Method and device for training power plant operating parameter prediction model
By preprocessing and online training of the power unit's operating parameters, combined with multi-threading technology and BP neural network optimization, the problem of accuracy in predicting power unit operating data under varying operating conditions was solved, achieving higher prediction accuracy and safety.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA STATE SHIPBUILDING CORP LTD RESEARCH INSTITUTE 719
- Filing Date
- 2024-06-05
- Publication Date
- 2026-04-10
AI Technical Summary
Existing technologies train a single neural network by extracting data sequences of power plant operating parameters over a historical period. This cannot meet the needs of predicting operating data under varying operating conditions, resulting in a mismatch between the prediction model and the power plant, and significant deviations in trend prediction results, which affects the safety of the ship's power plant.
Multithreading technology is used to control the parallel operation of the first and second neural networks. By preprocessing and online training of multiple operating parameters, the event sequence is transformed using a sliding time window algorithm, the weight matrix is dynamically updated, and the training of the dynamic prediction model is achieved by combining the optimization of the number of hidden layer nodes and the learning rate of the BP neural network.
It improves the accuracy of predicting the operating data of the power plant under varying operating conditions, enhances the model's adaptability to real-time changes in operating conditions, reduces the risk of misoperation, and improves the safety of the ship's power plant.
Smart Images
Figure CN118859765B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, and in particular to a training method and apparatus for a predictive model of operating parameters of a power plant. Background Technology
[0002] Marine propulsion systems are highly complex, involving hundreds of different types and quantities of equipment. To effectively control the operation of these systems under normal conditions, it is necessary to measure various operating parameters. Changes in these parameters directly reflect the operational status of the propulsion system; therefore, studying the trends of these parameters is crucial and meaningful.
[0003] In related technologies, a single neural network is trained by extracting data sequences of power plant operating parameters over a historical period and converting them into a training sample set of a certain size. Once the training metrics meet the requirements, the prediction model is used to predict subsequent time series. However, since the mapping relationship between samples that are closer in time is more similar, the prediction model's ability to track the input-output relationship weakens over time. Furthermore, the real-time operating conditions of ship power plants change frequently, and the parameters fluctuate significantly under changing conditions. Moreover, it is impossible to obtain operating data under changing conditions in advance, preventing the model from learning prior knowledge under changing conditions. This leads to a mismatch between the trend prediction model and the power plant, resulting in significant deviations in the trend prediction results. This can mislead operators, causing erroneous operations and affecting the safety of the ship's power plant.
[0004] Invention content / Utility model content
[0005] This invention provides a training method and apparatus for a power plant operating parameter prediction model, which solves the problem that the prediction model obtained by training a single neural network by extracting a data sequence of power plant operating parameters over a historical period cannot meet the needs of predicting operating data under varying operating conditions, and improves the accuracy of predicting operating data of power plants under varying operating conditions.
[0006] This invention provides a training method for a power plant operating parameter prediction model, comprising:
[0007] Multiple operating parameters of the power unit are acquired, with different operating parameters corresponding to different timestamps, and the multiple operating parameters are arranged in chronological order of the timestamps.
[0008] The multiple operating parameters are preprocessed, and the first neural network is trained online based on the preprocessed operating parameters. After the online training time reaches the target duration, the weight matrix output by the first neural network is sent to the second neural network. The second neural network is used to dynamically predict the operating parameters of the power unit.
[0009] Discard the running parameter corresponding to the shortest timestamp from the plurality of running parameters, and add a running parameter corresponding to a target time of the power device to obtain updated plurality of running parameters, and dynamically update the weight matrix of the output of the first neural network according to the updated plurality of running parameters and the prediction step corresponding to the second neural network to obtain the power device running parameter prediction model; the target time is greater than the maximum value of the plurality of running parameters corresponding to the timestamp.
[0010] According to the power device running parameter prediction model training method provided by the application, the preprocessing of the plurality of running parameters comprises:
[0011] The plurality of running parameters are preprocessed by the following formula:
[0012]
[0013] Wherein, u(z) is the preprocessed running parameter, z is the running parameter, h and g are constant terms, the values of h and g are in the range of [-1, 1], and g < h, x max is the maximum value of the running parameter, and x min is the minimum value of the running parameter.
[0014] According to the power device running parameter prediction model training method provided by the application, the first neural network and the second neural network adopt the same network structure; the method further comprises:
[0015] The first neural network and the second neural network are controlled to run side by side based on multi-thread technology, and the network parameters of the first neural network are adjusted according to the prediction result output by the second neural network.
[0016] According to the power device running parameter prediction model training method provided by the application, after the preprocessing of the plurality of running parameters, the method further comprises:
[0017] In the case that the number of preprocessed plurality of running parameters exceeds the first number threshold, the plurality of running parameters are converted into event sequences according to the sliding time window algorithm, and the first neural network is trained online according to the event sequences.
[0018] According to the power device running parameter prediction model training method provided by the application, after the plurality of running parameters are converted into event sequences according to the sliding time window algorithm, the method further comprises:
[0019] In the case that the number of samples in the event sequence is greater than or equal to the preset training sample number, the first neural network is trained online based on the event sequence.
[0020] In a case that the number of samples in the event sequence is greater than a second number threshold, a single-step prediction is performed on the event sequence based on a second neural network, and the single-step prediction result is displayed and stored; the second number threshold is greater than the first number threshold.
[0021] According to the power device operation parameter prediction model provided by the application, after the power device operation parameter prediction model is obtained, the method further comprises:
[0022] The to-be-tested operation parameter is input into the power device operation parameter prediction model for single-step prediction, and a plurality of prediction results are obtained.
[0023] The plurality of prediction results are respectively counted according to a target performance index, and a target prediction result is obtained, the target prediction result being one of the plurality of prediction results; the target performance index comprising at least one of mean square error (MSE) and average value of absolute value of relative error (AMPE).
[0024] According to the power device operation parameter prediction model provided by the application, the first neural network and the second neural network are both BP neural networks, and the number of hidden layer nodes, the number of input layer nodes and the number of output layer nodes of the BP neural network satisfy the following relationship:
[0025]
[0026] wherein, l is the number of hidden layer nodes, m is the number of input layer nodes, n is the number of output layer nodes, and a is a constant, the value of a being between 1 and 10;
[0027] The learning rate of the BP neural network is obtained by trial and error method.
[0028] The application further provides a training device of a power device operation parameter prediction model, comprising:
[0029] A parameter acquisition module is configured to acquire a plurality of operation parameters of a power device, different operation parameters corresponding to different time stamps, and the plurality of operation parameters being arranged in chronological order according to the time stamps;
[0030] A first processing module is configured to pre-process the plurality of operation parameters, and perform online training on a first neural network according to the pre-processed plurality of operation parameters, and send a weight matrix of an output of the first neural network to a second neural network after a target time length is reached during the online training; the second neural network is configured to perform dynamic prediction on the operation parameters of the power device.
[0031] The second processing module is configured to remove the running parameter corresponding to the shortest timestamp from the plurality of running parameters, add a running parameter corresponding to a target time of the power device, obtain updated plurality of running parameters, and dynamically update the weight matrix of the output of the first neural network according to the updated plurality of running parameters and a prediction step corresponding to the second neural network, to obtain the power device running parameter prediction model; and the target time is greater than the maximum value of the plurality of time stamps corresponding to the plurality of running parameters.
[0032] The application further provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the training method of the power device running parameter prediction model according to any one of the above when executing the program.
[0033] The application further provides a non-transitory computer-readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement the training method of the power device running parameter prediction model according to any one of the above.
[0034] The application further provides a computer program product, which includes a computer program, and the computer program is executable on a processor to implement the training method of the power device running parameter prediction model according to any one of the above.
[0035] The training method and device of the power device running parameter prediction model provided by the application improve the prediction accuracy of the running data of the power device under variable working conditions by preprocessing the plurality of running parameters, training the first neural network according to the plurality of preprocessed running parameters, sending the weight matrix of the output of the first neural network to the second neural network when the online training time reaches the target time length, removing the running parameter corresponding to the shortest timestamp from the plurality of running parameters, adding the running parameter corresponding to the target time of the power device, obtaining the updated plurality of running parameters, and dynamically updating the weight matrix of the output of the first neural network according to the updated plurality of running parameters and the prediction step corresponding to the second neural network, to obtain the power device running parameter prediction model. BRIEF DESCRIPTION OF DRAWINGS
[0036] In order to more clearly illustrate the technical solutions of the application or the prior art, the following will briefly introduce the drawings needed in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.
[0037] Figure 1 is one of the flowcharts of the training method of the power device running parameter prediction model provided by the application;
[0038] Figure 2 Figure 2 is a flowchart of a method for training a power plant operating parameter prediction model according to the present application;
[0039] Figure 3 Figure 3 is a schematic diagram of a training device for a power plant operating parameter prediction model according to the present application;
[0040] Figure 4 Figure 4 is a schematic diagram of an electronic device according to the present application. DETAILED DESCRIPTION
[0041] In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be described clearly and completely below with reference to the drawings of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.
[0042] The method and device for training a power plant operating parameter prediction model according to the present application will be described below. Figures 1-3
[0043] Figure 1 Figure 1 is a flowchart of a method for training a power plant operating parameter prediction model according to the present application. As shown in Figure 1, the method for training a power plant operating parameter prediction model comprises the following steps: Figure 1
[0044] In step 110, a plurality of operating parameters of a power plant are obtained, different operating parameters corresponding to different time stamps, and the plurality of operating parameters being arranged in chronological order according to the time stamps.
[0045] In this step, the power plant includes a steam boiler and an engine of a ship, etc., and the operating parameters corresponding to the power plant include factors affecting the water level of the steam boiler, such as feedwater flow, feedwater temperature, reactor coolant temperature, steam flow, etc.
[0046] In this embodiment, the plurality of operating parameters include historical data of the operating parameters of the power plant, such as operating parameters collected in the past few weeks, days or hours, and also include real-time collected operating parameters.
[0047] In this embodiment, each operating parameter corresponds to a time stamp, for example, the feedwater flow and the feedwater temperature of the water level of the steam boiler of the power plant are collected every time t.
[0048] Step 120, pre-process the plurality of operating parameters, and perform online training on the first neural network according to the pre-processed plurality of operating parameters, and send the weight matrix of the output of the first neural network to the second neural network after the online training time reaches the target length; the second neural network is used for dynamically predicting the operating parameters of the power device.
[0049] In this step, the first neural network includes a back propagation neural network (BPNN), and can also be a GBDT (Gradient Boosting Decision Tree) random forest algorithm, etc.
[0050] In this embodiment, the target length can be set according to user demand, for example, the target length can be 100 milliseconds (ms).
[0051] In this embodiment, the second neural network can be the same as or different from the first neural network; for example, the second neural network also includes a BPNN, a GBDT or a random forest algorithm, etc.
[0052] In this embodiment, the online training of the first neural network means updating the training data input to the first neural network by updating samples at different time sequences, and then updating the network parameters of the first neural network, and according to the prediction result of the second neural network and the parameters updated in real time by the first neural network, the first neural network is updated in real time to ensure that the first neural network can learn the operating parameter information containing effective time sequence.
[0053] In this step, the pre-processing includes normalizing the operating parameters, removing outliers and other data cleaning methods.
[0054] In this embodiment, the pre-processing of the plurality of operating parameters includes pre-processing the plurality of operating parameters by the following formula:
[0055]
[0056] Wherein, u(z) is the pre-processed operating parameter, z is the operating parameter, h and g are constant terms, the values of h and g are in the range of [-1, 1], and g < h, x max is the maximum value of the operating parameter, x min is the minimum value of the operating parameter.
[0057] In this embodiment, the activation function used by the first neural network and the second neural network is a bipolar Sigmoid function, as shown in the following formula:
[0058] c is a constant.
[0059] Step 130, the running parameter corresponding to the shortest timestamp is removed from the plurality of running parameters, and the running parameter corresponding to the target time of the power device is added to obtain updated plurality of running parameters, the weight matrix of the output of the first neural network is dynamically updated according to the updated plurality of running parameters and the prediction step corresponding to the second neural network, and the power device running parameter prediction model is obtained; the target time is greater than the maximum value of the plurality of running parameters corresponding to the timestamp.
[0060] In this step, the target time can be set according to user demand.
[0061] In this embodiment, the dynamic prediction of the power device running parameter time sequence is realized according to the online BP neural network (Online Back Propagation Neural Network, Online BPNN) prediction model, and the matching degree of the trend prediction model and the power device under variable working conditions is improved.
[0062] Specifically, dynamic prediction means that the prediction model will be dynamically updated over time, and the update of the prediction model depends on the dynamic update of the training sample set and the online training of the neural network.
[0063] In this embodiment, the power device prediction model of the ship adopts two BP neural networks, one of which is used for online training, and the other is used for prediction; in addition, since the time interval for data sampling from the simulation software is ns (for example, n=5), it means that the training sample set can be updated once every ns, that is, the first sample in the training sample set is removed and the latest sample is added to the training sample set, and at the same time, the training of the neural network and the prediction of the time sequence are realized within the ns.
[0064] In this embodiment, in order to effectively utilize the ns time, a multi-threading technology is adopted, that is, the neural network training of the sub-thread and the prediction of the main thread run side by side.
[0065] The training method of the power device running parameter prediction model provided by the embodiment of the application, by preprocessing the plurality of running parameters, and according to the plurality of running parameters after preprocessing, the first neural network is trained online, after the online training time reaches the target time, the weight matrix of the output of the first neural network is sent to the second neural network, and the running parameter corresponding to the shortest timestamp is removed from the plurality of running parameters, and the running parameter corresponding to the target time of the power device is added to obtain updated plurality of running parameters, the weight matrix of the output of the first neural network is dynamically updated according to the updated plurality of running parameters and the prediction step corresponding to the second neural network, and the power device running parameter prediction model is obtained, which improves the prediction accuracy of the running data of the power device under variable working conditions.
[0066] In some embodiments, the first neural network and the second neural network adopt the same network structure; the method further comprises: controlling the first neural network and the second neural network to run in parallel based on a multi-thread technology, and adjusting the network parameters of the first neural network according to the prediction result output by the second neural network.
[0067] In this embodiment, the first neural network and the second neural network can adopt the same network structure, for example, the first neural network and the second neural network are both BP neural networks with the same number of network layers.
[0068] In this embodiment, by adopting two BP neural networks with the same structure, one is used for online training, and the other is used for prediction; in addition, since the time interval for sampling data from the simulation software is ns, it means that the training sample set can be updated once every ns, that is, the first sample in the training sample set is removed and the latest sample is added to the training sample set, and at the same time, the training of the neural network and the prediction of the time series are realized within the ns, in order to effectively utilize the ns time, the multi-thread technology is adopted in the programming, that is, the neural network training of the sub-thread and the prediction of the main thread run in parallel, so as to realize the dynamic updating of the corresponding network parameters of the first neural network and the second neural network.
[0069] The training method of the power device operation parameter prediction model provided by the embodiment of the application controls the first neural network and the second neural network to run in parallel through the multi-thread technology, and adjusts the network parameters of the first neural network according to the prediction result output by the second neural network, thereby improving the prediction performance of the second neural network.
[0070] In some embodiments, after the plurality of operation parameters are preprocessed, the method further comprises: in the case that the number of the preprocessed plurality of operation parameters exceeds a first number threshold, converting the plurality of operation parameters into an event sequence according to a sliding time window algorithm, and performing online training on the first neural network according to the event sequence.
[0071] In this embodiment, the first number threshold can be set according to the actual demand of the user, for example, the first number threshold is 4.
[0072] In this embodiment, when the number of the obtained operation parameters reaches the first number threshold n, that is, {x t1 , x t2 , …x tn}, the operation parameters are converted into n-m (n
[0073] In this embodiment, for the time series prediction of the key parameters of the power plant, the steepest descent method (also known as the gradient descent method) is used to train the BP neural network. Since the BP neural network used in the time series prediction is trained online, the training termination condition of the BP neural network can be set as: the training time of the neural network reaches its preset value.
[0074] Specifically, the first neural network is trained online using the event sequence. When the training duration reaches the preset duration tms, the current round of training is completed, and the weight matrix output by the first neural network at the current time is transmitted to the second neural network. The second neural network updates the network parameters according to the weight matrix to obtain a real-time updated prediction model.
[0075] In this embodiment, after the plurality of operating parameters are converted into the event sequence according to the sliding time window algorithm, the method further includes: training the first neural network online based on the event sequence when the number of samples in the event sequence is greater than or equal to a preset training sample number; performing single-step prediction on the event sequence based on the second neural network when the number of samples in the event sequence is greater than a second number threshold; and displaying and storing the single-step prediction result. The second number threshold is greater than the first number threshold.
[0076] In this embodiment, the second number threshold can be set according to actual user needs. For example, the second number threshold is 10.
[0077] It should be noted that the future values of these parameters affecting the water level of the steam boiler of the ship power plant are unknown and need to be predicted. However, it can be considered that the influence of these factors is concentrated in the observation value time series of the steam boiler water level. Any time series can be regarded as an input-output system determined by a nonlinear mechanism, as shown in the following formula: (the mapping relationship of the formula is solved by using the prediction model to perform time series prediction)
[0078] y t+1 =f(y t ,y t-1 ,...,y t-s+1 );
[0079] where y t is the observation value at time t, and s is the length of historical data, also called the length of original data. Since the correlation between different time series is different, especially for complex time series, the prediction results of different historical data lengths are quite different, and therefore the selection of s value is relatively important. Generally, the s value can be selected by the trial-and-error method to select different historical data lengths for fitting and detection by the prediction model, and the length with relatively good fitting and detection effect is selected. The prediction value of y at time t+1 is represented by represent, while y t+1 Generally called the actual value or true value, represent, while y t Compared in time, it is called single-step prediction; if (y t+1 , y t , y t-1 , … y t-s+2 ) is taken as the input of the prediction model, then It is called two-step prediction, and by analogy, three-step, four-step or multi-step prediction can be realized.
[0080] In this embodiment, although the multi-step prediction increases the prediction time, its accuracy and stability are relatively poor, and the application only studies single-step, two-step and three-step prediction. Since the prediction model only finds the mapping relationship of a single variable time series, it is called a single variable time series problem. In actual application, sometimes multiple variables need to be predicted at the same time, which is called multi-variable time series prediction. Since all information in time series prediction comes from a single sequence, the back propagation direction can be used for memory training and prediction in application. That is, the BP network can be used for nonlinear time series prediction. The good nonlinear function fitting and approximation trend of the BP neural network can better reveal the correlation of the nonlinear time series of the power device in the time delay state space, so as to predict the key parameters of the power system.
[0081] The training method of the power device operation parameter prediction model provided by the embodiment of the application realizes dynamic updating of the corresponding network parameters of the first neural network and the second neural network, and further improves the accuracy of the power device parameter prediction, by converting the plurality of operation parameters into an event sequence according to a sliding time window algorithm when the number of the plurality of operation parameters after preprocessing exceeds a first number threshold, and performing online training on the first neural network according to the event sequence.
[0082] In some embodiments, after obtaining the power device operation parameter prediction model, the method further includes: inputting the to-be-tested operation parameter into the power device operation parameter prediction model for single-step prediction to obtain a plurality of prediction results; and respectively performing statistics on the plurality of prediction results according to a target performance index to obtain a target prediction result, the target prediction result being one of the plurality of prediction results; and the target performance index includes at least one of a mean square error (MSE) and an average value of absolute values of relative errors (AMPE).
[0083] In this embodiment, by statistically analyzing the mean square error (MSE) and the average value of absolute values of relative errors (AMPE) of the prediction results, and drawing the corresponding effect diagram, the user can observe the parameter change of the online updating of the two neural networks and the prediction performance of the overall model.
[0084] In some embodiments, the first neural network and the second neural network are both BP neural networks, and the number of hidden layer nodes, the number of input layer nodes and the number of output layer nodes of the BP neural network satisfy the following relationship:
[0085]
[0086] wherein, l is the number of hidden layer nodes, m is the number of input layer nodes, n is the number of output layer nodes, a is a constant, and the value of a is between [1, 10]; the learning rate of the BP neural network is obtained by trial and error method.
[0087] In this embodiment, before performing time series prediction, the parameters of the BP neural network need to be determined, including the number of input layer nodes, the number of hidden layer nodes, the learning rate, etc. The setting of each parameter is described in detail below.
[0088] In this embodiment, the number of output layer nodes of the BP neural network is 1, and the number of input layer nodes (length of historical data) and the number of hidden layer nodes are determined without a unified formula to follow. Generally, empirical formula and trial and error method are used. In the operation of this study, the following is used:
[0089] (1) First, the length of the historical time series is set in the range of [a, b], that is, the number of input layer nodes of the BP neural network is in the range of [a, b]; and because the number of output layer nodes is 1, the range of the number of hidden layer nodes is [c, d] according to the following formula:
[0090]
[0091] (2) A part of the data under the steady state of the power device is intercepted to form a plurality of training sample sets, and each training sample set contains n training samples.
[0092] (3) Keeping the other parameters unchanged, different numbers of input layer nodes (length of historical data) and different numbers of hidden layer nodes are used to perform single-step prediction by the BP neural network. When the training time reaches tms, the training is stopped and the value of the mean square error (MSE) is recorded;
[0093] (4) In Table 1 below, the number of input layer nodes and the number of hidden layer nodes corresponding to the minimum MSE are selected as the best setting values to improve the prediction performance of the network. Through comparative analysis, the best settings are determined: the number of input layer nodes and the number of hidden layer nodes.
[0094] Table 1 Training of BPNN with different learning rates
[0095]
[0096] The training method of the power device operation parameter prediction model provided by the embodiment of the present application comprises the following steps: setting the number of hidden layer nodes, the number of input layer nodes and the number of output layer nodes of the BP neural network to meet a certain relationship to construct a first neural network and a second neural network, and providing parameter support for network training and prediction.
[0097] Figure 2 is a flowchart of the training method of the power device operation parameter prediction model provided by the present application, and in the embodiment shown in Figure 2 In the embodiment shown in the figure, after obtaining a plurality of operation parameters (corresponding to data acquisition), it is first determined whether the number of operation parameters exceeds 4. If it exceeds, the plurality of operation parameters are converted into an event sequence through a sliding time window. On the one hand, the number of samples in the event sequence is accumulated. When the number of samples in the event sequence reaches the number of training samples, the first neural network is trained using the event sequence. In the case where the training time is greater than tms, the weight matrix of the first neural network is transmitted to the second neural network, and the first sample in the training sample set is deleted to update the training sample of the first neural network. On the other hand, when the number of samples in the event sequence is greater than n, the second neural network is used to predict the event sequence, and the prediction result is displayed and stored. According to the prediction result and the updated training sample, it is determined whether the iteration of the first neural network and the second neural network is stopped. In the case where the iteration is stopped, the trained second neural network is used for prediction and effect evaluation.
[0098] The training device of the power device operation parameter prediction model provided by the present application is described below. The training device of the power device operation parameter prediction model described below can be correspondingly referred to the training method of the power device operation parameter prediction model described above.
[0099] Figure 3 is a structural schematic diagram of the training device of the power device operation parameter prediction model provided by the present application, as Figure 3 The training device of the power device operation parameter prediction model comprises a parameter acquisition module 310, a first processing module 320 and a second processing module 330.
[0100] The parameter acquisition module 310 is used to acquire a plurality of operation parameters of the power device. Different operation parameters correspond to different time stamps, and the plurality of operation parameters are arranged in chronological order according to the time stamps.
[0101] The first processing module 320 is used to pre-process the plurality of operation parameters, and to perform online training on the first neural network according to the plurality of pre-processed operation parameters. After the online training time reaches the target length of time, the weight matrix of the output of the first neural network is sent to the second neural network. The second neural network is used to dynamically predict the operation parameters of the power device.
[0102] The second processing module 330 is configured to remove the running parameter corresponding to the shortest timestamp from the plurality of running parameters, add the running parameter corresponding to the target moment of the power device to obtain updated plurality of running parameters, dynamically update the weight matrix of the output of the first neural network according to the updated plurality of running parameters and the prediction step corresponding to the second neural network, and obtain the power device running parameter prediction model; the target moment is greater than the maximum value of the plurality of running parameters corresponding to the timestamps.
[0103] The training device of the power device running parameter prediction model provided by the embodiment of the application improves the prediction accuracy of the running data of the power device under the variable working condition by pre-processing the plurality of running parameters, online training the first neural network according to the plurality of pre-processed running parameters, sending the weight matrix of the output of the first neural network to the second neural network after the online training time reaches the target length, removing the running parameter corresponding to the shortest timestamp from the plurality of running parameters, adding the running parameter corresponding to the target moment of the power device to obtain the updated plurality of running parameters, and dynamically updating the weight matrix of the output of the first neural network according to the updated plurality of running parameters and the prediction step corresponding to the second neural network.
[0104] The application further provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the training method of any of the power device running parameter prediction models when executing the program.
[0105] Figure 4 is a structural schematic diagram of the electronic device provided by the application, as Figure 4As shown, the electronic device can include a processor 410, a communications interface 420, a memory 430, and a communications bus 440, wherein the processor 410, the communications interface 420, and the memory 430 complete mutual communication through the communications bus 440. The processor 410 can invoke a logic instruction in the memory 430 to execute a training method of a power device running parameter prediction model, the method comprising: obtaining a plurality of running parameters of the power device, different running parameters corresponding to different time stamps, the plurality of running parameters arranged in chronological order according to the time stamps; preprocessing the plurality of running parameters, and performing online training on a first neural network according to the preprocessed plurality of running parameters, sending a weight matrix of an output of the first neural network to a second neural network after an online training time reaches a target duration; the second neural network is used for dynamically predicting the running parameters of the power device; removing the running parameter corresponding to the shortest time stamp from the plurality of running parameters, and adding the running parameter corresponding to the target time of the power device, to obtain updated plurality of running parameters, dynamically updating the weight matrix of the output of the first neural network according to the updated plurality of running parameters and the prediction step corresponding to the second neural network, to obtain the power device running parameter prediction model; the target time is greater than the maximum value of the time stamps corresponding to the plurality of running parameters.
[0106] In addition, the logic instruction in the memory 430 described above can be implemented in the form of a software function unit and sold or used as an independent product, which can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.
[0107] In another aspect, the present application also provides a computer program product comprising a computer program, which can be stored on a non-transitory computer-readable storage medium, and which, when executed by a processor, enables a computer to perform the method for training a power plant operating parameter prediction model as described above, which comprises: obtaining a plurality of operating parameters of a power plant, different operating parameters corresponding to different time stamps, the plurality of operating parameters being arranged in chronological order according to the time stamps; pre-processing the plurality of operating parameters, and performing online training on a first neural network according to the pre-processed plurality of operating parameters, and sending a weight matrix of an output of the first neural network to a second neural network after the online training time reaches a target duration; the second neural network being configured to dynamically predict the operating parameters of the power plant; removing an operating parameter corresponding to the shortest time stamp from the plurality of operating parameters, and adding an operating parameter corresponding to a target time of the power plant, to obtain updated plurality of operating parameters, and dynamically updating the weight matrix of the output of the first neural network according to the updated plurality of operating parameters and a prediction step corresponding to the second neural network, to obtain the power plant operating parameter prediction model; the target time being greater than a maximum value of the time stamps corresponding to the plurality of operating parameters.
[0108] In yet another aspect, the present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for training a power plant operating parameter prediction model as described above, which comprises: obtaining a plurality of operating parameters of a power plant, different operating parameters corresponding to different time stamps, the plurality of operating parameters being arranged in chronological order according to the time stamps; pre-processing the plurality of operating parameters, and performing online training on a first neural network according to the pre-processed plurality of operating parameters, and sending a weight matrix of an output of the first neural network to a second neural network after the online training time reaches a target duration; the second neural network being configured to dynamically predict the operating parameters of the power plant; removing an operating parameter corresponding to the shortest time stamp from the plurality of operating parameters, and adding an operating parameter corresponding to a target time of the power plant, to obtain updated plurality of operating parameters, and dynamically updating the weight matrix of the output of the first neural network according to the updated plurality of operating parameters and a prediction step corresponding to the second neural network, to obtain the power plant operating parameter prediction model; the target time being greater than a maximum value of the time stamps corresponding to the plurality of operating parameters.
[0109] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.
[0110] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.
[0111] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A training method for a predictive model of operating parameters of a power unit, characterized in that, The method comprises the following steps: acquiring a plurality of operating parameters of a power device, different operating parameters corresponding to different timestamps, the plurality of operating parameters being arranged in chronological order according to the timestamps; preprocessing the plurality of operating parameters, and performing online training on a first neural network according to the plurality of preprocessed operating parameters, and sending a weight matrix of an output of the first neural network to a second neural network after the online training time reaches a target length of time; the second neural network is used for dynamically predicting the operating parameters of the power device; removing an operating parameter corresponding to the shortest timestamp from the plurality of operating parameters, and adding an operating parameter corresponding to a target time of the power device to obtain updated plurality of operating parameters, and dynamically updating the weight matrix of the output of the first neural network according to the updated plurality of operating parameters and a prediction step corresponding to the second neural network to obtain a power device operating parameter prediction model; the target time is greater than the maximum value of the timestamps corresponding to the plurality of operating parameters.
2. The method of claim 1, wherein, The preprocessing of the plurality of operating parameters comprises: the plurality of operating parameters are preprocessed by the following formula: wherein u(z) is the pre-processed operating parameter, z is the operating parameter, h and g are constant terms, the values of h and g are in the range of [-1, 1], and g < h, x max is the maximum value of the operating parameter, and x min is the minimum value of the operating parameter.
3. The method of claim 1, wherein, the first neural network and the second neural network adopt the same network structure; the method further comprises: controlling the first neural network and the second neural network to run side by side based on multi-threading technology, and adjusting network parameters of the first neural network according to a prediction result output by the second neural network.
4. The method of claim 1, wherein, After the preprocessing of the plurality of operating parameters, the method further comprises: in a case where the number of the plurality of preprocessed operating parameters exceeds a first number threshold, converting the plurality of operating parameters into an event sequence according to a sliding time window algorithm, and performing online training on the first neural network according to the event sequence.
5. The method of claim 4, wherein, After the plurality of operating parameters are converted into the event sequence according to the sliding time window algorithm, the method further comprises: in a case where the number of samples in the event sequence is greater than or equal to a preset training sample number, performing online training on the first neural network based on the event sequence; in a case where the number of samples in the event sequence is greater than a second number threshold, performing single-step prediction on the event sequence based on a second neural network, and displaying and storing the single-step prediction result; the second number threshold is greater than the first number threshold.
6. The method of claim 1, wherein, After obtaining the power device operating parameter prediction model, the method further comprises: inputting a to-be-tested operating parameter into the power device operating parameter prediction model to perform single-step prediction, and obtaining a plurality of prediction results; statistically processing the plurality of prediction results according to a target performance index to obtain a target prediction result, the target prediction result being one of the plurality of prediction results; the target performance index comprises at least one of mean square error (MSE) and average value of absolute value of relative error (AMPE).
7. The method of claim 1, wherein, The first neural network and the second neural network are both BP neural networks, and the number of nodes in the hidden layer, the number of nodes in the input layer and the number of nodes in the output layer of the BP neural network satisfy the following relationship: Wherein, l is the number of nodes of the hidden layer, m is the number of nodes of the input layer, n is the number of nodes of the output layer, a is a constant, and the value of a is between 1 and 10. The learning rate of the BP neural network is obtained by trial and error method.
8. A device for training a power plant operating parameter prediction model, characterized by Comprise: A parameter acquisition module is configured to acquire a plurality of operating parameters of a power device, different operating parameters corresponding to different time stamps, and the plurality of operating parameters being arranged in chronological order according to the time stamps. A first processing module is configured to pre-process the plurality of operating parameters, and perform online training on a first neural network according to the pre-processed plurality of operating parameters, and send a weight matrix of an output of the first neural network to a second neural network when a training time of the online training reaches a target length. The second neural network is configured to dynamically predict operating parameters of the power device. A second processing module is configured to exclude an operating parameter corresponding to a shortest time stamp from the plurality of operating parameters, and add an operating parameter corresponding to a target time of the power device to obtain updated plurality of operating parameters, and dynamically update the weight matrix of the output of the first neural network according to the updated plurality of operating parameters and a prediction step corresponding to the second neural network to obtain a power device operating parameter prediction model; the target time is greater than a maximum value of the time stamps corresponding to the plurality of operating parameters.
9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to realize the training method of the power device operating parameter prediction model according to any one of claims 1 to 7. 10.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the training method of the power device operating parameter prediction model according to any one of claims 1 to 7.
Citation Information
Patent Citations
Microgrid frequency adaptive learning control method
CN108448594A
Traffic illegal behavior identification method for applying artificial intelligence to intelligent traffic
CN114092878A