[0031] Obtain the test set, input the test set into the converged initial model, and obtain the test prediction results;

[0032] The confidence level of the test prediction results is measured using ACE. If the model confidence level meets the preset confidence threshold, the trained load prediction model is output.

[0033] In some possible embodiments, the use of ACE to measure the confidence level of the test prediction results specifically includes:

[0034] Obtain the test prediction results, define the prediction confidence intervals corresponding to the test prediction results, traverse the prediction confidence intervals, and extract the quantile levels corresponding to the quantile curves that serve as the upper and lower bounds within the prediction confidence intervals.

[0035] Calculate the nominal coverage probability and the empirical coverage probability of the predicted confidence interval;

[0036] The nominal coverage probability and the interval empirical coverage probability are calculated using the following formulas:

[0037]

[0038] Among them, E NCP E represents the nominal coverage probability. ECP This represents the interval empirical coverage probability. These represent the quantile levels corresponding to the quantile curves that serve as the upper and lower bounds, respectively. x represents the predicted value corresponding to the quantile level of the quantile curves that serve as the upper and lower bounds, respectively. t+l Indicates the size of the prediction interval;

[0039] Obtain the nominal coverage probability and the interval empirical coverage probability, and calculate the confidence level of the test prediction results based on the nominal coverage probability and the interval empirical coverage probability;

[0040]

[0041] Among them, E ACE (x) represents the confidence level of the test prediction result. These are the actual values ​​corresponding to the upper and lower bounds of the predicted values, respectively.

[0042] In some possible embodiments, the method of executing the load forecasting model and outputting the combined cycle power generation system load curve for the next cycle specifically includes:

[0043] Obtain real-time working parameters for the day;

[0044] The real-time working parameters within the day are normalized to obtain the real-time dataset;

[0045] The real-time dataset is input into the input layer of the load prediction model. The gating structure of the input layer is optimized for the next cycle of the real-time dataset based on a swarm intelligence algorithm to obtain an optimized data set.

[0046] In some possible embodiments, the method of executing the load forecasting model and outputting the combined cycle power generation system load curve for the next cycle further includes:

[0047] Obtain the optimized data set, input the optimized data set into the hidden layer, and the hidden layer extracts features from the optimized data set to obtain the hidden feature set;

[0048] The activation layer traverses the hidden feature set, calculates the predicted load value of the combined cycle power generation system corresponding to the hidden feature set based on the activation function, integrates at least one set of predicted load values ​​of the combined cycle power generation system, and outputs the combined cycle power generation system load curve for the next cycle.

[0049] In some possible embodiments, the method for calculating distributed regulation of a combined cycle power generation system based on load output constraints and load regulation constraints specifically includes:

[0050] Obtain the load curve of the combined cycle power generation system, and define load output constraints and load regulation constraints based on the load curve and time-series operating parameters of the combined cycle power generation system;

[0051] The load output constraint is expressed as follows:

[0052] P l,min ≤P l ≤P l,max (9)

[0053]

[0054] P l P represents the load output constraint. l,min P l,max These represent the lower and upper limits of the load output constraint, respectively, β. up ,β down p represents the system unit's frequency regulation penalty cost coefficients for up- and down-regulation, and p represents the system unit's peak shaving cost coefficients for up- and down-regulation, respectively. up ,p down This represents the average value of the peak-shaving cost coefficient;

[0055] Load regulation constraints are expressed as follows:

[0056] S l,min ≤S l ≤S l,max (11)

[0057] Among them, S l S represents the load regulation constraint. l,min S l,max These represent the lower limit constraint and upper limit constraint for load regulation, respectively.

[0058] Calculate the distributed regulation difference of the combined cycle power generation system based on the load output constraints and load regulation constraints of the combined cycle power generation system;

[0059] Obtain the distributed regulation difference of the combined cycle power generation system, and realize the distributed regulation of the combined cycle power generation system based on the distributed regulation difference.

[0060] Secondly, embodiments of the present invention provide a load forecasting and regulation device for a combined cycle power generation system, used to implement the load forecasting and regulation method for a combined cycle power generation system as described above. The load forecasting and regulation device for the combined cycle power generation system specifically includes:

[0061] The data acquisition module is used to acquire the timing operating parameters of the combined cycle power generation system;

[0062] The model training module is used to obtain a normalized dataset, pre-build a load prediction model that combines swarm intelligence algorithm and backpropagation neural network. The load prediction model uses the backpropagation neural network as the initial model, iteratively trains the load prediction model using the normalized dataset, and outputs the trained load prediction model.

[0063] The load curve output module is used to collect real-time operating parameters during the day, normalize the real-time operating parameters to obtain a real-time dataset, use the real-time dataset as input to execute the load forecasting model, and output the load curve of the combined cycle power generation system for the next cycle.

[0064] The load regulation module defines the load output constraints and load regulation constraints of the combined cycle power generation system by combining the time-series operating parameters, and realizes the distributed regulation calculation of the combined cycle power generation system based on the load output constraints and load regulation constraints.

[0065] In some possible embodiments, the data acquisition module specifically includes:

[0066] The missing value processing unit loads the time-series working parameters, fills in the missing values ​​of the time-series working parameters, and removes outliers;

[0067] The data augmentation unit is used to perform normalization processing on time-series working parameters and uses a hybrid augmentation method to perform label hybrid augmentation on the normalized values;

[0068] The vectorized representation unit is used to vectorize the normalized values ​​after the hybrid enhancement to obtain the normalized dataset.

[0069] Compared with the prior art, the embodiments of the present invention have the following main advantages:

[0070] In this embodiment of the invention, a load forecasting model is constructed by combining swarm intelligence algorithms and backpropagation neural networks. This model combines the advantages of both swarm intelligence algorithms and backpropagation neural networks in processing time-series operating parameters, thereby accurately outputting the load curve of the combined cycle power generation system for the next cycle, thus providing a basis for distributed regulation of the combined cycle power generation system. This overcomes the problems of existing methods, such as poor adaptability of LSTM neural network models to unit data, inability to avoid gradient vanishing, and consequently, inability to retain longer-term effective memory information.

[0071] In this embodiment of the invention, normalizing the time-series working parameters facilitates the modeling and training of the load forecasting model. Furthermore, the normalization process employs a hybrid enhancement method combined with vectorization, which helps the model integrate data from different sources, solves the data heterogeneity problem, and achieves seamless integration of multi-source data. This provides a more comprehensive information foundation for the integrated analysis of the load forecasting model and helps to discover deeper relationships and patterns between data.

[0072] In this embodiment of the invention, a normalized dataset is used to iteratively train the load prediction model, which consists of an input layer, a hidden layer, and an activation layer. The input layer introduces a gating structure based on a swarm intelligence algorithm, which facilitates the construction of the topology of the cyclic power generation system graph, improves the efficiency and quality of data processing in the load prediction model, and provides strong support for the analysis and decision-making of complex systems.

[0073] In this embodiment of the invention, ACE is used to measure the confidence of the test prediction results, so that the model evaluation does not depend on the probability threshold output by the classifier. Therefore, the performance of the model can be evaluated more comprehensively and the trade-off between sensitivity and specificity can be handled well. Attached Figure Description

[0074] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0075] Figure 1 This is a schematic diagram illustrating the implementation process of the load forecasting and regulation method for a combined cycle power generation system provided by the present invention.

[0076] Figure 2 This is a schematic diagram illustrating the implementation process of the method for normalizing timing parameters provided by the present invention.

[0077] Figure 3This is a schematic diagram illustrating the implementation process of the load prediction model iterative training method using a normalized dataset provided by the present invention.

[0078] Figure 4 This is a schematic diagram of the implementation process of using ACE to measure the confidence of test prediction results provided by the present invention;

[0079] Figure 5 This is a schematic diagram illustrating the implementation process of the load prediction model provided by the present invention, which outputs the load curve of the combined cycle power generation system in the next cycle.

[0080] Figure 6 This is a schematic diagram illustrating the implementation process of the distributed regulation calculation method for a combined cycle power generation system based on load output constraints and load regulation constraints provided by the present invention.

[0081] Figure 7 This is a schematic diagram of the load forecasting and regulation device for a combined cycle power generation system provided by the present invention. Detailed Implementation

[0082] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0083] Unless otherwise specifically stated, the technical or scientific terms used in the embodiments of this invention should be understood in their ordinary sense by those skilled in the art to which this invention pertains. The terms "comprising" or "including," as used in the embodiments of this invention, do not limit the shapes, numbers, steps, actions, operations, components, elements, and / or groups thereof mentioned, nor do they exclude the appearance or addition of one or more other different shapes, numbers, steps, actions, operations, components, elements, and / or groups thereof, or the addition of these. 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 and order of the indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the embodiments of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0084] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale, and techniques, methods, and apparatus known to those skilled in the art may not be discussed in detail; however, where appropriate, the illustrated techniques, methods, and apparatus should be considered part of the specification. In all the examples shown and discussed herein, any other specific example may have different values. It should be noted that similar symbols and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0085] In the description of the embodiments of the present invention, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In the embodiments of the present invention, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in the embodiments of the present invention, as well as the features of different embodiments or examples.

[0086] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0087] Existing methods, such as LSTM neural network models, suffer from poor adaptability to unit data and cannot avoid gradient vanishing, thus failing to retain longer-term effective memory information. To address these issues, we propose a load forecasting and regulation method and system for combined cycle power generation systems. In short, the method first acquires the time-series operating parameters of the combined cycle power generation system and pre-constructs a load forecasting model combining swarm intelligence algorithms and backpropagation neural networks. Then, it collects real-time operating parameters during the day, normalizes these parameters to obtain a real-time dataset, and uses this dataset as input to execute the load forecasting model, outputting the combined cycle power generation system load curve for the next cycle. Finally, it performs distributed regulation calculations based on the combined cycle power generation system's load output constraints and load regulation constraints. In this embodiment, the load forecasting model is constructed by combining swarm intelligence algorithms and backpropagation neural networks. This model combines the advantages of both algorithms in processing time-series operating parameters, enabling accurate output of the combined cycle power generation system load curve for the next cycle, thus providing a basis for distributed regulation of the combined cycle power generation system. This method overcomes the problems of poor adaptability of LSTM neural network models to unit data and inability to avoid gradient vanishing when processing unit data, which in turn prevents the retention of longer effective memory information.

[0088] This invention provides a load forecasting and regulation method for combined cycle power generation systems. Figure 1 A schematic diagram illustrating the implementation process of a load forecasting and regulation method for a combined cycle power generation system is shown. The load forecasting and regulation method for the combined cycle power generation system specifically includes:

[0089] Step S10: Obtain the time-series operating parameters of the combined cycle power generation system, and normalize the time-series operating parameters to obtain a normalized dataset;

[0090] It should be noted that combined cycle power generation systems include, but are not limited to, gas turbine units, steam turbines, waste heat boilers, and generators, while time-series operating parameters include the system equipment's output power, load, speed, vibration, temperature, load curve, peak load, load fluctuation, solar radiation, ambient temperature, and wind speed / direction.

[0091] Step S20: Obtain the normalized dataset, pre-build a load prediction model combining swarm intelligence algorithm and backpropagation neural network, the load prediction model uses backpropagation neural network as the initial model, iteratively trains the load prediction model using the normalized dataset, and outputs the trained load prediction model.

[0092] Step S30: Collect intraday real-time operating parameters, normalize the intraday real-time operating parameters to obtain a real-time dataset, use the real-time dataset as input, execute the load forecasting model, and output the load curve of the combined cycle power generation system for the next cycle.

[0093] It should be noted that the normalization process for intraday real-time operating parameters is the same as that for time-series operating parameters.

[0094] Step S40: Obtain the load curve of the combined cycle power generation system, define the load output constraints and load regulation constraints of the combined cycle power generation system in combination with the time-series operating parameters, and realize the distributed regulation calculation of the combined cycle power generation system based on the load output constraints and load regulation constraints of the combined cycle power generation system.

[0095] In this embodiment of the invention, a load forecasting model is constructed by combining swarm intelligence algorithms and backpropagation neural networks. This model combines the advantages of both swarm intelligence algorithms and backpropagation neural networks in processing time-series operating parameters, thereby accurately outputting the load curve of the combined cycle power generation system for the next cycle, thus providing a basis for distributed regulation of the combined cycle power generation system. This overcomes the problems of existing methods, such as poor adaptability of LSTM neural network models to unit data, inability to avoid gradient vanishing, and consequently, inability to retain longer-term effective memory information.

[0096] This invention provides a method for normalizing timing-related operating parameters. Figure 2 The diagram illustrates the implementation flow of a method for normalizing timing parameters. Specifically, the method for normalizing timing parameters includes:

[0097] Step S101: Load the timing parameters, fill in the missing values ​​of the timing parameters, and remove outliers.

[0098] It should be noted that missing values ​​of time-series working parameters can be filled by deletion, filling, and modeling prediction, while outliers can be removed by standard deviation method, interquartile range method, regression residual method, or isolated forest method.

[0099] Step S102: Perform normalization processing on the timing parameters. The normalization formula is as follows:

[0100] The timing parameters are normalized using the following formula:

[0101]

[0102] Where x(t) is the normalized value after normalization, and t is the input time-series working parameter. max The maximum value of the timing parameter, t min This represents the minimum value of the timing-related operating parameters.

[0103] Step S103: Apply a hybrid enhancement method to the normalized values ​​using label hybrid enhancement. The label hybrid enhancement formula is as follows:

[0104] mix-batch x(t) =α×batch x(n) +(1-α)batch x(t-n) (2)

[0105] mix-batch y(t) =α×batch y(n) +(1-α)batch y(t-n) (3)

[0106] Among them, batch x(n) To process a batch of data, batch x(t-n) For another batch of data to process, batch y(n) A batch of data is used to create augmented labels. y(t-n) For another batch of data, the labels are enhanced by mixing, and α is the mixing enhancement coefficient, which satisfies the following relationship;

[0107]

[0108] Where t is the number of data points. λ represents the mean variance of the normalized value, and λ is the mixing enhancement constant. In this embodiment, the mixing enhancement constant is an integer between 1 and 6.

[0109] Step S104: Vectorize the normalized values ​​after the hybrid enhancement to obtain the normalized dataset.

[0110] In this embodiment of the invention, normalizing the time-series working parameters facilitates the modeling and training of the load forecasting model. Furthermore, the normalization process employs a hybrid enhancement method combined with vectorization, which helps the model integrate data from different sources, solves the data heterogeneity problem, and achieves seamless integration of multi-source data. This provides a more comprehensive information foundation for the integrated analysis of the load forecasting model and helps to discover deeper relationships and patterns between data.

[0111] This invention provides a method for iteratively training a load prediction model using a normalized dataset. Figure 3The diagram illustrates the implementation flow of a method for iteratively training a load prediction model using a normalized dataset. Specifically, this method includes:

[0112] Step S201: Using a backpropagation neural network as the initial model, the initial model consists of an input layer, a hidden layer, and an activation layer. A gating structure based on a swarm intelligence algorithm is introduced into the input layer of the initial model.

[0113] Step S202: Freeze the hidden layers of the initial model. Introduce a forget gate, an input gate, and an output gate into the hidden layers of the initial model. The hidden layers extract candidate values ​​for hidden features from the training samples. The formula for calculating the candidate values ​​is as follows:

[0114] C~ t =tanh(W C x(t)+U C h t-1 +b c (5)

[0115] Among them, C~ t To extract candidate values ​​for the hidden features of the training samples, tanh(·) is the activation function, U C Let b be the weight matrix. c This represents the bias term.

[0116] Step S203: Construction of the combined cycle power generation system graph topology. The correlation between the normalized dataset and the load curve of the combined cycle power generation system is mined through graph topology representation. An adaptive matrix is ​​introduced into the graph topology of the combined cycle power generation system to realize the dynamic change of the graph topology.

[0117] Step S204: Obtain the normalized dataset, divide the normalized dataset into a training set and a test set, and initialize the hyperparameters of the initial model.

[0118] Step S205: Using the training set as input, execute the initial model, iteratively update the internal parameters of the model, and obtain a converged initial model.

[0119] Step S206: Obtain the test set and input the test set into the converged initial model to obtain the test prediction results.

[0120] Step S207: Use ACE to measure the confidence of the test prediction results. If the model confidence meets the preset confidence threshold, output the trained load prediction model.

[0121] In this embodiment of the invention, a normalized dataset is used to iteratively train the load prediction model, which consists of an input layer, a hidden layer, and an activation layer. The input layer introduces a gating structure based on a swarm intelligence algorithm, which facilitates the construction of the topology of the cyclic power generation system graph, improves the efficiency and quality of data processing in the load prediction model, and provides strong support for the analysis and decision-making of complex systems.

[0122] This invention provides a method for measuring the confidence level of test prediction results using ACE. Figure 4 The diagram illustrates the implementation process of using ACE to measure the confidence level of test prediction results. Specifically, using ACE to measure the confidence level of test prediction results includes:

[0123] Step S2071: Obtain the test prediction result, define the prediction confidence interval corresponding to the test prediction result, traverse the prediction confidence interval, and extract the quantile level corresponding to the quantile curve that serves as the upper and lower bounds within the prediction confidence interval.

[0124] Step S2072: Calculate the nominal coverage probability and the empirical coverage probability of the interval corresponding to the predicted confidence interval;

[0125] The nominal coverage probability and the interval empirical coverage probability are calculated using the following formulas:

[0126]

[0127] Among them, E NCP E represents the nominal coverage probability. ECP This represents the interval empirical coverage probability. These represent the quantile levels corresponding to the quantile curves that serve as the upper and lower bounds, respectively. x represents the predicted value corresponding to the quantile level of the quantile curves that serve as the upper and lower bounds, respectively. t+l This indicates the size of the prediction interval.

[0128] Step S2073: Obtain the nominal coverage probability and the interval empirical coverage probability, and calculate the confidence level of the test prediction result based on the nominal coverage probability and the interval empirical coverage probability.

[0129]

[0130] Among them, E ACE (x) represents the confidence level of the test prediction result. These are the actual values ​​corresponding to the upper and lower bounds of the predicted values, respectively.

[0131] In this embodiment of the invention, ACE is used to measure the confidence of the test prediction results, so that the model evaluation does not depend on the probability threshold output by the classifier. Therefore, the performance of the model can be evaluated more comprehensively and the trade-off between sensitivity and specificity can be handled well.

[0132] This invention provides a method for executing a load forecasting model and outputting the load curve of a combined cycle power generation system for the next cycle. Figure 5 The diagram illustrates the implementation flow of a method for executing a load forecasting model and outputting the load curve of a combined cycle power generation system for the next cycle. The method specifically includes:

[0133] Step S301: Obtain real-time working parameters for the day.

[0134] Step S302: Normalize the real-time working parameters during the day to obtain the real-time dataset.

[0135] Step S303: Input the real-time dataset into the input layer of the load prediction model. The gating structure of the input layer is optimized for the next cycle of the real-time dataset based on a swarm intelligence algorithm to obtain an optimized data set.

[0136] Step S304: Obtain the optimized data set, input the optimized data set into the hidden layer, and extract features from the optimized data set to obtain the hidden feature set.

[0137] Step S305: The activation layer traverses the hidden feature set, calculates the predicted load value of the combined cycle power generation system corresponding to the hidden feature set based on the activation function, integrates at least one set of predicted load values ​​of the combined cycle power generation system, and outputs the combined cycle power generation system load curve for the next cycle.

[0138] This invention provides a method for distributed regulation calculation of a combined cycle power generation system based on load output constraints and load regulation constraints. Figure 6 This document illustrates a flowchart of a method for calculating distributed regulation of a combined cycle power generation system based on load output constraints and load regulation constraints. The method specifically includes:

[0139] Step S401: Obtain the load curve of the combined cycle power generation system, and define load output constraints and load regulation constraints based on the load curve and time-series operating parameters of the combined cycle power generation system;

[0140] The load output constraint is expressed as follows:

[0141] P l,min ≤Pl ≤P l,max (9)

[0142]

[0143] P l P represents the load output constraint. l,min P l,max These represent the lower and upper limits of the load output constraint, respectively, β. up ,β down p represents the system unit's frequency regulation penalty cost coefficients for up- and down-regulation, and p represents the system unit's peak shaving cost coefficients for up- and down-regulation, respectively. up ,p down This represents the average value of the peak-shaving cost coefficient;

[0144] Load regulation constraints are expressed as follows:

[0145] S l,min ≤S l ≤S l,max (11)

[0146] Among them, S l S represents the load regulation constraint. l,min S l,max These represent the lower limit constraint and upper limit constraint of load regulation, respectively.

[0147] Step S402: Calculate the distributed regulation difference of the combined cycle power generation system based on the load output constraint and load regulation constraint of the combined cycle power generation system.

[0148] Step S403: Obtain the distributed regulation difference of the combined cycle power generation system, and realize the distributed regulation of the combined cycle power generation system based on the distributed regulation difference of the combined cycle power generation system.

[0149] This invention also provides a load forecasting and regulation device for a combined cycle power generation system. Figure 7 A schematic diagram of a load forecasting and regulation device for a combined cycle power generation system is shown. The load forecasting and regulation device for the combined cycle power generation system specifically includes:

[0150] The data acquisition module 100 is used to acquire the timing operating parameters of the combined cycle power generation system.

[0151] The model training module 200 is used to acquire a normalized dataset, pre-build a load prediction model that combines swarm intelligence algorithm and backpropagation neural network. The load prediction model uses the backpropagation neural network as the initial model, iteratively trains the load prediction model using the normalized dataset, and outputs the trained load prediction model.

[0152] The load curve output module 300 is used to collect real-time operating parameters during the day, normalize the real-time operating parameters to obtain a real-time dataset, use the real-time dataset as input to execute the load forecasting model, and output the load curve of the combined cycle power generation system for the next cycle.

[0153] The load regulation module 400 defines the load output constraints and load regulation constraints of the combined cycle power generation system in combination with the time-series operating parameters, and realizes the distributed regulation calculation of the combined cycle power generation system based on the load output constraints and load regulation constraints of the combined cycle power generation system.

[0154] This invention provides a load forecasting and regulation device for a combined cycle power generation system. The device comprises a data acquisition module 100, a model training module 200, a load curve output module 300, and a load regulation module 400. These components work together to construct a load forecasting model using swarm intelligence algorithms and backpropagation neural networks. This model combines the advantages of both swarm intelligence algorithms and backpropagation neural networks in handling time-series operating parameters, enabling it to accurately output the load curve of the combined cycle power generation system for the next cycle, thus providing a basis for distributed regulation of the combined cycle power generation system. This overcomes the problems of poor adaptability of LSTM neural network models to unit data, the inability to avoid gradient vanishing, and the inability to retain longer-term effective memory information when processing unit data.

[0155] In this embodiment, the data acquisition module 100 specifically includes:

[0156] The missing value processing unit 110 loads the time-series working parameters, fills in missing values ​​for the time-series working parameters, and removes outliers. The data augmentation unit 120 performs normalization processing on the time-series working parameters, and uses a hybrid augmentation method to perform label hybrid augmentation on the normalized values. The vectorization representation unit 130 performs vectorization processing on the normalized values ​​after hybrid augmentation to obtain a normalized dataset.

[0157] On the other hand, embodiments of the present invention also provide a schematic diagram of a computer device, which includes a display screen, a memory, a processor, and a computer program. The memory stores the computer program, and when the computer program is executed by the processor, the processor performs the steps of the load forecasting and regulation method for the combined cycle power generation system.

[0158] It is understood that, in the preferred embodiments provided by the present invention, the computer device may also be a laptop computer, a personal digital assistant (PDA), a mobile phone, or other devices capable of communication.

[0159] In another aspect, the present invention provides a computer-readable storage medium storing computer program instructions that can be executed by a processor. When executed, the computer program instructions implement the method of any of the above embodiments.

[0160] For example, a computer program can be divided into one or more modules, one or more of which are stored in memory and executed by a processor to perform the present invention. One or more modules can be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in a terminal device. For example, the aforementioned computer program can be divided into units or modules of the load forecasting and regulation device for the combined cycle power generation system provided in the various system embodiments described above.

[0161] Finally, it should be noted that the computer-readable storage medium (e.g., memory) described herein can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. By way of example, and not limitation, non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM), which can act as external cache memory. By way of example, and not limitation, RAM can be obtained in various forms, such as synchronous RAM (DRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The storage devices disclosed herein are intended to include, but are not limited to, these and other suitable types of memory.

[0162] The various exemplary logic blocks, modules, and circuits described herein can be implemented or performed using the following components designed to perform the functions herein: general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic, discrete hardware components, or any combination of these components. A general-purpose processor may be a microprocessor, but alternatively, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors combined with a DSP, and / or any other such configuration.

[0163] In summary, this invention provides a load forecasting and regulation method and system for combined cycle power generation systems. In the embodiments of this invention, a load forecasting model is constructed by combining swarm intelligence algorithms and backpropagation neural networks. This load forecasting model combines the advantages of swarm intelligence algorithms and backpropagation neural networks in processing time-series operating parameters, thereby accurately outputting the load curve of the combined cycle power generation system for the next cycle, thus providing a basis for distributed regulation of the combined cycle power generation system. This overcomes the problems of existing methods, such as poor adaptability of LSTM neural network models to unit data, inability to avoid gradient vanishing, and consequently, inability to retain longer-term effective memory information.

[0164] It should be noted that, for the sake of simplicity, the foregoing embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0165] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A load forecasting and regulation method for a combined cycle power generation system, characterized in that, The load forecasting and regulation method for the combined cycle power generation system includes: Obtain the time-series operating parameters of the combined cycle power generation system; wherein the combined cycle power generation system includes a gas turbine unit, a steam turbine, a waste heat boiler, and a generator, and the time-series operating parameters include the output power, load, speed, vibration, temperature, load curve, peak load, load fluctuation, solar radiation, ambient temperature, and wind speed / direction of the system equipment; normalize the time-series operating parameters to obtain a normalized dataset; Obtain the normalized dataset, pre-construct a load prediction model combining swarm intelligence algorithm and backpropagation neural network. The load prediction model uses backpropagation neural network as the initial model, iteratively trains the load prediction model using the normalized dataset, and outputs the trained load prediction model. Collect intraday real-time operating parameters, normalize the intraday real-time operating parameters to obtain a real-time dataset, use the real-time dataset as input to execute the load forecasting model, and output the load curve of the combined cycle power generation system for the next cycle. The method further includes: obtaining the load curve of the combined cycle power generation system, defining the load output constraint and load regulation constraint of the combined cycle power generation system in combination with the time-series operating parameters, and realizing the distributed regulation calculation of the combined cycle power generation system based on the load output constraint and load regulation constraint of the combined cycle power generation system; The method for calculating distributed regulation of a combined cycle power generation system based on load output constraints and load regulation constraints specifically includes: Obtain the load curve of the combined cycle power generation system, and define load output constraints and load regulation constraints based on the load curve and time-series operating parameters of the combined cycle power generation system; The load output constraint is expressed as follows: P lmin ≤P l ≤P lmax P l P represents the load output constraint. l,min P l,max These represent the lower and upper limits of the load output constraint, respectively, β. up ,β down p represents the frequency regulation penalty cost coefficients of the system unit, respectively. up ,p down These represent the average values ​​of the peak shaving cost coefficients for the upper and lower peak shaving operations, respectively. Load regulation constraints are expressed as follows: S l,min ≤S l ≤S l,max Among them, S l S represents the load regulation constraint. l,min S l,max These represent the lower limit constraint and upper limit constraint for load regulation, respectively. Calculate the distributed regulation difference of the combined cycle power generation system based on the load output constraints and load regulation constraints of the combined cycle power generation system; Obtain the distributed regulation difference of the combined cycle power generation system, and realize the distributed regulation of the combined cycle power generation system based on the distributed regulation difference.

2. The load forecasting and regulation method for a combined cycle power generation system according to claim 1, characterized in that, The method for normalizing the timing parameters specifically includes: Load the time-series working parameters, fill in missing values ​​for the time-series working parameters, and remove outliers; The timing parameters are normalized using the following formula: Where x(t) is the normalized value after normalization, and t is the input time-series working parameter. max The maximum value of the timing parameter, t min This represents the minimum value of the timing-related operating parameters; A hybrid enhancement method is used to perform label hybrid enhancement on the normalized values. The label hybrid enhancement formula is as follows: mix_batch x(t) =α×batch x(n) +(1-α)batch x(t-n) (2) mix_batch y(t) =α×batch y(n) +(1-α)batch y(t-n) (3) Among them, batch x(n) To process a batch of data, batch x(t-n) For another batch of data to process, batch y(n) A batch of data is used to create augmented labels. y(t-n) For another batch of data, the labels are enhanced by mixing, and α is the mixing enhancement coefficient, which satisfies the following relationship; Where t is the number of data points. λ represents the mean variance of the normalized values, and λ is the mixing enhancement constant. The normalized values ​​after hybrid enhancement are vectorized to obtain the normalized dataset.

3. The load forecasting and regulation method for a combined cycle power generation system according to claim 2, characterized in that, The method of iteratively training the load prediction model using a normalized dataset specifically includes: Using a backpropagation neural network as the initial model, the initial model consists of an input layer, a hidden layer, and an activation layer. A gating structure based on a swarm intelligence algorithm is introduced into the input layer of the initial model. Freeze the hidden layers of the initial model. Introduce forget gate, input gate, and output gate into the hidden layers of the initial model. The hidden layers extract candidate values ​​for hidden features from the training samples. The formula for calculating the candidate values ​​is as follows: in, To extract candidate values ​​for the hidden features of the training samples, tanh(·) is the activation function, U C Let b be the weight matrix. c Represents the bias term; The graph topology of the combined cycle power generation system is constructed by mining the correlation between the normalized dataset and the load curve of the combined cycle power generation system through graph topology representation, and an adaptive matrix is ​​introduced into the graph topology of the combined cycle power generation system to realize the dynamic change of the graph topology; Obtain the normalized dataset, divide the normalized dataset into training and testing sets, and initialize the hyperparameters of the initial model; Using the training set as input, the initial model is executed, and the internal parameters of the model are iteratively updated and learned to obtain a converged initial model. Obtain the test set, input the test set into the converged initial model, and obtain the test prediction results; The confidence level of the test prediction results is measured using ACE. If the model confidence level meets the preset confidence threshold, the trained load prediction model is output.

4. The load forecasting and regulation method for a combined cycle power generation system according to claim 3, characterized in that, The use of ACE to measure the confidence level of the test prediction results specifically includes: Obtain the test prediction results, define the prediction confidence intervals corresponding to the test prediction results, traverse the prediction confidence intervals, and extract the quantile levels corresponding to the quantile curves that serve as the upper and lower bounds within the prediction confidence intervals. Calculate the nominal coverage probability and the empirical coverage probability of the predicted confidence interval; The nominal coverage probability and the interval empirical coverage probability are calculated using the following formulas: Among them, E NCP E represents the nominal coverage probability. ECP This represents the interval empirical coverage probability. f These represent the quantile levels corresponding to the quantile curves that serve as the upper and lower bounds, respectively. x represents the predicted value corresponding to the quantile level of the quantile curves that serve as the upper and lower bounds, respectively. t+l Indicates the size of the prediction interval; Obtain the nominal coverage probability and the interval empirical coverage probability, and calculate the confidence level of the test prediction results based on the nominal coverage probability and the interval empirical coverage probability; Among them, E ACE (x) represents the confidence level of the test prediction result. z f These are the actual values ​​corresponding to the upper and lower bounds of the predicted values, respectively.

5. The load forecasting and regulation method for a combined cycle power generation system according to claim 1, characterized in that, The method for executing the load forecasting model and outputting the load curve of the combined cycle power generation system for the next cycle specifically includes: Obtain real-time working parameters for the day; The real-time working parameters within the day are normalized to obtain the real-time dataset; The real-time dataset is input into the input layer of the load prediction model. The gating structure of the input layer is optimized for the next cycle of the real-time dataset based on a swarm intelligence algorithm to obtain an optimized data set.

6. The load forecasting and regulation method for a combined cycle power generation system according to claim 5, characterized in that, The method for executing the load forecasting model and outputting the load curve of the combined cycle power generation system for the next cycle further includes: Obtain the optimized data set, input the optimized data set into the hidden layer, and the hidden layer extracts features from the optimized data set to obtain the hidden feature set; The activation layer traverses the hidden feature set, calculates the predicted load value of the combined cycle power generation system corresponding to the hidden feature set based on the activation function, integrates at least one set of predicted load values ​​of the combined cycle power generation system, and outputs the combined cycle power generation system load curve for the next cycle.

7. A load forecasting and regulation device for a combined cycle power generation system, used to implement the load forecasting and regulation method for a combined cycle power generation system as described in any one of claims 1 to 6, characterized in that, The load forecasting and regulation device for the combined cycle power generation system specifically includes: The data acquisition module is used to acquire the timing operating parameters of the combined cycle power generation system; The model training module is used to obtain a normalized dataset, pre-build a load prediction model that combines swarm intelligence algorithm and backpropagation neural network. The load prediction model uses the backpropagation neural network as the initial model, iteratively trains the load prediction model using the normalized dataset, and outputs the trained load prediction model. The load curve output module is used to collect real-time operating parameters during the day, normalize the real-time operating parameters to obtain a real-time dataset, use the real-time dataset as input to execute the load forecasting model, and output the load curve of the combined cycle power generation system for the next cycle. The load regulation module defines the load output constraints and load regulation constraints of the combined cycle power generation system by combining the time-series operating parameters, and realizes the distributed regulation calculation of the combined cycle power generation system based on the load output constraints and load regulation constraints.

8. The load forecasting and regulation device for a combined cycle power generation system according to claim 7, characterized in that, The data acquisition module specifically includes: The missing value processing unit loads the time-series working parameters, fills in the missing values ​​of the time-series working parameters, and removes outliers; The data augmentation unit is used to perform normalization processing on time-series working parameters and uses a hybrid augmentation method to perform label hybrid augmentation on the normalized values; The vectorized representation unit is used to vectorize the normalized values ​​after the hybrid enhancement to obtain the normalized dataset.

Images