Method and device for constructing digital real-time simulation model of generator stator

Abstract

The invention relates to the technical field of generator stator simulation, and discloses a generator stator digital real-time simulation model construction method and device, and the method comprises the steps: obtaining the field actual working condition data of a target pumped storage generator, and constructing an operation working condition data set based on the field actual working condition data of the target pumped storage generator; constructing an MLP neural network prediction model; wherein the MLP neural network prediction model takes an air cooler outlet air temperature, a stator winding current and a stator winding voltage in the operation condition data set as independent variables, and takes a stator winding average temperature and a stator core average temperature as dependent variables; and based on the operation condition data set, performing parameter tuning on the MLP neural network prediction model by using layered optimization and a single-layer multi-objective optimization algorithm to obtain a generator stator digital real-time simulation model. According to the method, the practicability and the accuracy of the generator stator digital real-time simulation model are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of generator stator simulation, and particularly relates to a method and device for constructing a digital real-time simulation model of a generator stator. Background Art

[0002] Building a physical model for real-time simulation operation of a digital model that operates in parallel with the actual generator of a hydropower or pumped storage unit, and realizing real-time simulation, online visualization, and blind-spot-free operation monitoring of pumped storage unit equipment from the technical direction of online simulation of physical fields has become an important development direction for the future intelligent operation of hydropower or pumped storage units.

[0003] However, the construction methods of related simulation models have certain limitations in the digital twin of hydropower and pumped storage unit equipment. They only consider the theoretical methods for building surrogate models, or completely ignore whether on-site monitoring data can play a driving role, resulting in unreasonable selection of model parameters, low efficiency, and insufficient accuracy, affecting the efficiency and accuracy of the equipment-level digital twin model. Summary of the Invention

[0004] In view of this, the present invention provides a method and device for constructing a digital real-time simulation model of a generator stator to solve the problems of unreasonable selection of model parameters, low efficiency, and insufficient accuracy of the simulation model.

[0005] In a first aspect, the present invention provides a method for constructing a digital real-time simulation model of a generator stator, the method comprising: Obtaining on-site actual operating condition data of a target pumped storage generator, and constructing an operating condition data set based on the on-site actual operating condition data of the target pumped storage generator; Constructing an MLP neural network prediction model; wherein, the MLP neural network prediction model takes the air temperature at the outlet of the air cooler, the stator winding current, and the stator winding voltage in the operating condition data set as independent variables, and takes the average temperature of the stator winding and the average temperature of the stator core as dependent variables; Based on the operating condition data set, using a hierarchical optimization and single-layer multi-objective optimization algorithm to optimize the parameters of the MLP neural network prediction model, and obtaining a digital real-time simulation model of the generator stator.

[0006] A method for constructing a digital real-time simulation model of a generator stator provided in this embodiment constructs an operating condition data set through the on-site actual operating condition data of the target pumped storage generator, increases the feasibility of on-site layout without adding new sensors, uses the on-site actual operating condition data as the inlet boundary for driving the model, and realizes the accurate selection of model parameters of the simulation model; and uses the hierarchical optimization and single-layer multi-objective optimization algorithms to optimize the parameters of the MLP neural network prediction model, obtains the digital real-time simulation model of the generator stator, reduces the dimension of the search space, improves the practicability and accuracy of the digital real-time simulation model of the generator stator, promotes the intelligent and digital processes in the fields of hydropower and pumped storage projects, and provides support for the innovation and development of project management.

[0007] In an alternative embodiment, constructing an operating condition data set based on the on-site actual operating condition data of the target pumped storage generator includes: Determining the steady-state operating condition type based on the on-site actual operating condition data of the target pumped storage generator; Performing data cleaning, normalization processing, and timestamp correction on the on-site actual operating condition data of the target pumped storage generator to obtain operating condition data; Constructing an operating condition data set based on the operating condition data with the same steady-state operating condition type.

[0008] A method for constructing a digital real-time simulation model of a generator stator provided in this embodiment, since the data volume of the on-site actual operating condition data corresponding to different steady-state operating condition types is extremely large, when constructing the digital real-time simulation model of the generator stator, training is performed with the operating condition data sets corresponding to different steady-state operating condition types. Furthermore, when driving the input with data, first judge the steady-state operating condition category, and then call the corresponding model, which can reduce the error and construction difficulty of constructing the same model without classifying by operating conditions; and, judge the steady-state operating condition type based on the on-site actual operating condition data, so as to realize the real-time display of the corresponding proxy model driven by the operating data, and improve the efficiency, accuracy, and safety of on-site operation and maintenance of the digital real-time simulation model of the generator stator.

[0009] In an alternative embodiment, based on the operating condition data set, using the hierarchical optimization and single-layer multi-objective optimization algorithms to optimize the parameters of the MLP neural network prediction model to obtain the digital real-time simulation model of the generator stator includes: Based on the operating condition data set, performing structural layer hyperparameter optimization on the MLP neural network prediction model to obtain an optimized MLP neural network prediction model; wherein, the structural layer hyperparameters include the number of hidden layers of the MLP neural network and the number of neurons in each hidden layer; Based on the operating condition data set, the hyperparameters of the training layer of the optimized MLP neural network prediction model are optimized to obtain the digital real-time simulation model of the generator stator; among them, the hyperparameters of the training layer include the learning rate, the learning decay coefficient, the number of training times, and the regularization coefficient.

[0010] For the construction method of the digital real-time simulation model of the generator stator provided in this embodiment, since the selection of the hyperparameters of the MLP neural network prediction model is too random, the hyperparameters of the structure layer are first optimized, and after the hyperparameters of the structure layer are fixed, the hyperparameters of the training layer are optimized, reducing the search space dimension, being able to provide a more efficient and intelligent search strategy, and improving the optimization efficiency and accuracy of the corresponding parameters of the MLP neural network prediction model.

[0011] The accuracy of the digital real-time simulation model of the generator stator is ensured.

[0012] In an optional implementation manner, based on the operating condition data set, the hyperparameters of the structure layer of the MLP neural network prediction model are optimized to obtain the optimized MLP neural network prediction model, including: Taking the MLP neural network prediction model as the objective function to be optimized, taking the hyperparameters of the structure layer as independent variables, and taking the average relative error between the average predicted temperature of the stator winding and the average temperature of the stator winding in the operating condition data set as the target result; Randomly select multiple groups of hyperparameter groups of the structure layer as initial sample points, and use the objective function to be optimized to determine the average relative error corresponding to the initial sample points; Based on the initial sample points and the average relative error, use Gaussian process to fit the surrogate probability function to obtain the variance of each group of hyperparameter groups of the structure layer; Compare the variance of each group of hyperparameter groups of the structure layer with the variance threshold. If the variance of the hyperparameter group of the structure layer is greater than the variance threshold, sample the hyperparameter group of the structure layer using the first acquisition function to obtain updated sample points; Use the updated sample points to iteratively evaluate the objective function to be optimized until the average relative error is less than the preset error threshold to obtain the optimal hyperparameter group of the structure layer; Determine the optimized MLP neural network prediction model based on the optimal hyperparameter group of the structure layer.

[0013] For the construction method of the digital real-time simulation model of the generator stator provided in this embodiment, the parameter space is effectively explored through the surrogate probability function, reducing the number of direct calculations of the objective function, being suitable for computationally expensive optimization problems. Through the acquisition function, it can well balance exploring the unknown area and exploiting the known area, avoiding falling into local optima. The Gaussian process can provide prediction uncertainty, which helps to better guide the sampling process, making the optimal hyperparameter group of the structure layer more accurate and improving the parameter tuning efficiency of the MLP neural network prediction model.

[0014] In an alternative embodiment, based on the operating condition data set, hyperparameter optimization of the training layer of the optimized MLP neural network prediction model is performed to obtain a digital real-time simulation model of the generator stator, including: Randomly select hyperparameters of the training layer to establish multiple groups of training layer hyperparameter groups, and use the optimized MLP neural network prediction model to determine the target values corresponding to the multiple groups of training layer hyperparameter groups; Take the multiple groups of training layer hyperparameter groups and target values as initial data, and use the initial data to train the multi-objective surrogate model to establish the joint distribution of the objective function; Evaluate the multiple groups of training layer hyperparameter groups using the joint distribution of the objective function. Based on the evaluation results, use the second acquisition function to select the next training layer hyperparameter group to be evaluated; Re-evaluate the next training layer hyperparameter group to be evaluated until the current iteration number reaches the preset iteration number to obtain the optimal training layer hyperparameter group; Input the optimal training layer hyperparameter group into the optimized MLP neural network prediction model to obtain a digital real-time simulation model of the generator stator.

[0015] A method for constructing a digital real-time simulation model of a generator stator provided in this embodiment performs hyperparameter optimization of the training layer of the optimized MLP neural network prediction model, and substitutes the optimal structural layer hyperparameter group into the optimization process of the training layer hyperparameters, which can guide the search process to converge to the optimal solution faster, improve the optimization effect and efficiency, and by constructing the joint distribution of the objective function, comprehensively consider the information of each objective, provide a set of optimal solutions, provide more choice space for decision-makers, help find the optimal trade-off solutions that meet different requirements and preferences, and can adaptively adjust the search direction in the entire search space, improving the parameter optimization efficiency and accuracy of the digital real-time simulation model of the generator stator.

[0016] In an alternative embodiment, it further includes: Evaluate the digital real-time simulation model of the generator stator, and perform model optimization on the digital real-time simulation model of the generator stator based on the evaluation results to obtain an optimized digital real-time simulation model of the generator stator.

[0017] A method for constructing a digital real-time simulation model of a generator stator provided in this embodiment improves the accuracy and generalization ability of the digital real-time simulation model of the generator stator by evaluating the digital real-time simulation model of the generator stator and then performing model optimization on the digital real-time simulation model of the generator stator.

[0018] In a second aspect, the present invention provides a device for constructing a digital real-time simulation model of a generator stator, and the device includes: An acquisition module, configured to acquire on-site actual operating condition data of a target pumped-storage generator, and construct an operating condition data set based on the on-site actual operating condition data of the target pumped-storage generator; A construction module, configured to construct an MLP neural network prediction model; wherein, the MLP neural network prediction model uses the air temperature at the outlet of the air cooler, the stator winding current, and the stator winding voltage in the operating condition data set as independent variables, and uses the average stator winding temperature and the average stator core temperature as dependent variables; A parameter tuning module, configured to perform parameter tuning on the MLP neural network prediction model based on the operating condition data set by using a hierarchical optimization and single-layer multi-objective optimization algorithm to obtain a digital real-time simulation model of the generator stator.

[0019] In a third aspect, the present invention provides a computer device, including: a memory and a processor, which are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to execute the method for constructing a digital real-time simulation model of the generator stator according to the first aspect or any corresponding embodiment thereof.

[0020] In a fourth aspect, the present invention provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to cause a computer to execute the method for constructing a digital real-time simulation model of the generator stator according to the first aspect or any corresponding embodiment thereof.

[0021] In a fifth aspect, the present invention provides a computer program product, including computer instructions, and the computer instructions are used to cause a computer to execute the method for constructing a digital real-time simulation model of the generator stator according to the first aspect or any corresponding embodiment thereof. Description of the Drawings

[0022] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0023] Figure 1 is a schematic flowchart of a method for constructing a digital real-time simulation model of a generator stator according to an embodiment of the present invention; Figure 2 is a schematic flowchart of another method for constructing a digital real-time simulation model of a generator stator according to an embodiment of the present invention; Figure 3 is a schematic flowchart of yet another method for constructing a digital real-time simulation model of a generator stator according to an embodiment of the present invention; Figure 4 It is a schematic flowchart of a method for constructing another digital real - time simulation model of a generator stator according to an embodiment of the present invention; Figure 5 It is a structural block diagram of a device for constructing a digital real - time simulation model of a generator stator according to an embodiment of the present invention; Figure 6 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed implementation manners

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0025] The arrangement of generator sensors in hydropower stations or pumped - storage power stations is limited. The maintenance personnel of the power station can only understand and evaluate the operating state of the equipment through scattered and single local measurement point data, and cannot perceive the equipment in real - time and comprehensively in three - dimensions. They cannot understand and master the detailed state of the equipment in the entire spatial and temporal distribution, and there are certain blind spots in the current equipment status monitoring, such as the underwater components of the unit and the inside of the generator; the relevant monitoring means cannot visually display the operating state of the unit equipment, which is not convenient for the operation and maintenance personnel to view and analyze the operating conditions of the equipment in real - time.

[0026] In the field of digital twin intelligent operation and maintenance of generators in hydropower and pumped - storage units, different methods need to be used to construct different real - time simulation models, so as to achieve the dimensionality reduction of the numerical calculation order or data fitting, and realize the real - time display of the multi - physical fields of each component through visualization means.

[0027] The main methods for realizing the reduction of the multi - physical field model order are the numerical calculation method and the simulation data - driven method: the numerical calculation method includes mathematical dimensionality reduction such as proper orthogonal decomposition. By constructing a surrogate model between the input conditions and the orthogonal basis coefficients, the purpose of quickly obtaining the response of the entire three - dimensional field is achieved; the simulation data - driven method includes methods such as support vector machines and deep learning to realize the fitting of the input - output relationship, thereby constructing the corresponding relationship.

[0028] The driving parameters for constructing the real - time model of the generator digital twin include loss power, inlet temperature, inlet flow rate, etc.

[0029] The above - mentioned method for constructing the simulation model has the following problems: 1) Unreasonable selection of driving parameters: The condition monitoring system of the power station does not include real-time data on power loss and wind speed. Therefore, for the construction method with power loss as the driving factor, even if a surrogate model is constructed, it cannot be combined with on-site data to construct real-time driving based on on-site data and can only remain at the theoretical research level.

[0030] 2) Single comparison between the results output by the digital real-time simulation model and conventional simulation data: When the dataset of the digital real-time simulation model is large enough, theoretically, the error between the reduced-order simulation and the conventional simulation will continuously decrease. Therefore, for the monitoring system applied on-site, the comparison with the data collected by on-site actual sensors needs to be considered.

[0031] 3) The construction of the digital real-time simulation model is not classified according to working conditions: The working conditions of pumped-storage motors are complex. For steady-state working conditions, they include power generation, motor operation, power generation phase regulation, and pumping phase regulation; for transient working conditions, they include black start, etc. Therefore, corresponding surrogate models need to be constructed for different working conditions respectively; if constructed in a mixed manner, the datasets of different working conditions are huge, increasing the calculation difficulty, and it is difficult to meet the characteristic points of the working conditions.

[0032] In summary, there are certain limitations in the implementation method relying on simulation model technology for the digital twin of hydropower and pumped-storage unit equipment. It only considers the theoretical method of constructing a surrogate model or completely ignores whether on-site monitoring data can play a driving role; therefore, a new technical solution is needed to improve the above implementation method and enhance the efficiency and accuracy of the equipment-level digital twin model.

[0033] The embodiment of the present invention provides a method for constructing a digital real-time simulation model of a generator stator. By combining the on-site condition monitoring system, considering the situation of not adding new sensors, using on-site data as the inlet boundary of the driving model, and at the same time constructing corresponding surrogate models for different working conditions respectively, and adding a small amount of extreme working condition data for training to solve the problem of unreasonable selection of parameters for constructing the surrogate model.

[0034] The embodiment of the present invention provides a method for constructing a digital real-time simulation model of a generator stator. It should be noted that for the method for constructing a digital real-time simulation model of a generator stator provided by the embodiment of the present invention, the execution subject can be a device for constructing a digital real-time simulation model of a generator stator. This device for constructing a digital real-time simulation model of a generator stator can be implemented as part or all of an electronic device through software, hardware, or a combination of software and hardware. Among them, the electronic device can be a server or a terminal. Among them, the server in the embodiment of the present application can be a single server or a server cluster composed of multiple servers. The terminal in the embodiment of the present application can be other intelligent hardware devices such as a smart phone, a personal computer, a tablet computer, a wearable device, and a smart robot. In the following method embodiments, the execution subject is taken as an electronic device as an example for illustration.

[0035] According to an embodiment of the present invention, an embodiment of a method for constructing a digital real-time simulation model of a generator stator is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0036] In this embodiment, a method for constructing a digital real-time simulation model of a generator stator is provided, which can be used for the above-mentioned electronic device. Figure 1 It is a flowchart of a method for constructing a digital real-time simulation model of a generator stator according to an embodiment of the present invention, as Figure 1 shown, the process includes the following steps: Step S101, obtain the on-site actual operating condition data of the target pumped storage generator, and construct an operating condition data set based on the on-site actual operating condition data of the target pumped storage generator.

[0037] Step S102, construct an MLP neural network prediction model; wherein, the MLP neural network prediction model takes the air temperature at the outlet of the air cooler, the stator winding current, and the stator winding voltage in the operating condition data set as independent variables, and takes the average temperature of the stator winding and the average temperature of the stator core as dependent variables.

[0038] Specifically, since the air temperature at the outlet of the air cooler , the stator winding current , and the stator winding voltage are monitorable and accessible data in the power station status monitoring system, they can be used as driving factors for real-time update of the surrogate model; taking the construction of the surrogate model of the stator winding and core temperature field of the pumped storage motor as an example, select the air temperature at the outlet of the air cooler corresponding to the same operating condition , the stator winding current , and the stator winding voltage as independent variables, and the average temperature of the stator winding and the average temperature of the stator core as two independent dependent variables.

[0039] Furthermore, the MLP (Multilayer Perceptron) neural network is a type of feedforward neural network, which includes an input layer, at least one intermediate hidden layer, and an output layer. By adjusting the neuron weights, the prediction error of the model is minimized, thereby realizing the training of the model and the prediction of the results.

[0040] Further, the MLP neural network prediction model adopts a multi-layer perceptron MLP layer, which is composed of three fully connected layers and two activation functions. The output of the first fully connected layer is: (1) The output of the second fully connected layer is: (2) The output of the multi-layer perceptron (i.e., the third fully connected layer) is: (3) Among them, 、 、 respectively represent the weight matrices of the three fully connected layers, 、 、 respectively represent the bias terms of the fully connected layers, represents the activation function, is the processed input independent variable, including the air temperature at the outlet of the air cooler 、the stator winding current and the stator winding voltage 。

[0041] Step S103: Use the hierarchical optimization and single-layer multi-objective optimization algorithm to optimize the parameters of the MLP neural network prediction model to obtain the digital real-time simulation model of the generator stator.

[0042] Specifically, the Bayesian optimization algorithm is used to optimize the parameters of the MLP neural network prediction model. The parameters to be optimized of the MLP neural network prediction model include the number of hidden layers of the MLP neural network, the number of neurons in each hidden layer, the learning rate, the learning decay coefficient, the number of training times, and the regularization coefficient.

[0043] Further, input the data matrix corresponding to the operating condition data set into the MLP neural network model training based on Bayesian optimization to find the corresponding hyperparameter combinations, including hierarchical optimization and single-layer multi-objective optimization methods. The hyperparameters of the structure layer are optimized using the EI (Expected Improvement) probability function, and the hyperparameters of the training layer are optimized using multi-objective Bayesian optimization.

[0044] A method for constructing a digital real-time simulation model of a generator stator provided in this embodiment. Since the amount of on-site actual condition data corresponding to different steady-state condition types is extremely large, when constructing the digital real-time simulation model of the generator stator, it is trained with the operating condition data sets corresponding to different steady-state condition types. Then, when the data-driven input is performed, the steady-state condition category is first determined, and then the corresponding model is called, which can reduce the error and construction difficulty of constructing the same model without classifying by working conditions. Moreover, based on the on-site actual condition data, the steady-state condition type is judged, so as to realize the real-time display of the corresponding proxy model driven by the operating data, and improve the efficiency, accuracy and safety of on-site operation and maintenance of the digital real-time simulation model of the generator stator.

[0045] In this embodiment, a method for constructing a digital real-time simulation model of a generator stator is provided, which can be used for the above-mentioned electronic device. Figure 2 It is a flowchart of a method for constructing a digital real-time simulation model of a generator stator according to an embodiment of the present invention, as Figure 2 shown. The process includes the following steps: Step S201, obtain the on-site actual condition data of the target pumped-storage generator, and construct an operating condition data set based on the on-site actual condition data of the target pumped-storage generator.

[0046] Specifically, the above step S201 includes: Step S2011, determine the steady-state condition type based on the on-site actual condition data of the target pumped-storage generator.

[0047] Specifically, obtain the condition signals (i.e., on-site actual condition data) collected by the on-site state monitoring system, and judge the steady-state condition type of the target pumped-storage generator according to the condition signals. Among them, the steady-state condition types of the target pumped-storage generator include six steady-state conditions: shutdown (S), spinning reserve (SR), power generation (G), power generation and phase modulation (GC), pumping (P), and pumping and phase modulation (PC).

[0048] Furthermore, the condition determination can also be performed according to the connected electrical parameter variables, as shown in Table 1 below.

[0049] Table 1:

[0050] Furthermore, the on-site actual condition data in the access state monitoring system is sent to the data acquisition box, transmitted to the data switch through the network cable, and then uploaded to the real-time simulation monitoring system arranged at the power station site through the network cable. The real-time simulation monitoring system can determine the corresponding condition type according to the real-time operating data of the on-site unit.

[0051] Further, historical data of the motor under different working conditions are obtained to provide sufficient basic data for the training and prediction of the model. The historical data of different working conditions include multi-dimensional features such as air temperature, winding current, and winding voltage in the power station status monitoring system, ensuring the comprehensiveness and representativeness of the data.

[0052] Step S2012: Clean, normalize, and correct the time stamps of the on-site actual working condition data of the target pumped-storage generator to obtain the operating condition data.

[0053] Specifically, to ensure the quality and reliability of the data, necessary preprocessing is performed on the collected original data (i.e., the operating condition data of the target pumped-storage generator), including data cleaning, normalization, and time stamp correction.

[0054] Further, data cleaning includes: handling missing values, outliers, and noise data; for missing values, interpolation, mean filling, or time-series-based prediction methods are used for completion; for outliers, through statistical analysis and threshold setting, data points deviating from the normal range are identified and removed. The specific steps for removing error points include: dividing the operating condition data of the target pumped-storage generator into segments, taking the data segment as an example, calculating the average value and standard deviation of this segment, setting the threshold range of this segment as and then removing the capacity values exceeding this threshold range.

[0055] Further, the normalization process uses the min-max normalization method, that is, scaling the data to the interval [0, 1]; using Z-score standardization to make the data have zero mean and unit variance.

[0056] Further, the steps for time stamp correction include: the power station status monitoring system records one data point per minute. Relying on manual correction and unification of the time stamps, the consistency of each feature data in the time dimension is ensured.

[0057] Step S2013: Construct an operating condition data set based on the operating condition data with the same steady-state working condition type.

[0058] Specifically, based on the above six steady-state working conditions, corresponding operating condition data sets are established respectively. The operating condition data set can be expressed as: where is the matrix of the corresponding relationship between the independent variable and the dependent variable at different moments under the corresponding working condition. The on-site data stores one corresponding data per second to form the corresponding data set; among them, taking the power generation working condition as an example, its corresponding data matrix is: (4) wherein, represents the number of variables.

[0059] Furthermore, based on the Latin hypercube sampling (LHS) experimental design method, the calculation formula for the number of sample units P is as follows: P = (N + 1)(N + 2) / 2 (5) Furthermore, the above dataset is randomly sampled using the random starting point equidistant sampling method. The specific steps include segmenting the samples in the dataset. The sample distance K = the total number of units G / the number of sample units P. On the premise that the population is divided into K segments, , first randomly select a sample unit from the 1st to Kth population units in the first segment, and then select a sample unit every K units until Y units are selected, obtaining the dataset for different operating conditions.

[0060] Furthermore, by combining with the on-site condition monitoring system, without adding new sensors, using the on-site data as the inlet boundary for driving the model, and considering constructing corresponding surrogate models under different operating conditions respectively, and adding a small amount of extreme operating condition data for training.

[0061] Step S202, construct an MLP neural network prediction model; wherein, the MLP neural network prediction model takes the air temperature at the outlet of the air cooler, the stator winding current, and the stator winding voltage in the operating condition dataset as independent variables, and takes the average stator winding temperature and the average stator core temperature as dependent variables. For details, please refer to Figure 1 Step S102 of the embodiment shown, which will not be elaborated here.

[0062] Step S203, use the hierarchical optimization and single-layer multi-objective optimization algorithm to optimize the parameters of the MLP neural network prediction model, obtaining the digital real-time simulation model of the generator stator. For details, please refer to Figure 1 Step S103 of the embodiment shown, which will not be elaborated here.

[0063] For the method for constructing a digital real-time simulation model of a generator stator provided in this embodiment, since the selection of hyperparameters of the MLP neural network prediction model is too random, the hyperparameters of the structural layer are first optimized, and after fixing the hyperparameters of the structural layer, the hyperparameters of the training layer are optimized, reducing the search space dimension, being able to provide a more efficient and intelligent search strategy, and improving the optimization efficiency and accuracy of the corresponding parameters of the MLP neural network prediction model.

[0064] In this embodiment, a method for constructing a digital real-time simulation model of a generator stator is provided, which can be used for the above-mentioned electronic device. Figure 3 It is a flowchart of a method for constructing a digital real-time simulation model of a generator stator according to an embodiment of the present invention, as Figure 3 shown. The process includes the following steps: Step S301: Obtain the on-site actual operating condition data of the target pumped-storage generator, and construct an operating condition data set based on the on-site actual operating condition data of the target pumped-storage generator. For details, please refer to Figure 2 Step S201 of the embodiment shown, which will not be elaborated here.

[0065] Step S302: Construct an MLP neural network prediction model; among them, the MLP neural network prediction model takes the air temperature at the outlet of the air cooler, the stator winding current, and the stator winding voltage in the operating condition data set as independent variables, and takes the average stator winding temperature and the average stator core temperature as dependent variables. For details, please refer to Figure 2 Step S202 of the embodiment shown, which will not be elaborated here.

[0066] Step S303: Use the hierarchical optimization and single-layer multi-objective optimization algorithm to optimize the parameters of the MLP neural network prediction model to obtain a digital real-time simulation model of the generator stator.

[0067] Specifically, divide the operating condition data set into an 80% training set and a 20% test set, adopt hierarchical optimization, divide the hyperparameters of the MLP neural network prediction model into structural layer hyperparameters and training layer hyperparameters, and optimize in stages, that is, first optimize the structural parameters, and then optimize the training parameters after fixing, so as to reduce the search space dimension.

[0068] Among them, the above step S303 includes: Step S3031: Optimize the structural layer hyperparameters of the MLP neural network prediction model based on the operating condition data set to obtain an optimized MLP neural network prediction model; among them, the structural layer hyperparameters include the number of hidden layers of the MLP neural network and the number of neurons in each hidden layer.

[0069] In some optional implementation manners, the above step S3031 includes: Step a1: Take the MLP neural network prediction model as the objective function to be optimized, take the structural layer hyperparameters as independent variables, and take the average relative error between the average predicted stator winding temperature of the MLP neural network prediction model and the average stator winding temperature in the operating condition data set as the target result.

[0070] Specifically, take the MLP neural network prediction model as the objective function to be optimized , the objective function to be optimized can be expressed as: (6) Further, the average relative error between the average stator winding temperature predicted by the MLP neural network prediction model and the average stator winding temperature in the test set is used as the target result; alternatively, the average relative error between the average stator core temperature predicted by the MLP neural network prediction model and the average stator core temperature in the test set can be used as the target result.

[0071] Step a2, randomly select multiple groups of structure layer hyperparameter groups as initial sample points, and use the objective function to be optimized to determine the average relative error corresponding to the initial sample points.

[0072] Specifically, the selected initial sample points are used as the initial structure layer hyperparameters in the objective function to be optimized, and the independent variable data in the operating condition dataset is input into the objective function to be optimized to obtain the average predicted stator winding temperature. The average relative error is calculated based on the average predicted stator winding temperature and the average stator winding temperature in the operating condition dataset.

[0073] Step a3, based on the initial sample points and the average relative error, use Gaussian process to fit the surrogate probability function to obtain the variance of each group of structure layer hyperparameter groups.

[0074] Specifically, use Gaussian process to fit the surrogate probability function, establish the probability distribution corresponding to the objective function to be optimized, and obtain the expected mean and variance of each group of structure layer hyperparameter groups. Among them, the surrogate probability function can be expressed as: (7) where represents the expected mean, and the expected mean represents the final expected effect of the structure layer hyperparameter group. represents the variance, and the variance represents the uncertainty of the effect of the structure layer hyperparameter group. The larger the variance, the greater the uncertainty of the effect of the structure layer hyperparameter group.

[0075] Step a4, compare the variance of each group of structure layer hyperparameter groups with the variance threshold. If the variance of the structure layer hyperparameter group is greater than the variance threshold, sample the structure layer hyperparameter group using the first acquisition function to obtain updated sample points.

[0076] Specifically, the first acquisition function uses the EI probability function. If the variance of the structure layer hyperparameter group is greater than the variance threshold, it is necessary to further accurately fit the surrogate probability function, that is, use the EI probability function to sample in the region with a larger variance to obtain updated sample points.

[0077] Further, the improvement of the initial sampling point on the current optimal objective function value is expressed as: (8) Furthermore, by maximizing searching for the next sample point, the updated sample point can be expressed as: (9) where and represent the current sampling point and the next sampling point respectively, is the search space of the structural layer hyperparameters, represents the optimal function value of the current objective function, represents the expectation for the conditional distribution taking the expectation.

[0078] Step a5, re-evaluate the objective function to be optimized using the updated sample points until the average relative error is less than the preset error threshold, and obtain the optimal structural layer hyperparameter set.

[0079] Specifically, repeat the above steps a2 and a3 based on the updated sample points, so that the surrogate probability function continuously approximates the objective function. Through continuous iteration of the above process, the optimal structural layer hyperparameter set can be obtained.

[0080] Furthermore, when the average relative error of the objective function to be optimized changes very little or reaches the preset threshold in several consecutive iterations, it can be considered that convergence has occurred, and at this time, the optimization is stopped, that is, when the average relative error between the average stator winding temperature predicted by the MLP neural network prediction model and the average stator winding temperature in the test set is less than 10%, the optimization is stopped.

[0081] Step a6, determine the optimized MLP neural network prediction model based on the optimal structural layer hyperparameter set.

[0082] Specifically, the optimal structural layer hyperparameter combination is obtained through the fitted surrogate probability function, that is, the primary surrogate model of the neural network is constructed. The number of hidden layers of the MLP neural network and the number of neurons in each hidden layer in the optimal structural layer hyperparameter set are used as fixed values of the MLP neural network prediction model to obtain the optimized MLP neural network prediction model.

[0083] The construction method of a generator stator digital real-time simulation model provided in this embodiment effectively explores the parameter space through the surrogate probability function, reduces the number of direct calculations of the objective function, is suitable for computationally expensive optimization problems, can well balance exploring unknown regions and exploiting known regions through the acquisition function, avoids falling into local optima, and the Gaussian process can provide prediction uncertainty, which helps to better guide the sampling process, making the optimal structural layer hyperparameter set more accurate and improving the parameter tuning efficiency of the MLP neural network prediction model.

[0084] Step S3032: Based on the operating condition data set, optimize the hyperparameters of the training layer of the optimized MLP neural network prediction model to obtain a digital real-time simulation model of the generator stator; wherein, the hyperparameters of the training layer include the learning rate, learning decay coefficient, number of training times, and regularization coefficient.

[0085] Specifically, in combination with the warm-start strategy, use the optimal structure layer hyperparameter group in the first stage as the prior knowledge in the second stage, and adopt multi-objective Bayesian optimization to optimize the hyperparameters of the training layer while optimizing the model performance (such as accuracy), computational cost (such as the number of parameters), and training efficiency (such as time).

[0086] In some optional embodiments, the above step S3032 includes: Step b1: Randomly select hyperparameters of the training layer to establish multiple groups of hyperparameter groups of the training layer, and use the optimized MLP neural network prediction model to determine the target values corresponding to the multiple groups of hyperparameter groups of the training layer.

[0087] Step b2: Use the multiple groups of hyperparameter groups of the training layer and the target values as initial data, and use the initial data to train the multi-objective surrogate model to establish the joint distribution of the objective function.

[0088] Specifically, the multi-objective surrogate model is composed of multiple objective functions, and the multi-objective surrogate model can be expressed as: (10) wherein, = 1 - validation set accuracy (minimize error); = number of model parameters (minimize complexity); = training time (minimize time).

[0089] Specifically, randomly sample a small number (e.g., 10) of groups of hyperparameter groups of the training layer, train the optimized MLP neural network prediction model and record the multi-objective values, and then use the multi-objective optimization algorithm to solve the approximation of the Pareto front on the multi-objective surrogate model to obtain a set of non-dominated solutions, that is, an approximate set of Pareto optimal solutions.

[0090] Furthermore, use the initial data to train the multi-objective surrogate model (such as multi-output Gaussian process) to model the joint distribution of the objective function. Through the joint distribution, the probability density of the approximate set of Pareto optimal solutions in the multi-objective space can be calculated, so as to evaluate its closeness to the Pareto front. Solutions with higher probability density are usually closer to the Pareto front and have relatively higher quality.

[0091] Step b3: Evaluate the multiple groups of hyperparameter groups of the training layer using the joint distribution of the objective function. Based on the evaluation results, use the second acquisition function to select the next hyperparameter group of the training layer to be evaluated.

[0092] Specifically, if the joint distribution of the objective function does not meet the preset conditions, the next training layer hyperparameter group to be evaluated is selected through the second acquisition function; among them, the second acquisition function can adopt an acquisition function based on EHVI (Expected Hypervolume Improvement).

[0093] Step b4: Re-evaluate the next training layer hyperparameter group to be evaluated until the current number of iterations reaches the preset number of iterations, and obtain the optimal training layer hyperparameter group.

[0094] Specifically, repeat the above steps b2 and b3 iteratively based on the next training layer hyperparameter group to be evaluated, train the optimized MLP neural network prediction model and calculate the multi-objective values, and iteratively update the data set until the preset number of iterations (such as 100 iterations) is reached or the Pareto front converges.

[0095] Step b5: Input the optimal training layer hyperparameter group into the optimized MLP neural network prediction model to obtain the generator stator digital real-time simulation model.

[0096] Specifically, through the hierarchical optimization method, the fitted surrogate probability function obtains the optimal hyperparameter combination, and the generator stator digital real-time simulation model is constructed. That is, the optimal training layer hyperparameter group is input into the optimized MLP neural network prediction model, and the optimized MLP neural network prediction model of the training layer is used as the generator stator digital real-time simulation model.

[0097] The construction method of a generator stator digital real-time simulation model provided in this embodiment optimizes the training layer hyperparameters of the optimized MLP neural network prediction model, and substitutes the optimal structural layer hyperparameter group into the optimization process of the training layer hyperparameters, which can guide the search process to converge to the optimal solution faster, improve the optimization effect and efficiency, and by constructing the joint distribution of the objective function, comprehensively consider the information of each objective, provide a set of optimal solutions, provide more choice space for decision-makers, help to find the optimal trade-off solution that meets different needs and preferences, and can adaptively adjust the search direction in the entire search space, improving the parameter optimization efficiency and accuracy of the generator stator digital real-time simulation model.

[0098] In this embodiment, a construction method of a generator stator digital real-time simulation model is provided, which can be used for the above-mentioned electronic device. Figure 4 It is a flowchart of a construction method of a generator stator digital real-time simulation model according to an embodiment of the present invention, as Figure 4 shown, and this process includes the following steps: Step S401: Obtain the on-site actual operating condition data of the target pumped-storage generator, and construct an operating condition data set based on the on-site actual operating condition data of the target pumped-storage generator. For details, please refer to Figure 1 Step S301 of the embodiment shown, which will not be elaborated here.

[0099] Step S402: Construct an MLP neural network prediction model; among them, the MLP neural network prediction model takes the air temperature at the outlet of the air cooler, the stator winding current, and the stator winding voltage in the operating condition data set as independent variables, and takes the average stator winding temperature and the average stator core temperature as dependent variables. For details, please refer to Figure 1 Step S302 of the embodiment shown, which will not be elaborated here.

[0100] Step S403: Use the hierarchical optimization and single-layer multi-objective optimization algorithm to optimize the parameters of the MLP neural network prediction model to obtain a digital real-time simulation model of the generator stator. For details, please refer to Figure 1 Step S303 of the embodiment shown, which will not be elaborated here.

[0101] Step S404: Evaluate the digital real-time simulation model of the generator stator, and optimize the model of the digital real-time simulation model of the generator stator based on the evaluation results to obtain an optimized digital real-time simulation model of the generator stator.

[0102] Specifically, randomly select the on-site condition data of n target pumped-storage generators and the corresponding output temperature values; input the on-site condition data into the digital real-time simulation model of the generator stator to obtain the predicted stator temperature values; calculate the average error value based on the predicted stator temperature values and the output temperature values, and compare the average error value with the preset efficiency value; if the average error value exceeds 10% of the preset efficiency value, then add the sampling points with larger errors in the n on-site conditions to the training set, and retrain the digital real-time simulation model of the generator stator to optimize the digital real-time simulation model of the generator stator; if the average error value does not exceed 10% of the preset efficiency value, then output the digital real-time simulation model of the generator stator.

[0103] Furthermore, the calculation formula of the average error value is as follows: (11) (12) Where represents the error, represents the output temperature value, represents the predicted stator temperature value, represents the average error value.

[0104] A construction method of a digital real-time simulation model for a generator stator, by evaluating the digital real-time simulation model of the generator stator, and then optimizing the model of the digital real-time simulation model of the generator stator, improves the accuracy and generalization ability of the digital real-time simulation model of the generator stator.

[0105] In this embodiment, a construction device for a digital real-time simulation model of a generator stator is also provided. This device is used to implement the above embodiments and preferred implementation manners, and those that have been described will not be repeated. As used hereinafter, the term "module" can be a combination of software and / or hardware that realizes a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0106] This embodiment provides a construction device for a digital real-time simulation model of a generator stator, as Figure 5 shown, including: An acquisition module 501, configured to acquire on-site actual operating condition data of a target pumped-storage generator, and construct an operating condition data set based on the on-site actual operating condition data of the target pumped-storage generator; A construction module 502, configured to construct an MLP neural network prediction model; wherein, the MLP neural network prediction model uses the air temperature at the outlet of the air cooler, the stator winding current, and the stator winding voltage in the operating condition data set as independent variables, and uses the average stator winding temperature and the average stator core temperature as dependent variables; A parameter tuning module 503, configured to perform parameter tuning on the MLP neural network prediction model based on the operating condition data set by using a hierarchical optimization and single-layer multi-objective optimization algorithm to obtain a digital real-time simulation model of the generator stator.

[0107] In some optional implementation manners, the acquisition module 501 includes: A determination unit, configured to determine a steady-state operating condition type based on the on-site actual operating condition data of the target pumped-storage generator; A processing unit, configured to perform data cleaning, normalization processing, and timestamp correction on the on-site actual operating condition data of the target pumped-storage generator to obtain operating condition data; A construction unit, configured to construct an operating condition data set based on the operating condition data with the same steady-state operating condition type.

[0108] In some optional implementation manners, the parameter tuning module 503 includes: A structure layer hyperparameter optimization unit, configured to perform structure layer hyperparameter optimization on the MLP neural network prediction model based on the operating condition data set to obtain an optimized MLP neural network prediction model; wherein, the structure layer hyperparameters include the number of hidden layers of the MLP neural network and the number of neurons in each hidden layer; A training layer hyperparameter optimization unit is used to optimize the hyperparameters of the training layer of the optimized MLP neural network prediction model based on the operating condition dataset to obtain a digital real-time simulation model of the generator stator; wherein, the hyperparameters of the training layer include the learning rate, the learning decay coefficient, the number of training times, and the regularization coefficient.

[0109] In some alternative embodiments, the structure layer hyperparameter optimization unit includes: An acquisition subunit is used to take the MLP neural network prediction model as the objective function to be optimized, the structure layer hyperparameters as independent variables, and the average relative error between the average predicted temperature of the stator winding and the average temperature of the stator winding in the operating condition dataset as the target result; A first selection subunit is used to randomly select multiple groups of structure layer hyperparameter groups as initial sample points, and use the objective function to be optimized to determine the average relative error corresponding to the initial sample points; A fitting subunit is used to fit the surrogate probability function by Gaussian process based on the initial sample points and the average relative error to obtain the variance of each group of structure layer hyperparameter groups; A comparison subunit is used to compare the variance of each group of structure layer hyperparameter groups with the variance threshold. If the variance of the structure layer hyperparameter group is greater than the variance threshold, the structure layer hyperparameter group is sampled using the first acquisition function to obtain updated sample points; A first evaluation subunit is used to iteratively evaluate the objective function to be optimized using the updated sample points until the average relative error is less than the preset error threshold to obtain the optimal structure layer hyperparameter group; A first determination subunit is used to determine the optimized MLP neural network prediction model based on the optimal structure layer hyperparameter group.

[0110] In some alternative embodiments, the training layer hyperparameter optimization unit includes: A second selection subunit is used to randomly select training layer hyperparameters to establish multiple groups of training layer hyperparameter groups, and use the optimized MLP neural network prediction model to determine the target values corresponding to the multiple groups of training layer hyperparameter groups; A training subunit is used to use the multiple groups of training layer hyperparameter groups and the target values as initial data to train the multi-objective surrogate model to establish the joint distribution of the objective function; A third selection subunit is used to evaluate the multiple groups of training layer hyperparameter groups using the joint distribution of the objective function, and based on the evaluation results, select the next training layer hyperparameter group to be evaluated using the second acquisition function; A second evaluation subunit is used to re-evaluate the next training layer hyperparameter group to be evaluated until the current iteration number reaches the preset iteration number to obtain the optimal training layer hyperparameter group; A second determination subunit, configured to input the optimal training layer hyperparameter group into the optimized MLP neural network prediction model to obtain a digital real-time simulation model of the generator stator.

[0111] In some alternative embodiments, it further includes: An evaluation module, configured to evaluate the digital real-time simulation model of the generator stator, and optimize the model of the digital real-time simulation model of the generator stator based on the evaluation result to obtain an optimized digital real-time simulation model of the generator stator.

[0112] The further function descriptions of the above-mentioned various modules and units are the same as those in the corresponding foregoing embodiments, and will not be elaborated herein.

[0113] The construction device of a digital real-time simulation model of a generator stator in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0114] An embodiment of the present invention further provides a computer device having the above-mentioned Figure 5 construction device of a digital real-time simulation model of a generator stator as shown.

[0115] Please refer to Figure 6 , Figure 6 which is a schematic structural diagram of a computer device provided by an alternative embodiment of the present invention. As shown in Figure 6 , the computer device includes: one or more processors 10, a memory 20, and an interface for connecting each component, including a high-speed interface and a low-speed interface. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (such as an array of servers, a set of blade servers, or a multi-processor system). Figure 6 One processor 10 is taken as an example in

[0116] The processor 10 can be a central processing unit, a network processor, or a combination thereof. Among them, the processor 10 can further include a hardware chip. The above-mentioned hardware chip can be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The above-mentioned programmable logic device can be a complex programmable logic device, a field programmable gate array, a generic array logic, or any combination thereof.

[0117] Among them, the memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiments.

[0118] The memory 20 can include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the computer device, etc. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 can optionally include a memory remotely set relative to the processor 10, and these remote memories can be connected to the computer device through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0119] The memory 20 can include a volatile memory, such as a random access memory; the memory can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state drive; the memory 20 can also include a combination of the above types of memories.

[0120] The computer device further includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30, and the output device 40 can be connected through a bus or other means, Figure 6 Taking the connection through the bus as an example.

[0121] The input device 30 can receive input digital or character information, and generate key signal inputs related to the user settings and function controls of the computer device, such as a touch screen, a keypad, a mouse, a trackpad, a touchpad, a pointing stick, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 can include a display device, an auxiliary lighting device (such as an LED), and a tactile feedback device (such as a vibration motor), etc. The above display device includes but is not limited to a liquid crystal display, a light-emitting diode, a display, and a plasma display. In some alternative embodiments, the display device can be a touch screen.

[0122] Embodiments of the present invention also provide a computer-readable storage medium. The methods according to the embodiments of the present invention can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented as computer code that is originally stored in a remote storage medium or a non-transitory machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the methods described herein can be stored as such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code, and when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the methods shown in the above embodiments are implemented.

[0123] A part of the present invention can be applied as a computer program product, for example, computer program instructions, which when executed by a computer, can call or provide the methods and / or technical solutions according to the present invention through the operation of the computer. Those skilled in the art should be able to understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to the computer.

[0124] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for constructing a digital real-time simulation model of a generator stator, characterized in that: The method comprises: Acquire actual on-site operating condition data of the target pumped-storage generator, and construct an operating condition data set based on the actual on-site operating condition data of the target pumped-storage generator; Constructing an MLP neural network prediction model; wherein the MLP neural network prediction model takes the air cooler outlet air temperature, stator winding current and stator winding voltage in the operating condition data set as independent variables, and takes the stator winding average temperature and the stator core average temperature as dependent variables; Based on the operating condition data set, the parameters of the MLP neural network prediction model are tuned using hierarchical optimization and single-layer multi-objective optimization algorithms to obtain a digital real-time simulation model of the generator stator.

2. The method according to claim 1, characterized in that The constructing of the operating condition data set based on the actual on-site operating condition data of the target pumped-storage generator comprises: Determining a steady-state operating condition type based on actual on-site operating condition data of the target pumped-storage generator; Performing data cleaning, normalization processing and timestamp correction on the actual on-site operating condition data of the target pumped-storage generator to obtain operating condition data; The operating condition data set is constructed based on the operating condition data having the same steady-state condition type.

3. The method according to claim 1, characterized in that Based on the operating condition data set, the MLP neural network prediction model is tuned by using a hierarchical optimization and a single-layer multi-objective optimization algorithm to obtain a generator stator digital real-time simulation model, including: Based on the operating condition data set, the structural layer hyperparameters of the MLP neural network prediction model are optimized to obtain an optimized MLP neural network prediction model; wherein the structural layer hyperparameters include the number of hidden layers of the MLP neural network and the number of neurons in each hidden layer; Based on the operating condition data set, the training layer hyperparameters of the optimized MLP neural network prediction model are optimized to obtain the generator stator digital real-time simulation model; wherein the training layer hyperparameters include learning rate, learning attenuation coefficient, number of training times and regularization coefficient.

4. The method according to claim 3, characterized in that The method of performing structural layer hyperparameter optimization on the MLP neural network prediction model based on the operating condition data set to obtain an optimized MLP neural network prediction model includes: The MLP neural network prediction model is used as the objective function to be optimized, the structural layer hyperparameter is used as an independent variable, and the average relative error between the average predicted temperature of the stator winding and the average temperature of the stator winding in the operating condition data set is used as the target result; Randomly select multiple groups of structural layer hyperparameter groups as initial sample points, and use the objective function to be optimized to determine the average relative errors corresponding to the initial sample points; Based on the initial sample points and the average relative error, a Gaussian process is used to fit the proxy probability function to obtain the variance of each group of structural layer hyperparameter groups; Compare the variance of each group of structure layer hyperparameter groups with the variance threshold, and if the variance of the structure layer hyperparameter group is greater than the variance threshold, sample the structure layer hyperparameter group using the first acquisition function to obtain updated sample points; Iteratively evaluating the objective function to be optimized using the updated sample points until the average relative error is less than a preset error threshold, thereby obtaining an optimal structural layer hyperparameter group; The optimized MLP neural network prediction model is determined based on the optimal structure layer hyperparameter group.

5. The method according to claim 4, characterized in that The method of performing training layer hyperparameter optimization on the optimized MLP neural network prediction model based on the operating condition data set to obtain the generator stator digital real-time simulation model includes: Randomly selecting the training layer hyperparameters to establish multiple sets of training layer hyperparameter groups, and using the optimized MLP neural network prediction model to determine target values ​​corresponding to the multiple sets of training layer hyperparameter groups; Using the multiple sets of training layer hyperparameter groups and the target values ​​as initial data, training a multi-objective proxy model using the initial data, and establishing a joint distribution of the objective function; The plurality of training layer hyperparameter groups are evaluated using the joint distribution of the objective function, and based on the evaluation result, a next training layer hyperparameter group to be evaluated is selected using the second acquisition function; Re-evaluate the next training layer hyperparameter group to be evaluated until the current number of iterations reaches a preset number of iterations, and obtain the optimal training layer hyperparameter group; The optimal training layer hyperparameter group is input into the optimized MLP neural network prediction model to obtain the generator stator digital real-time simulation model.

6. The method according to claim 1, characterized in that Also includes: The generator stator digital real-time simulation model is evaluated, and based on the evaluation result, the generator stator digital real-time simulation model is optimized to obtain an optimized generator stator digital real-time simulation model.

7. A device for constructing a digital real-time simulation model of a generator stator, characterized in that: The device comprises: An acquisition module, used for acquiring actual on-site operating condition data of a target pumped-storage generator, and constructing an operating condition data set based on the actual on-site operating condition data of the target pumped-storage generator; A construction module is used to construct an MLP neural network prediction model; wherein the MLP neural network prediction model uses the air cooler outlet air temperature, stator winding current and stator winding voltage in the operating condition data set as independent variables, and uses the stator winding average temperature and the stator core average temperature as dependent variables; The parameter tuning module is used to tune the parameters of the MLP neural network prediction model based on the operating condition data set by using hierarchical optimization and single-layer multi-objective optimization algorithms to obtain a digital real-time simulation model of the generator stator.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method for constructing a digital real-time simulation model of a generator stator according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method for constructing a digital real-time simulation model of a generator stator according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises computer instructions, wherein the computer instructions are used to enable a computer to execute the method for constructing a digital real-time simulation model of a generator stator according to any one of claims 1 to 6.

Images