Gas turbine high load and low load switching control system and method and storage medium

Abstract

The embodiment of the invention provides a gas turbine high-load and low-load switching control system and method and a storage medium, and solves the problem that the high-load and low-load switching response speed is low due to the fact that an existing method cannot accurately predict the load of a thermal generator set in combination with the historical load of the thermal generator set. The method comprises the steps of obtaining working parameters of the gas turbine, executing a load prediction model, outputting a load prediction curve of the gas turbine, adjusting the load prediction curve by optimizing the working parameters of the gas turbine, obtaining an adjusted prediction curve, and generating a high load and low load switching strategy. According to the method, the load prediction model is constructed by combining the multi-implicit strata feedforward neural network and the LSTM model, the load prediction model can effectively utilize the gas turbine working parameter placement time sequence characteristic to carry out accurate load prediction, the load prediction model is high in convergence speed and good in robustness, the high load and low load switching response speed is increased, and the load prediction efficiency is improved. And the efficiency of the thermal generator set is ensured.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of gas turbine control, and in particular to a gas turbine high-load and low-load switching control system, method and storage medium. Background Art

[0002] During the operation of the gas turbine, as the load changes, the combustion mode may switch, which is crucial to maintaining stable combustion of the gas turbine and meeting emission standards. In order to achieve effective coordination between the gas turbine and the steam turbine, it is necessary to design a suitable control strategy and consider the actual load of the steam turbine to achieve a balance of load commands.

[0003] Chinese patent CN107528349A discloses a load speed regulation control method and system for thermal power generating units based on the real-time power generation load of regional new energy. According to the new energy power generation capacity in the region and the power grid operation requirements, a switching threshold of the load speed regulation of the thermal power generating units is set. When the new energy power generation load in the region is greater than the threshold, the associated thermal power generating units automatically complete the disturbance-free switching of load regulation from low-speed regulation to high-speed regulation; when the new energy power generation load in the region is less than the threshold, the associated thermal power generating units automatically complete the disturbance-free switching of load regulation from high-speed regulation to low-speed regulation; however, the existing method cannot accurately predict the load of the thermal power generating units in combination with the historical load of the thermal power generating units, resulting in a slow response speed for switching between high load and low load, affecting the efficiency of the thermal power generating units. In view of the above problems, we propose a gas turbine high-load and low-load switching control system, method and storage medium. Summary of the invention

[0004] The embodiments of the present invention aim to solve at least one of the technical problems existing in the prior art, and provide a gas turbine high-load and low-load switching control system, method and storage medium.

[0005] In a first aspect, an embodiment of the present invention provides a high-load and low-load switching control method for a gas turbine, the high-load and low-load switching control method for a gas turbine comprising:

[0006] Obtaining gas turbine operating parameters; wherein the gas turbine operating parameters include unit temperature, pressure, speed, power output, fuel flow, vibration, environment, and load parameters;

[0007] A load prediction model is constructed by combining a multi-hidden layer feedforward neural network and an LSTM model, and the load prediction model is executed with the gas turbine operating parameters as input to output a gas turbine load prediction curve;

[0008] A load forecast curve is obtained, and the load forecast curve is decomposed into at least one group of intrinsic mode functions based on an empirical mode decomposition algorithm. A load switching model is established using a Boltzmann machine algorithm. The load switching model uses time-sharing energy consumption and power output as decision variables. The load forecast curve is adjusted by optimizing the operating parameters of the gas turbine to obtain an adjusted forecast curve.

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

[0010] The adjustment prediction curve is obtained, and based on the adjustment prediction curve as a constraint, the high load and low load periods of the gas turbine are divided into time periods, and the high load and low load switching strategies are generated.

[0011] In some possible embodiments, the method of building a load forecasting model by combining a multi-hidden layer feedforward neural network and an LSTM model specifically includes:

[0012] The LSTM model is used as the initial model of the load forecasting model; wherein the initial model is composed of an autoencoder, two groups of LSTM layers, and a fully connected layer, the LSTM layer includes three groups of LSTM blocks, and a multi-hidden layer feedforward neural network is introduced between the autoencoder and the LSTM layer in the initial model;

[0013] Obtaining gas turbine operating parameters and standardizing the gas turbine operating parameters;

[0014] Among them, during the standardization process, the standardization process is performed on a daily basis, and the gas turbine operating parameters in a day are standardized using the following formula:

[0015]

[0016] Among them, x i is the standardized gas turbine operating parameter, r i is the input gas turbine operating parameters, max{r i}、min{r i} is the maximum parameter and minimum parameter in the gas turbine operating parameter set, l is the number of parameter types, and α is the parameter weight;

[0017] Obtain the standardized gas turbine operating parameters, and divide the gas turbine operating parameters into a training set and a test set using a bagging method, with the ratio of the training set to the test set being 80% and 20%;

[0018] The training set is input into the autoencoder, and the autoencoder performs one-hot encoding on the training samples in the training set;

[0019] Define the activation function and loss function of the multi-hidden layer feedforward neural network, iteratively train the training samples through the loss function combined with the gradient descent method to generate the output of the multi-hidden layer feedforward neural network, where the activation function of the multi-hidden layer feedforward neural network is the ReLU activation function, and the loss function is the MAPE loss function;

[0020] Set the number of neural network nodes in the LSTM layer to 100, the initial learning rate to 0.01, and use the LSTM layer to train the training samples to obtain the LSTM layer output;

[0021] The fully connected layer concatenates and fuses the output samples of the LSTM layer to update the parameters of the load forecasting model until convergence;

[0022] Get the test set, use the test set as input, execute the load forecasting model, get the predicted value of the test sample, and calculate the MAPE value between the predicted value and the true value;

[0023] Divide the MAPE value between the predicted value and the true value into 16 equal parts, count the proportion of the number of test samples in each interval, and calculate the PSI value of the load forecasting model;

[0024] The PSI value is calculated by the following formula:

[0025]

[0026] Among them, Ac d , Ec d are the probability values ​​of the dth interval in the training sample and the test sample respectively;

[0027] Determine whether the PSI value of the load forecasting model exceeds the preset probability threshold. If it exceeds the preset probability threshold, output the load forecasting model.

[0028] In some possible embodiments, the method of executing the load prediction model and outputting the gas turbine load prediction curve specifically includes:

[0029] Obtaining gas turbine operating parameters, standardizing the gas turbine operating parameters, and obtaining a standardized set;

[0030] Perform dimensionality reduction processing on the standardized set to obtain a data dimensionality reduction set;

[0031] Obtain a data dimension reduction set, and the autoencoder performs one-hot encoding on the data dimension reduction set to obtain a data encoding set;

[0032] Loading a data encoding set, the multi-hidden layer feedforward neural network divides the data encoding set into at least one group of initial prediction centers;

[0033] The correlation between the data coding set and the initial prediction center is calculated based on the FCM algorithm, wherein the correlation between the data coding set is calculated by the following formula;

[0034]

[0035] Among them, m is the number of prediction centers, x b Represents the input data encoding set, x g Represents the association degree of the data encoding set, represents the mean of all prediction centers;

[0036] Obtain the relevance of the data coding set, and update the prediction center to which the data coding set belongs based on the relevance;

[0037] Update the posterior expectation variable of the data in the data coding set corresponding to the next period based on the prediction center to which the data coding set belongs;

[0038]

[0039] Among them, u i represents the posterior expected variable of the next period, θ m represents the input representation of the prediction center, and q represents the prediction center weight penalty parameter;

[0040]

[0041] Among them, q represents the prediction center weight penalty parameter, and Q represents the prediction center weight penalty constant.

[0042] In some possible embodiments, the method of executing the load prediction model and outputting the gas turbine load prediction curve specifically further includes:

[0043] Obtain the posterior expected variable of the next period corresponding to the data in the data encoding set, and calculate the kernel matrix of the posterior expected variable based on the covariance function;

[0044] The acquisition function is used to evaluate the confidence upper bound, confidence lower bound of the kernel matrix and the expected candidate points defined by the confidence upper bound and confidence lower bound;

[0045] The expected candidate points limited by the upper and lower confidence bounds are loaded, and the LSTM layer outputs the predicted load value associated with the expected candidate points;

[0046] The fully connected layer integrates at least one set of predicted load values, determines a load state of the predicted load values ​​based on the predicted load values, and outputs a gas turbine load prediction curve.

[0047] In some possible embodiments, the method of establishing a load switching model using a Boltzmann machine algorithm specifically includes:

[0048] Obtain a load forecast curve, decompose the load forecast curve using an empirical mode decomposition algorithm, and obtain a curve decomposition signal;

[0049] Among them, the empirical mode decomposition algorithm is expressed as:

[0050]

[0051] Among them, J i+t represents the decomposition signal of the load forecast curve, ω j represents the center frequency of the load forecast curve interval frequency domain, j i represents the input data of the load forecasting curve, τ represents the noise tolerance of the empirical mode decomposition algorithm;

[0052] The curve decomposition signal is obtained, and the load switching model is established by combining the Boltzmann machine algorithm. The objective function of the load switching model is defined as:

[0053]

[0054] in, Output the load switching value for the load switching model, J(Q) represents the power output constraint, f(j) represents the time-sharing energy consumption constraint, and λ represents the switching coefficient of the load switching model.

[0055] In some possible embodiments, the method for adjusting the load prediction curve by optimizing the gas turbine operating parameters specifically includes:

[0056] Obtain the load forecast curve, initialize the nonlinear convergence factor associated with the load forecast curve by the load switching model, and generate the initial position of the ant colony;

[0057] The fitness and convergence factor of the ant colony are calculated by combining the Boltzmann machine function.

[0058] In some possible embodiments, the method for adjusting the load prediction curve by optimizing the gas turbine operating parameters further includes:

[0059] The load forecasting curve is adjusted by taking time-sharing energy consumption and power output as decision variables and combining ant colony fitness and convergence factor to obtain an adjusted forecasting curve.

[0060] In a second aspect, an embodiment of the present invention provides a gas turbine high load and low load switching control system, which is used to implement the gas turbine high load and low load switching control method as described above, and the gas turbine high load and low load switching control system specifically includes:

[0061] A data acquisition module is used to obtain the gas turbine operating parameters, wherein the gas turbine operating parameters include unit temperature, pressure, speed, power output, fuel flow, vibration, environment, and load parameters;

[0062] The load forecasting module combines the multi-hidden layer feedforward neural network and the LSTM model to build a load forecasting model, takes the gas turbine operating parameters as input, executes the load forecasting model, and outputs the gas turbine load forecasting curve;

[0063] A load adjustment module is used to obtain a load forecast curve, decompose the load forecast curve into at least one set of intrinsic mode functions based on an empirical mode decomposition algorithm, establish a load switching model using a Boltzmann machine algorithm, and adjust the load forecast curve by optimizing the working parameters of the gas turbine to obtain an adjusted forecast curve.

[0064] The strategy output module is used to obtain the adjustment prediction curve, divide the gas turbine into high load and low load time periods based on the adjustment prediction curve as a constraint, and generate high load and low load switching strategies.

[0065] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium storing computer program instructions, which, when executed, implement the gas turbine high-load and low-load switching control method as described above.

[0066] Compared with the prior art, the embodiments of the present invention mainly have the following beneficial effects:

[0067] In the embodiment of the present invention, a load prediction model is constructed by combining a multi-hidden layer feedforward neural network and an LSTM model. The load prediction model can effectively utilize the gas turbine operating parameters to enter the time series characteristics to perform accurate load prediction, and the load prediction model has fast convergence speed and good robustness, which improves the high-load and low-load switching response speed, ensures the efficiency of the thermal power generating unit, and overcomes the problem that the existing method cannot accurately predict the load of the thermal power generating unit in combination with the historical load of the thermal power generating unit, resulting in a slow high-load and low-load switching response speed, affecting the efficiency of the thermal power generating unit.

[0068] In an embodiment of the present invention, a load forecasting model and a training method thereof are provided. The load forecasting model has the advantages of a multi-hidden layer feedforward neural network and an LSTM model, so that it can utilize a multi-layer structure to learn complex nonlinear relationships in input data, thereby effectively capturing long-term dependencies, which not only improves the accuracy and efficiency of the prediction, but also enhances the generalization ability and adaptability of the model.

[0069] In an embodiment of the present invention, a load switching model is provided. The load switching model is constructed using a Boltzmann machine algorithm, so that the load switching model is based on time-sharing energy consumption and power output as constraints, thereby ensuring the efficiency and response speed of the gas turbine when switching between high load and low load, and at the same time ensuring that energy consumption is reduced and power output is more stable when switching between high load and low load. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0071] Figure 1 It is a schematic diagram of the implementation flow of the high-load and low-load switching control method of a gas turbine provided by an embodiment of the present invention;

[0072] Figure 2 It is a schematic diagram of the implementation process of the method for building a load forecasting model by combining a multi-hidden layer feedforward neural network and an LSTM model provided by an embodiment of the present invention;

[0073] Figure 3 It is a schematic diagram of the implementation flow of a method for executing a load prediction model and outputting a gas turbine load prediction curve provided by an embodiment of the present invention;

[0074] Figure 4 It is a schematic diagram of the implementation flow of the method for establishing a load switching model using the Boltzmann machine algorithm provided by an embodiment of the present invention;

[0075] Figure 5 It is a schematic diagram of the implementation flow of the method for adjusting the load prediction curve by optimizing the working parameters of the gas turbine provided in an embodiment of the present invention;

[0076] Figure 6 It is a structural schematic diagram of a gas turbine high-load and low-load switching control system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0077] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work are within the scope of protection of the present invention.

[0078] Unless otherwise specified, the technical terms or scientific terms used in the embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. "Including" or "comprising" used in the embodiments of the present invention neither limit the shapes, numbers, steps, actions, operations, components, originals and / or their groups mentioned, nor exclude the appearance or addition of one or more other different shapes, numbers, steps, actions, operations, components, originals and / or their groups, or the addition of these. In addition, the terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number and order of the indicated technical features. Thus, the features defined as "first" and "second" can explicitly or implicitly include one or more of the features. In the description of the embodiments of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0079] Unless otherwise specifically stated, the relative arrangement of the components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention. At the same time, it should be understood that for ease of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship, and the techniques, methods and devices known to ordinary technicians in the relevant fields may not be discussed in detail, but where appropriate, the techniques, methods and devices shown should be considered as part of the authorized specification. In all examples shown and discussed here, any specific other examples may have different values. It should be noted that similar symbols and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0080] In the description of the embodiments of the present invention, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In the embodiments of the present invention, the schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in the embodiments of the present invention and the features of the different embodiments or examples, without contradiction.

[0081] Below, the exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention, and it should be understood that the present invention is not limited to the exemplary embodiments described here.

[0082] The existing methods cannot accurately predict the load of thermal power generating units in combination with the historical load of the thermal power generating units, resulting in a slow response speed for switching between high load and low load, which affects the efficiency of the thermal power generating units. To address the above problems, we propose a gas turbine high load and low load switching control system, method and storage medium. In short, when the method is executed, the gas turbine operating parameters are first obtained, and a load prediction model is constructed by combining a multi-hidden layer feedforward neural network and an LSTM model. The gas turbine operating parameters are used as input, the load prediction model is executed, and a gas turbine load prediction curve is output. Then, based on the empirical mode decomposition algorithm, the load prediction curve is decomposed into at least one set of intrinsic mode functions, and a load switching model is established using the Boltzmann machine algorithm. Finally, the load switching model uses time-sharing energy consumption and power output as decision variables, and the load prediction curve is adjusted by optimizing the gas turbine operating parameters to obtain an adjusted prediction curve. In the embodiment of the present invention, a load prediction model is constructed by combining a multi-hidden layer feedforward neural network and an LSTM model. The load prediction model can effectively utilize the gas turbine operating parameters to enter the time series characteristics to perform accurate load prediction, and the load prediction model has fast convergence speed and good robustness, which improves the high-load and low-load switching response speed, ensures the efficiency of the thermal power generating unit, and overcomes the problem that the existing method cannot accurately predict the load of the thermal power generating unit in combination with the historical load of the thermal power generating unit, resulting in a slow high-load and low-load switching response speed, affecting the efficiency of the thermal power generating unit.

[0083] The embodiment of the present invention provides a high load and low load switching control method for a gas turbine. Figure 1 The schematic diagram of the implementation flow of the high load and low load switching control method of the gas turbine is shown. The high load and low load switching control method of the gas turbine specifically includes:

[0084] Step S10, obtaining gas turbine operating parameters.

[0085] It should be noted that the gas turbine operating parameters include but are not limited to unit temperature, pressure, speed, power output, fuel flow, vibration, environment, and load parameters, and the gas turbine operating parameters are acquired through temperature sensors, flow sensors, pressure sensors, and vibration sensors based on a preset acquisition period, and the preset acquisition period can be 1s-10min.

[0086] Step S20, combining a multi-hidden layer feedforward neural network and an LSTM model to construct a load prediction model, taking the gas turbine operating parameters as input, executing the load prediction model, and outputting a gas turbine load prediction curve.

[0087] Step S30, obtaining a load forecast curve, decomposing the load forecast curve into at least one set of intrinsic mode functions based on an empirical mode decomposition algorithm, establishing a load switching model using a Boltzmann machine algorithm, wherein the load switching model uses time-sharing energy consumption and power output as decision variables, and adjusting the load forecast curve by optimizing the gas turbine operating parameters to obtain an adjusted forecast curve.

[0088] Step S40, obtaining an adjustment prediction curve, dividing the gas turbine into high load and low load time periods based on the adjustment prediction curve as a constraint, and generating a high load and low load switching strategy.

[0089] In the embodiment of the present invention, a load prediction model is constructed by combining a multi-hidden layer feedforward neural network and an LSTM model. The load prediction model can effectively utilize the gas turbine operating parameters to enter the time series characteristics to perform accurate load prediction, and the load prediction model has fast convergence speed and good robustness, which improves the high-load and low-load switching response speed, ensures the efficiency of the thermal power generating unit, and overcomes the problem that the existing method cannot accurately predict the load of the thermal power generating unit in combination with the historical load of the thermal power generating unit, resulting in a slow high-load and low-load switching response speed, affecting the efficiency of the thermal power generating unit.

[0090] The embodiment of the present invention provides a method for building a load forecasting model by combining a multi-hidden layer feedforward neural network and an LSTM model. Figure 2 The schematic diagram of the implementation process of the method for building a load forecasting model by combining a multi-hidden layer feedforward neural network and an LSTM model is shown. The method for building a load forecasting model by combining a multi-hidden layer feedforward neural network and an LSTM model specifically includes:

[0091] Step S101, taking the LSTM model as the initial model of the load forecasting model, the initial model consists of an autoencoder, two groups of LSTM layers, and a fully connected layer. The LSTM layer includes three groups of LSTM blocks. A multi-hidden layer feedforward neural network is introduced between the autoencoder and the LSTM layer in the initial model.

[0092] Step S102, obtaining gas turbine operating parameters and standardizing the gas turbine operating parameters.

[0093] Among them, during the standardization process, the standardization process is performed on a daily basis, and the gas turbine operating parameters in a day are standardized using the following formula:

[0094]

[0095] Among them, x i is the standardized gas turbine operating parameter, r i is the input gas turbine operating parameters, max{r i}、min{r i} is the maximum parameter and minimum parameter in the gas turbine operating parameter set, l is the number of parameter types, and α is the parameter weight. It should be noted that the number of parameter types is 2-15, and the parameter weight is set based on principal component analysis, linear analysis or expert consultation method.

[0096] Step S103, obtaining the standardized gas turbine operating parameters, and dividing the gas turbine operating parameters into a training set and a test set using a bagging method, with the ratio of the training set to the test set being 80% and 20%.

[0097] Step S104, input the training set into the autoencoder, and the autoencoder performs one-hot encoding on the training samples in the training set.

[0098] Step S105, defining the activation function and loss function of the multi-hidden layer feedforward neural network, iteratively training the training samples through the loss function combined with the gradient descent method, and generating the output of the multi-hidden layer feedforward neural network, wherein the activation function of the multi-hidden layer feedforward neural network is the ReLU activation function, and the loss function is the MAPE loss function.

[0099] Step S106, setting the number of neural network nodes of the LSTM layer to 100, setting the initial learning rate to 0.01, and using the LSTM layer to train the training samples to obtain the LSTM layer output.

[0100] Step S107: The fully connected layer concatenates and fuses the output samples of the LSTM layer to update the parameters of the load forecasting model until convergence.

[0101] In an embodiment of the present invention, a load forecasting model and a training method thereof are provided. The load forecasting model has the advantages of a multi-hidden layer feedforward neural network and an LSTM model, so that it can utilize a multi-layer structure to learn complex nonlinear relationships in input data, thereby effectively capturing long-term dependencies, which not only improves the accuracy and efficiency of the prediction, but also enhances the generalization ability and adaptability of the model.

[0102] Step S108, obtain a test set, use the test set as input, execute the load forecasting model, obtain the predicted value of the test sample, and calculate the MAPE value between the predicted value and the true value.

[0103] Step S109, divide the MAPE value between the predicted value and the true value into 16 equal parts, count the proportion of the number of test samples in each interval, and calculate the PSI value of the load forecasting model.

[0104] The PSI value is calculated by the following formula:

[0105]

[0106] Among them, Ac d , Ec dare the probability values ​​of the dth interval in the training sample and the test sample respectively.

[0107] Step S110, determining whether the PSI value of the load forecasting model exceeds a preset probability threshold.

[0108] Step S111: if the probability exceeds the preset threshold, the load forecasting model is outputted. If the probability does not exceed the preset threshold, the process returns to step S108 to continue iterative training of the model.

[0109] In this embodiment, the PSI value of the load forecasting model is 0.032, which is less than the preset probability threshold of 0.5, indicating that the load forecasting model trained by the embodiment of the present invention has good stability and robustness.

[0110] The embodiment of the present invention provides a method for executing a load prediction model and outputting a gas turbine load prediction curve. Figure 3 The schematic diagram of the implementation flow of the method for executing the load prediction model and outputting the gas turbine load prediction curve is shown. The method for executing the load prediction model and outputting the gas turbine load prediction curve specifically includes:

[0111] Step S201, obtaining gas turbine operating parameters, and standardizing the gas turbine operating parameters to obtain a standardized set.

[0112] Step S202, performing dimensionality reduction processing on the standardized set to obtain a data dimensionality reduction set.

[0113] It should be noted that the methods for reducing the dimensionality of the standardized set include but are not limited to local linear embedding and factor analysis. When using the local linear embedding method for dimensionality reduction, it is assumed that each data point and its neighbors are located on a local linear manifold. LLE maps high-dimensional data to a low-dimensional space by keeping the reconstruction weight between each data point and its neighbors unchanged.

[0114] Step S203, obtaining a data dimension reduction set, and performing one-hot encoding on the data dimension reduction set by the autoencoder to obtain a data encoding set.

[0115] In the embodiment of the present invention, the autoencoder performs one-hot encoding on the data dimension reduction set, which not only solves the problem of classification data processing, but also improves the performance of the model. At the same time, since the standardized set has been processed for dimension reduction, the disadvantages of one-hot encoding, such as increased dimension and information loss, can be overcome.

[0116] Step S204, loading the data code set, and the multi-hidden layer feedforward neural network divides the data code set into at least one group of initial prediction centers.

[0117] Step S205, calculating the correlation between the data code set and the initial prediction center based on the FCM algorithm, wherein the correlation between the data code set is calculated by the following formula;

[0118]

[0119] Among them, m is the number of prediction centers, x b Represents the input data encoding set, x g Represents the association degree of the data encoding set, represents the mean of all prediction centers.

[0120] Step S206, obtaining the relevance of the data coding set, and updating the prediction center to which the data coding set belongs based on the relevance.

[0121] Step S207, updating the posterior expectation variable of the data in the data coding set corresponding to the next period based on the prediction center to which the data coding set belongs;

[0122]

[0123] Among them, u i represents the posterior expected variable of the next period, θ m represents the input representation of the prediction center, and q represents the prediction center weight penalty parameter;

[0124]

[0125] Wherein, q represents the prediction center weight penalty parameter, Q represents the prediction center weight penalty constant, and in this embodiment, the prediction center weight penalty constant is 2-8.

[0126] Step S208, obtaining the posterior expected variable of the next period corresponding to the data in the data coding set, and calculating the kernel matrix of the posterior expected variable based on the covariance function.

[0127] Step S209: using an acquisition function to evaluate the confidence upper bound, confidence lower bound, and expected candidate points defined by the confidence upper bound and confidence lower bound of the kernel matrix.

[0128] Step S210, loading the expected candidate points defined by the upper confidence bound and the lower confidence bound, and the LSTM layer outputs the predicted load value associated with the expected candidate points.

[0129] Step S211, the fully connected layer integrates at least one set of predicted load values, determines the load state of the predicted load values ​​based on the predicted load values, and outputs a gas turbine load prediction curve.

[0130] The embodiment of the present invention provides a method for establishing a load switching model using a Boltzmann machine algorithm. Figure 4 The schematic diagram of the implementation process of the method for establishing a load switching model using the Boltzmann machine algorithm is shown. The method for establishing a load switching model using the Boltzmann machine algorithm specifically includes:

[0131] Step S301, obtaining a load forecast curve, and using an empirical mode decomposition algorithm to decompose the load forecast curve to obtain a curve decomposition signal;

[0132] Among them, the empirical mode decomposition algorithm is expressed as:

[0133]

[0134] Among them, J i+t represents the decomposition signal of the load forecast curve, ω j represents the center frequency of the load forecast curve interval in the frequency domain, j i represents the input data of the load forecasting curve, and τ represents the noise tolerance of the empirical mode decomposition algorithm.

[0135] Step S302, obtain the curve decomposition signal, and establish a load switching model in combination with the Boltzmann machine algorithm. The load switching model objective function is defined as:

[0136]

[0137] in, Output the load switching value for the load switching model, J(Q) represents the power output constraint, f(j) represents the time-sharing energy consumption constraint, and λ represents the switching coefficient of the load switching model.

[0138] In an embodiment of the present invention, a load switching model is provided. The load switching model is constructed using a Boltzmann machine algorithm, so that the load switching model is based on time-sharing energy consumption and power output as constraints, thereby ensuring the efficiency and response speed of the gas turbine when switching between high load and low load, and at the same time ensuring that energy consumption is reduced and power output is more stable when switching between high load and low load.

[0139] The embodiment of the present invention provides a method for adjusting the load prediction curve by optimizing the operating parameters of the gas turbine. Figure 5 A schematic diagram of the implementation flow of the method for adjusting the load prediction curve by optimizing the working parameters of the gas turbine is shown. The method for adjusting the load prediction curve by optimizing the working parameters of the gas turbine specifically includes:

[0140] Step S401, obtaining a load forecast curve, initializing a nonlinear convergence factor associated with the load forecast curve by a load switching model, and generating an initial position of an ant colony.

[0141] Step S402, calculating the fitness and convergence factor of the ant colony in combination with the Boltzmann machine function.

[0142] Step S403, taking the time-sharing energy consumption and the power output as decision variables, and combining the ant colony fitness and the convergence factor to adjust the load forecast curve, to obtain an adjusted forecast curve.

[0143] On the other hand, the embodiment of the present invention also provides a high load and low load switching control system for a gas turbine. Figure 6 The schematic diagram of the structure of the high load and low load switching control system of the gas turbine is shown. The high load and low load switching control system of the gas turbine specifically includes:

[0144] The data acquisition module 100 acquires the gas turbine operating parameters, wherein the gas turbine operating parameters include unit temperature, pressure, speed, power output, fuel flow, vibration, environment, and load parameters.

[0145] The load prediction module 200 constructs a load prediction model by combining a multi-hidden layer feedforward neural network and an LSTM model, takes the gas turbine operating parameters as input, executes the load prediction model, and outputs a gas turbine load prediction curve.

[0146] The load adjustment module 300 is used to obtain a load forecast curve, decompose the load forecast curve into at least one set of intrinsic mode functions based on an empirical mode decomposition algorithm, and establish a load switching model using a Boltzmann machine algorithm. The load switching model uses time-sharing energy consumption and power output as decision variables, and adjusts the load forecast curve by optimizing the gas turbine operating parameters to obtain an adjusted forecast curve.

[0147] The strategy output module 400 is used to obtain the adjustment prediction curve, divide the high load and low load time periods of the gas turbine based on the adjustment prediction curve as a constraint, and generate high load and low load switching strategies.

[0148] In this embodiment, a gas turbine high load and low load switching control system is provided. The gas turbine high load and low load switching control system is composed of a data acquisition module 100, a load prediction module 200, a load adjustment module 300, and a strategy output module 400. The data acquisition module 100, the load prediction module 200, the load adjustment module 300, and the strategy output module 400 work together to effectively utilize the time series characteristics of the gas turbine operating parameters to perform accurate load prediction, thereby improving the high load and low load switching response speed and ensuring the efficiency of the thermal power generating unit.

[0149] On the other hand, an embodiment of the present invention further provides a computer-readable storage medium, which stores computer program instructions. When the computer program instructions are executed, the high-load and low-load switching control method of the gas turbine is implemented.

[0150] As another aspect of the present invention, a computer device is provided. The computer device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the method of any one of the above embodiments is implemented.

[0151] As a non-volatile computer-readable storage medium, the memory can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the high-load and low-load switching control method of the gas turbine in the embodiment of the present invention. The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the use of the high-load and low-load switching control method of the gas turbine, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the local module via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0152] Finally, it should be noted that the computer-readable storage medium (e.g., memory) herein may be a volatile memory or a nonvolatile memory, or may include both a volatile memory and a nonvolatile memory. As an example and not by way of limitation, a nonvolatile memory may include a read-only memory (ROM), a programmable ROM (PROM), an electrically programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), or a flash memory. A volatile memory may include a random access memory (RAM), which may act as an external cache memory. As an example and not by way of limitation, RAM may be obtained in a variety of forms, such as synchronous RAM (DRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The storage devices of the disclosed aspects are intended to include, but are not limited to, these and other suitable types of memory.

[0153] In summary, the present invention provides a gas turbine high-load and low-load switching control system, method and storage medium. In the embodiment of the present invention, a load prediction model is constructed by combining a multi-hidden layer feedforward neural network and an LSTM model. The load prediction model can effectively utilize the gas turbine operating parameters into the time series characteristics to perform accurate load prediction, and the load prediction model has fast convergence speed and good robustness, which improves the high-load and low-load switching response speed, ensures the efficiency of the thermal power generating unit, and overcomes the problem that the existing method cannot accurately predict the load of the thermal power generating unit in combination with the historical load of the thermal power generating unit, resulting in a slow high-load and low-load switching response speed, affecting the efficiency of the thermal power generating unit.

[0154] It should be noted that, for the above-mentioned embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described order of actions, because according to the present invention, some steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0155] In the several embodiments provided by the present invention, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of the above units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be in the form of telecommunication or other forms.

[0156] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0157] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill 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 high load and low load switching control method for a gas turbine, characterized in that: The gas turbine high load and low load switching control method comprises: Obtaining gas turbine operating parameters; wherein the gas turbine operating parameters include unit temperature, pressure, speed, power output, fuel flow, vibration, environment, and load parameters; A load prediction model is constructed by combining a multi-hidden layer feedforward neural network and an LSTM model, and the load prediction model is executed with the gas turbine operating parameters as input to output a gas turbine load prediction curve; A load forecast curve is obtained, and the load forecast curve is decomposed into at least one group of intrinsic mode functions based on an empirical mode decomposition algorithm. A load switching model is established using a Boltzmann machine algorithm. The load switching model uses time-sharing energy consumption and power output as decision variables. The load forecast curve is adjusted by optimizing the operating parameters of the gas turbine to obtain an adjusted forecast curve.

2. The high load and low load switching control method of a gas turbine according to claim 1, characterized in that: The method further comprises: The adjustment prediction curve is obtained, and based on the adjustment prediction curve as a constraint, the high load and low load periods of the gas turbine are divided into time periods, and the high load and low load switching strategies are generated.

3. The high load and low load switching control method of a gas turbine according to claim 1, characterized in that: The method of building a load forecasting model by combining a multi-hidden layer feedforward neural network and an LSTM model specifically includes: The LSTM model is used as the initial model of the load forecasting model; wherein the initial model is composed of an autoencoder, two groups of LSTM layers, and a fully connected layer, the LSTM layer includes three groups of LSTM blocks, and a multi-hidden layer feedforward neural network is introduced between the autoencoder and the LSTM layer in the initial model; Obtaining gas turbine operating parameters and standardizing the gas turbine operating parameters; Among them, during the standardization process, the standardization process is performed on a daily basis, and the gas turbine operating parameters in a day are standardized using the following formula: Among them, x i is the standardized gas turbine operating parameter, r i is the input gas turbine operating parameters, max{r i }、min{r i } is the maximum parameter and minimum parameter in the gas turbine operating parameter set, l is the number of parameter types, and α is the parameter weight; Obtain the standardized gas turbine operating parameters, and divide the gas turbine operating parameters into a training set and a test set using a bagging method, with the ratio of the training set to the test set being 80% and 20%; The training set is input into the autoencoder, and the autoencoder performs one-hot encoding on the training samples in the training set; Define the activation function and loss function of the multi-hidden layer feedforward neural network, iteratively train the training samples through the loss function combined with the gradient descent method to generate the output of the multi-hidden layer feedforward neural network, where the activation function of the multi-hidden layer feedforward neural network is the ReLU activation function, and the loss function is the MAPE loss function; Set the number of neural network nodes in the LSTM layer to 100, the initial learning rate to 0.01, and use the LSTM layer to train the training samples to obtain the LSTM layer output; The fully connected layer concatenates and fuses the output samples of the LSTM layer to update the parameters of the load forecasting model until convergence; Get the test set, use the test set as input, execute the load forecasting model, get the predicted value of the test sample, and calculate the MAPE value between the predicted value and the true value; Divide the MAPE value between the predicted value and the true value into 16 equal parts, count the proportion of the number of test samples in each interval, and calculate the PSI value of the load forecasting model; The PSI value is calculated by the following formula: Among them, Ac d , Ec d are the probability values ​​of the dth interval in the training sample and the test sample respectively; Determine whether the PSI value of the load forecasting model exceeds the preset probability threshold. If it exceeds the preset probability threshold, output the load forecasting model.

4. The high load and low load switching control method for a gas turbine according to any one of claims 1 to 3, characterized in that: The method of executing the load prediction model and outputting the gas turbine load prediction curve specifically includes: Obtaining gas turbine operating parameters, standardizing the gas turbine operating parameters, and obtaining a standardized set; Perform dimensionality reduction processing on the standardized set to obtain a data dimensionality reduction set; Obtain a data dimension reduction set, and the autoencoder performs one-hot encoding on the data dimension reduction set to obtain a data encoding set; Loading a data encoding set, the multi-hidden layer feedforward neural network divides the data encoding set into at least one group of initial prediction centers; The correlation between the data coding set and the initial prediction center is calculated based on the FCM algorithm, wherein the correlation between the data coding set is calculated by the following formula; Among them, m is the number of prediction centers, x b Represents the input data encoding set, x g Represents the association degree of the data encoding set, represents the mean of all prediction centers; Obtain the relevance of the data coding set, and update the prediction center to which the data coding set belongs based on the relevance; Update the posterior expectation variable of the data in the data coding set corresponding to the next period based on the prediction center to which the data coding set belongs; Among them, u i represents the posterior expected variable of the next period, θ m represents the input representation of the prediction center, and q represents the prediction center weight penalty parameter; Among them, q represents the prediction center weight penalty parameter, and Q represents the prediction center weight penalty constant.

5. The high load and low load switching control method of a gas turbine according to claim 4, characterized in that: The method of executing the load prediction model and outputting the gas turbine load prediction curve specifically includes: Obtain the posterior expected variable of the next period corresponding to the data in the data encoding set, and calculate the kernel matrix of the posterior expected variable based on the covariance function; The acquisition function is used to evaluate the confidence upper bound, confidence lower bound of the kernel matrix and the expected candidate points defined by the confidence upper bound and confidence lower bound; The expected candidate points limited by the upper and lower confidence bounds are loaded, and the LSTM layer outputs the predicted load value associated with the expected candidate points; The fully connected layer integrates at least one set of predicted load values, determines a load state of the predicted load values ​​based on the predicted load values, and outputs a gas turbine load prediction curve.

6. The high load and low load switching control method for a gas turbine according to any one of claims 1 to 3, characterized in that: The method of establishing a load switching model using the Boltzmann machine algorithm specifically includes: Obtain a load forecast curve, decompose the load forecast curve using an empirical mode decomposition algorithm, and obtain a curve decomposition signal; Among them, the empirical mode decomposition algorithm is expressed as: Among them, J i+t represents the decomposition signal of the load forecast curve, ω j represents the center frequency of the load forecast curve interval in the frequency domain, j i represents the input data of the load forecasting curve, τ represents the noise tolerance of the empirical mode decomposition algorithm; The curve decomposition signal is obtained, and the load switching model is established by combining the Boltzmann machine algorithm. The objective function of the load switching model is defined as: in, Output the load switching value for the load switching model, J(Q) represents the power output constraint, f(j) represents the time-sharing energy consumption constraint, and λ represents the switching coefficient of the load switching model.

7. The high load and low load switching control method of a gas turbine according to claim 6, characterized in that: The method for adjusting the load prediction curve by optimizing the gas turbine operating parameters specifically includes: Obtain the load forecast curve, initialize the nonlinear convergence factor associated with the load forecast curve by the load switching model, and generate the initial position of the ant colony; The fitness and convergence factor of the ant colony are calculated by combining the Boltzmann machine function.

8. The high load and low load switching control method of a gas turbine according to claim 7, characterized in that: The method for adjusting the load prediction curve by optimizing the gas turbine operating parameters specifically includes: The load forecasting curve is adjusted by taking time-sharing energy consumption and power output as decision variables and combining ant colony fitness and convergence factor to obtain an adjusted forecasting curve.

9. A gas turbine high load and low load switching control system, used to implement the gas turbine high load and low load switching control method according to any one of claims 1 to 8, characterized in that: The gas turbine high load and low load switching control system specifically includes: A data acquisition module is used to obtain the gas turbine operating parameters, wherein the gas turbine operating parameters include unit temperature, pressure, speed, power output, fuel flow, vibration, environment, and load parameters; The load forecasting module combines the multi-hidden layer feedforward neural network and the LSTM model to build a load forecasting model, takes the gas turbine operating parameters as input, executes the load forecasting model, and outputs the gas turbine load forecasting curve; A load adjustment module is used to obtain a load forecast curve, decompose the load forecast curve into at least one set of intrinsic mode functions based on an empirical mode decomposition algorithm, establish a load switching model using a Boltzmann machine algorithm, and adjust the load forecast curve by optimizing the working parameters of the gas turbine to obtain an adjusted forecast curve. The strategy output module is used to obtain the adjustment prediction curve, divide the gas turbine into high load and low load time periods based on the adjustment prediction curve as a constraint, and generate high load and low load switching strategies.

10. A computer-readable storage medium, characterized in that: Computer program instructions are stored, and when the computer program instructions are executed, the high-load and low-load switching control method for a gas turbine according to any one of claims 1 to 8 is implemented.

Images