Optimization method and device of multi-feedforward prediction model, equipment and storage medium
By optimizing the multi-feedforward prediction model using a time-varying loss function, the problem of insufficient accuracy and reliability of existing power equipment prediction models is solved, and more accurate power equipment status prediction is achieved.
Patent Information
- Application Number
- CN202411052963.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-01
- Publication Date
- 2024-11-19
Smart Images

Figure CN118982279A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of power equipment indicator prediction, and specifically to an optimization method, device, equipment and storage medium for a multi-feedforward prediction model. Background Art
[0002] Power equipment state prediction is a vital task in the power system, which can effectively improve the operating efficiency and reliability of the power system. Traditional prediction methods are mostly focused on single-step prediction, but with the increasing complexity of the power system, multi-step feedforward prediction (Multi-stepAheadForecasting) has gradually become a research hotspot. Multi-step feedforward prediction aims to directly predict the data values of multiple future time points by inputting historical data at one time, rather than gradually rolling predictions for each time point. The research on this method is of great value. First, it can more comprehensively reflect the dynamic changes of the future state of the system and avoid the problem of cumulative errors in rolling predictions. Secondly, multi-step feedforward prediction is superior in computational efficiency because it reduces the repeated calculations of each step of the prediction and improves the overall operating efficiency of the model.
[0003] At present, multi-step feedforward forecasting is generally achieved through multi-feedforward forecasting models, but the loss function of multi-feedforward forecasting models mostly characterizes the error, while ignoring the impact of the volatility of input historical data on the output results, which leads to the deviation of the forecast results from the actual situation and the loss of the ability to accurately capture the dynamic characteristics of the power system. In this case, the model may over-rely on the current input data and ignore the change pattern of the data and the complex time series structure, thus affecting the accuracy and reliability of the forecast.
[0004] Therefore, how to improve the accuracy and reliability of the multi-feedforward prediction model in predicting future data is a technical problem that technicians in this field urgently need to solve. Summary of the invention
[0005] Based on the above problems, the present application provides an optimization method, device, equipment and storage medium for a multi-feedforward prediction model, which can improve the accuracy and reliability of the multi-feedforward prediction model when predicting future data.
[0006] The embodiments of the present application disclose the following technical solutions:
[0007] A method for optimizing a multi-feedforward prediction model, the method comprising:
[0008] Acquire a test data set of a target power device; the test data set includes a plurality of test samples with operation index labels; one of the test samples includes a plurality of historical operation indexes of the target power device at the same time point; the operation index label of the test sample is the actual operation index corresponding to the test sample;
[0009] Based on the test data set, a prediction data set of the target power equipment is obtained through a pre-built preliminary multi-feedforward prediction model; the test data set and the prediction data set are both time series; the time length of the prediction data set is less than or equal to the time length of the test operation indicator;
[0010] Calculating the error between the predicted data set and the running indicator label of the test sample to obtain a prediction error;
[0011] Calculating the loss function value between the predicted data set and the running indicator label of the test sample using a time-varying loss function;
[0012] The preliminary multi-feedforward prediction model is optimized based on the loss function value and the prediction error.
[0013] In a possible implementation, the time-varying loss function is:
[0014] L=α·L Huber +β·L Volatility +γ·L Reg ;
[0015] Wherein, L is the time-varying loss function; L Huber is the Huber loss function; L Volatility is the volatility loss function; L Reg is the regularization function; α, β, γ are the weights of each function;
[0016] y is the actual operation index; is the prediction data set; δ is the hyperparameter of Huber loss;
[0017] Any time point t for the test run indicator i The prediction dataset; for The corresponding actual operation index; t is the time length of the predicted data set;
[0018] m is the number of samples used when constructing the preliminary multi-feedforward prediction model; At any time point t k Samples of For sample y tk W is the weight parameter connecting different layers in the preliminary multi-feedforward prediction model; is the square of the Euclidean norm of W; λ is the hyperparameter of the regularization term.
[0019] In a possible implementation manner, the optimizing the preliminary multi-feedforward prediction model based on the loss function value and the prediction error includes:
[0020] Determine adjustment parameters and adjustment directions and adjustment ranges of the adjustment parameters according to the current prediction error and loss function value; the adjustment parameters include one or more of various hyperparameters and weights of various functions;
[0021] Based on the adjustment parameters and the adjustment direction and adjustment range of the adjustment parameters, the parameters of the preliminary multi-feedforward prediction model are updated using an optimization algorithm such as gradient descent.
[0022] In a possible implementation, the method further includes:
[0023] Acquire a verification data set, and obtain verification operation indicators on the verification data set through the updated preliminary multi-feedback prediction model; the verification data set includes multiple verification samples with operation indicator labels; one verification sample includes multiple historical operation indicators of the target power equipment at the same time point; the operation indicator label of the verification sample is the actual operation indicator corresponding to the verification sample;
[0024] Calculating the error between the verification running indicator and the running indicator label of the verification data set to obtain a verification error;
[0025] If the verification error is less than or equal to the error threshold, the updated parameters of the preliminary multi-feedforward prediction model are solidified to obtain the target multi-feedforward prediction model.
[0026] In a possible implementation, the construction process of the pre-built preliminary multi-feedforward prediction model includes:
[0027] Acquire a training data set; the training data set includes a plurality of training samples with operation index labels; one of the training samples includes a plurality of historical operation indexes of the target power equipment at the same time point; the operation index label of the training sample is the actual operation index corresponding to the training sample;
[0028] The training data set is input into a neural network structure for forward propagation to obtain the preliminary multi-feedforward prediction model.
[0029] An optimization device for a multi-feedforward prediction model, the device comprising:
[0030] A first acquisition unit is used to acquire a test data set of a target power device; the test data set includes a plurality of test samples with operation index labels; one of the test samples includes a plurality of historical operation indexes of the target power device at the same time point; the operation index label of the test sample is an actual operation index corresponding to the test sample;
[0031] A second acquisition unit is used to obtain a prediction data set of the target power equipment through a pre-built preliminary multi-feedforward prediction model based on the test data set; the test data set and the prediction data set are both time series; the time length of the prediction data set is less than or equal to the time length of the test operation indicator;
[0032] A first calculation unit, used to calculate the error between the predicted data set and the running indicator label of the test sample to obtain a prediction error;
[0033] A second calculation unit, used to calculate the loss function value between the prediction data set and the running indicator label of the test sample by using a time-varying loss function;
[0034] An optimization unit is used to optimize the preliminary multi-feedforward prediction model based on the loss function value and the prediction error.
[0035] In a possible implementation, the optimization unit specifically includes:
[0036] A determination unit, used to determine an adjustment parameter and an adjustment direction and an adjustment range of the adjustment parameter according to a current prediction error and a loss function value; the adjustment parameter includes one or more of various hyperparameters and weights of various functions;
[0037] A parameter updating unit is used to update the parameters of the preliminary multi-feedforward prediction model based on the adjustment parameters and the adjustment direction and adjustment range of the adjustment parameters using an optimization algorithm such as gradient descent.
[0038] In a possible implementation manner, the device further includes:
[0039] A third acquisition unit is used to acquire a verification data set, and obtain a verification operation index on the verification data set through the updated preliminary multi-feedforward prediction model; the verification data set includes a plurality of verification samples with operation index labels; one of the verification samples includes a plurality of historical operation indexes of the target power equipment at the same time point;
[0040] A third calculation unit, used to calculate the error between the verification operation index and the operation index label of the verification data set to obtain a verification error;
[0041] A parameter solidification unit is used to solidify the updated parameters of the preliminary multi-feedforward prediction model to obtain a target multi-feedforward prediction model if the verification error is less than or equal to an error threshold.
[0042] An optimization device for a multi-feedforward prediction model comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the optimization method for the multi-feedforward prediction model as described above is implemented.
[0043] A computer-readable storage medium stores instructions, which, when executed on a terminal device, enable the terminal device to execute the optimization method for the multi-feedforward prediction model as described above.
[0044] Compared with the prior art, this application has the following beneficial effects:
[0045] The present application provides an optimization method, device, equipment and storage medium for a multi-feedforward prediction model. Specifically, when executing the optimization method for the multi-feedforward prediction model provided in the embodiment of the present application, a test data set of the target power equipment is first obtained, which includes multiple test samples with operating indicator labels. Each test sample contains multiple historical operating indicators of the equipment at the same time point, and the operating indicator label of each test sample corresponds to its actual operating indicator. These data are used to generate a prediction data set for the target power equipment based on a pre-built preliminary multi-feedforward prediction model, and these data sets are all time series data. The time length of the prediction data set does not exceed the time length of the test operating indicator data. Then, the error between the predicted data set and the operating indicator label of the test sample is calculated to obtain the prediction error. The loss function value between the predicted data set and the operating indicator label of the test sample is evaluated using a time-varying loss function, and the preliminary multi-feedback prediction model is optimized by this value and the prediction error. The operating indicators of power equipment are often affected by a variety of dynamic factors, such as load changes, seasonal changes, emergencies, etc. By adopting a time-varying loss function, the prediction model can better capture the changing pattern behind these data. The time-varying loss function not only takes into account the size of the prediction error, but also adjusts its sensitivity to the error according to time and context, thereby more accurately measuring the model's predictive ability at different time points and situations. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] In order to more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0047] Figure 1 A schematic diagram of an exemplary application scenario provided for an embodiment of the present application;
[0048] Figure 2 A method flow chart of an optimization method for a multi-feedforward prediction model provided in an embodiment of the present application;
[0049] Figure 3 A schematic diagram of the structure of an optimization device for a multi-feedforward prediction model provided in an embodiment of the present application. DETAILED DESCRIPTION
[0050] To facilitate understanding of the technical solutions provided by the embodiments of the present application, the background technology involved in the embodiments of the present application will be described below.
[0051] Predicting the state of power equipment is critical in power systems because it can significantly improve the operating efficiency and reliability of power systems. Traditional prediction methods can usually only predict data at a single point in the future step by step. However, with the increasing complexity of power systems, researchers have begun to pay attention to the method of multi-step feedforward prediction. Multi-step feedforward prediction predicts the data values of multiple future time points at one time by inputting historical data, rather than predicting each time point step by step. Therefore, it can more comprehensively reflect the dynamic changes of the future state of the system and avoid the problem of cumulative errors in step-by-step prediction. In addition, multi-step feedforward prediction is also better in computational efficiency because it reduces repeated calculations and improves overall operating efficiency.
[0052] At present, multi-step feedforward forecasting is mainly achieved through multi-feedforward forecasting models. However, most of the existing models focus on the characterization of errors by the loss function, but ignore the impact of the volatility of input historical data on the output results. This leads to the lack of robustness and accuracy of multi-step feedforward forecasting models when processing complex power system data, and may show problems such as poor generalization ability, susceptibility to data noise interference, model overfitting or underfitting, and unstable forecasting results. These factors ultimately reduce the reliability and effectiveness of the forecast, making it difficult for the model to accurately capture the dynamic characteristics of the power system.
[0053] In order to solve this problem, an optimization method, device, equipment and storage medium for a multi-feedforward prediction model are provided in an embodiment of the present application. First, a test data set of the target power equipment is obtained. Based on these test data, a pre-built preliminary multi-step feedforward prediction model is used to generate a prediction data set for the target power equipment, ensuring that the prediction data set and the test data set are both time series and the time length of the prediction data set does not exceed the test data set. Next, the error between the prediction data set and the operating indicator label of the test sample, that is, the prediction error, is calculated, and the loss function value between the prediction data set and the test sample is calculated using a time-varying loss function. Finally, based on the loss function value and the prediction error, the preliminary multi-step feedforward prediction model is optimized. This application considers the volatility of multiple historical operating indicators in the power equipment test data set, combines the time-varying loss function to calculate and optimize the prediction error, and effectively improves the model's learning ability for the volatility characteristics in the power system data, thereby enhancing the model's accurate prediction and robustness of the actual operating conditions.
[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0055] See also Figure 1 , which is a flow chart of a method for optimizing a multi-feedforward prediction model provided in an embodiment of the present application, such as Figure 1 As shown, the optimization method of the multi-feedforward prediction model may include steps S101-S105:
[0056] S101: Acquire a test data set of a target power device.
[0057] A test data set is obtained from the target power equipment. This data set contains multiple test samples, each of which has a corresponding operating indicator label. Each test sample represents multiple historical operating indicator data recorded by the target power equipment at the same time point. The operating indicator label of each test sample corresponds to the actual operating indicator data recorded by the sample. In other words, this data set provides time series observation data, each time point contains multiple related operating indicator values, which are used to analyze and predict the behavior and performance of power equipment.
[0058] It should be noted that the test dataset is mutually exclusive with the training dataset used to train the preliminary multi-feedforward prediction model, i.e., the model has not been exposed to it during the training process.
[0059] S102: Based on the test data set, a prediction data set of the target power equipment is obtained through a pre-built preliminary multi-feedforward prediction model.
[0060] A pre-built multi-layer feedforward neural network (i.e., a multi-feedforward prediction model) is used to make predictions based on the provided test data set. Both the test data set and the prediction data set are time series data, and each time point contains multiple operating indicators of the target power equipment. The time span of the prediction data set is less than or equal to the time span recorded in the test data set, that is, the data set generated by the prediction model covers the prediction results within the time period already in the test data set. This means that the prediction model predicts the future operating status and indicator changes of power equipment by learning from historical data.
[0061] In one possible implementation, the construction process of the pre-built preliminary multi-feedforward prediction model includes: first obtaining a training data set, which includes multiple training samples. Each training sample has a corresponding operating indicator label, which represents the actual operating indicator data corresponding to the sample. In each training sample, multiple historical operating indicator data of the target power equipment at the same time point are recorded. Next, using these training data sets, a multi-layer feedforward neural network is constructed. The training data set is input into the neural network structure for forward propagation, and a preliminary multi-layer feedforward prediction model can be obtained through back-propagation optimization during the training process. The purpose of this model is to use historical operating indicator data to learn the operating mode of the power equipment, so as to predict the operating status and indicator changes of the target power equipment in the future.
[0062] This can be achieved by following the steps below:
[0063] Data preparation: Collect and organize historical operating indicator data of power equipment as a training data set. The training data set contains multiple training samples with operating indicator labels. One training sample includes multiple historical operating indicators of the target power equipment at the same time point. The operating indicator label of the training sample is the actual operating indicator corresponding to the training sample.
[0064] Data preprocessing: Standardize and normalize the samples in the training data set so that they can be used in the model.
[0065] Model selection: Choose a suitable neural network structure as the multi-step feedforward indicator prediction model, such as Multilayer Perceptron (MLP), etc. You can also consider using Recurrent Neural Network (RNN) or Long Short-Term Memory (LSTM) to process time series data.
[0066] Model training: Input the processed training data set into the selected neural network structure for forward propagation to obtain a preliminary multi-feedforward prediction model.
[0067] S103: Calculate the error between the predicted data set and the running indicator label of the test sample to obtain the prediction error.
[0068] Compare the predicted dataset with the running metric labels of the test samples. By comparing the difference between the predicted values and the actual values, we can calculate an error metric, called prediction error, which is used to evaluate the accuracy and reliability of the prediction model.
[0069] Optionally, the prediction error may be, but is not limited to, a root mean square error (RMSE) or a mean absolute error (MAE).
[0070] S104: Calculate the loss function value between the predicted data set and the running indicator label of the test sample using a time-varying loss function.
[0071] The time-varying loss function is used to calculate the loss function value between the predicted value in the prediction data set and the actual operating indicator label in the test sample. The time-varying loss function is a loss function that is adjusted according to time changes and can better consider the importance and weight of different time points in the time series data. By calculating the loss function value, the difference between the predicted data and the actual data can be evaluated, and it helps to optimize the prediction model to improve the accuracy and prediction ability of the performance of power equipment.
[0072] In a possible implementation, the time-varying loss function is:
[0073] L=α·L Huber +β·L Volatility +γ·L Reg ;
[0074] Wherein, L is the time-varying loss function; L Huber is the Huber loss function; L Volatility is the volatility loss function; L Reg is the regularization function; α, β, γ are the weights of each function;
[0075] y is the actual operation index; is the prediction data set; δ is the hyperparameter of Huber loss;
[0076] Any time point t for the test run indicator i The prediction dataset; for The corresponding actual operation index; t is the time length of the predicted data set;
[0077] m is the number of samples used when constructing the preliminary multi-feedforward prediction model; At any time point t k Samples of For sample W is the weight parameter connecting different layers in the preliminary multi-feedforward prediction model; is the square of the Euclidean norm of W; λ is the hyperparameter of the regularization term.
[0078] It should be noted that the parameter values of each parameter in the time-varying loss function are set according to actual needs, that is, the user can define the parameter values of each parameter according to actual needs.
[0079] S105: Optimizing the preliminary multi-feedforward prediction model based on the loss function value and the prediction error.
[0080] By analyzing the loss function value and prediction error, we can identify the deficiencies of the model in predicting the performance of power equipment, and adopt corresponding optimization strategies to improve the performance and accuracy of the model. By optimizing the model, we can improve its ability to predict the future operating status and performance of power equipment, thereby better guiding equipment maintenance, management and decision-making.
[0081] In a possible implementation, step S105 optimizes the preliminary multi-feedback prediction model based on the loss function value and the prediction error, including:
[0082] First, according to the current prediction error and loss function value, the adjustment parameter and the adjustment direction and adjustment range of the adjustment parameter are determined. Then, based on the adjustment parameter and the adjustment direction and adjustment range of the adjustment parameter, an optimization algorithm such as gradient descent is used to update the parameters of the preliminary multi-feedforward prediction model.
[0083] This can be achieved by following the steps below:
[0084] Evaluate the performance of the neural network: First, evaluate the performance of the preliminary multi-feedforward prediction model based on the current prediction error and loss function value. The loss function value is an indicator that measures the difference between the model prediction and the actual result, and the prediction error is one of the indicators that specifically measures this difference.
[0085] Determine the network parameters that need to be adjusted: Based on the evaluation results, determine the network parameters that need to be adjusted, including the weights and biases in the neural network, and possible hyperparameters (such as learning rate, regularization parameters, etc.). The goal of adjusting these parameters is to improve the predictive performance of the model by reducing the loss function value.
[0086] Adjustment direction and range: Determine the direction and range in which each parameter can be adjusted, that is, determine the step size or amplitude by which each parameter can be increased or decreased during the update process. These adjustments are made to effectively move in the direction of reducing the loss function value.
[0087] Update parameters using optimization algorithms: Select an appropriate optimization algorithm, such as gradient descent, to update the parameters of the neural network model according to the gradient of the loss function. The gradient descent algorithm calculates the gradient of the loss function with respect to the model parameters and updates the parameters in the opposite direction of the gradient to minimize the loss function value.
[0088] Apply calculated adjustment values: Apply the calculated parameter adjustment values to the structural parameters of the neural network and update the entire model.
[0089] Repeat the training process: The updated model is used again for a new training dataset, and the above steps are repeated until the model's performance reaches a predetermined standard or a specified training cycle has passed. This process is iterative, and the model is gradually optimized through multiple adjustments and training, so that it can better generalize to new datasets.
[0090] See also Figure 2 , Figure 2 A method for determining a multi-feedforward prediction model provided in an embodiment of the present application can be implemented through steps A1-A3:
[0091] A1: Obtain a verification data set, and obtain verification operation indicators on the verification data set through the updated preliminary multi-feedback prediction model.
[0092] After the model training is completed, the prediction accuracy of the model is usually tested on an independent validation dataset to ensure the generalization ability of the model on unseen data. The validation dataset contains multiple validation samples, each of which includes multiple historical operating indicators of the target power equipment at the same time point. Each validation sample has a value called an operating indicator label, which corresponds to the actual operating indicator data of the validation sample. By applying the updated prediction model, predictions can be made for each validation sample, and the prediction results can be compared with the actual operating indicator labels to evaluate the performance of the model on the validation dataset. This process is designed to verify the predictive ability and accuracy of the model to ensure that it can effectively predict the operating status and indicators of the target power equipment.
[0093] It should be noted that the validation dataset and the training dataset are mutually exclusive, that is, the model has not been exposed to them during the training process.
[0094] A2: Calculate the error between the verification running indicator and the running indicator label of the verification data set to obtain a verification error.
[0095] Calculating the error between the validation operating indicator and the operating indicator label of each sample in the validation data set is a key step in evaluating the accuracy of the updated prediction model. By comparing the predicted value of each validation sample with the actual operating indicator label, the validation error can be calculated, which reflects the prediction accuracy and deviation of the model on the validation data set. This process helps determine whether the model can accurately predict the operating status of the target power equipment and provides an important reference for further optimization and improvement of the model.
[0096] Optionally, the validation error may be, but is not limited to, a root mean square error (RMSE) or a mean absolute error (MAE).
[0097] A3: If the verification error is less than or equal to the error threshold, the updated parameters of the preliminary multi-feedforward prediction model are solidified to obtain the target multi-feedforward prediction model.
[0098] If the verification error is less than or equal to the preset error threshold, the parameters of the updated preliminary multi-feedforward prediction model will be solidified (i.e., confirmed or determined), thereby forming the target multi-feedforward prediction model. This means that after verification, the prediction ability of the model has reached the expected level and can be safely applied to actual data for prediction and analysis, providing a reliable foundation for subsequent work.
[0099] Based on the contents of S101-S105, it can be known that the test data set is first collected. Subsequently, using these data, the prediction data set of the target power equipment is obtained through the pre-established preliminary multi-feedforward prediction model. Further, after calculating the error between the prediction data set and the operating index label of the test sample, the prediction error is obtained and the time-varying loss function is applied to calculate the loss function value of the preliminary multi-feedforward prediction model. Finally, the preliminary multi-feedforward prediction model is optimized based on the loss function value and the prediction error to improve its prediction accuracy and generalization ability for the dynamic evolution of the power equipment state. This method comprehensively considers the correlation between data and time series characteristics, and achieves more accurate model adjustment by optimizing the loss function, thereby effectively improving the performance of the prediction model in the power system data prediction task. The time-varying loss function of the present application can adjust the calculation method of the loss function according to the volatility of the actual data. The operating indicators of the power equipment may be affected by seasonal changes, load fluctuations or other external factors, which will cause the volatility of the data. The traditional static loss function is difficult to effectively handle such fluctuations, while the time-varying loss function can adjust the loss function according to the real-time data, so that the model can more accurately reflect the prediction error under the current operating state.
[0100] See also Figure 3 , Figure 3The schematic diagram of the structure of a multi-feedforward prediction model optimization device provided in the embodiment of the present application. Figure 3 As shown, the optimization device of the multi-feedforward prediction model includes:
[0101] The first acquisition unit 301 is used to acquire a test data set of a target power device; the test data set includes multiple test samples with operation index labels; one of the test samples includes multiple historical operation indexes of the target power device at the same time point; the operation index label of the test sample is the actual operation index corresponding to the test sample;
[0102] A second acquisition unit 302 is used to obtain a prediction data set of the target power equipment based on the test data set through a pre-built preliminary multi-feedforward prediction model; the test data set and the prediction data set are both time series; the time length of the prediction data set is less than or equal to the time length of the test operation indicator;
[0103] A first calculation unit 303 is used to calculate the error between the predicted data set and the running indicator label of the test sample to obtain a prediction error;
[0104] A second calculation unit 304 is used to calculate the loss function value between the prediction data set and the running indicator label of the test sample by using a time-varying loss function;
[0105] The optimization unit 305 is used to optimize the preliminary multi-feedforward prediction model based on the loss function value and the prediction error.
[0106] In a possible implementation, the time-varying loss function is:
[0107] L=α·L Huber +β·L Volatility +γ·L Reg ;
[0108] Wherein, L is the time-varying loss function; L Huber is the Huber loss function; L Volatility is the volatility loss function; L Reg is the regularization function; α, β, γ are the weights of each function;
[0109] y is the actual operation index; is the prediction data set; δ is the hyperparameter of Huber loss;
[0110] Any time point t for the test run indicator i The prediction dataset; for The corresponding actual operation index; t is the time length of the predicted data set;
[0111] m is the number of samples used when constructing the preliminary multi-feedforward prediction model; At any time point t k Samples of For sample W is the weight parameter connecting different layers in the preliminary multi-feedforward prediction model; is the square of the Euclidean norm of W; λ is the hyperparameter of the regularization term.
[0112] In a possible implementation, the optimization unit 305 specifically includes:
[0113] A determination unit, used to determine an adjustment parameter and an adjustment direction and an adjustment range of the adjustment parameter according to a current prediction error and a loss function value; the adjustment parameter includes one or more of various hyperparameters and weights of various functions;
[0114] A parameter updating unit is used to update the parameters of the preliminary multi-feedforward prediction model based on the adjustment parameters and the adjustment direction and adjustment range of the adjustment parameters using an optimization algorithm such as gradient descent.
[0115] In a possible implementation manner, the device further includes:
[0116] A third acquisition unit is used to acquire a verification data set, and obtain a verification operation index on the verification data set through the updated preliminary multi-feedforward prediction model; the verification data set includes a plurality of verification samples with operation index labels; one of the verification samples includes a plurality of historical operation indexes of the target power equipment at the same time point;
[0117] A third calculation unit, used to calculate the error between the verification operation index and the operation index label of the verification data set to obtain a verification error;
[0118] A parameter solidification unit is used to solidify the updated parameters of the preliminary multi-feedforward prediction model to obtain a target multi-feedforward prediction model if the verification error is less than or equal to an error threshold.
[0119] In a possible implementation manner, the device further includes:
[0120] A fourth acquisition unit is used to acquire a training data set; the training data set includes a plurality of training samples with operation index labels; one of the training samples includes a plurality of historical operation indexes of the target power equipment at the same time point; the operation index label of the training sample is the actual operation index corresponding to the training sample;
[0121] An input unit is used to input the training data set into a neural network structure for forward propagation to obtain the preliminary multi-feedforward prediction model.
[0122] In addition, an embodiment of the present application also provides an optimization device for a multi-feedforward prediction model, including: a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the optimization method for the multi-feedforward prediction model as described above is implemented.
[0123] In addition, an embodiment of the present application also provides a computer-readable storage medium, in which instructions are stored. When the instructions are executed on a terminal device, the terminal device executes the optimization method of the multi-feedforward prediction model as described above.
[0124] The embodiment of the present application provides an optimization device for a multi-feedforward prediction model, first using a first acquisition unit 301 to acquire a test data set of a target power device, and using a second acquisition unit 302 to obtain a prediction data set of the target power device through a pre-built preliminary multi-feedforward prediction model based on the test data set. The first calculation unit 303 calculates the error between the prediction data set and the operating index label of the test sample to obtain the prediction error. Then, based on the second calculation unit 304, the loss function value between the prediction data set and the operating index label of the test sample is calculated using a time-varying loss function, so that the optimization unit 305 optimizes the preliminary multi-feedforward prediction model based on the loss function value and the prediction error. The time-varying loss function applied in the present application can dynamically adjust the prediction accuracy of the prediction model for the operating index of the target power device to adapt to the volatility and change trend of the data. By evaluating the error between the prediction data set and the actual operating index in real time, and combining the time series characteristics of the historical operating index, the time-varying loss function effectively optimizes the parameters of the multi-feedforward prediction model, improves the robustness and prediction accuracy of the model in different time periods, and thus enhances its practicality and reliability in a complex power equipment environment.
[0125] The above is a detailed introduction to the optimization method, device, equipment and storage medium of a multi-feedforward prediction model provided by the present application. The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the scope of protection of the claims of the present application.
[0126] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.
[0127] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.
Claims
1. A method for optimizing a multi-feedforward prediction model, characterized in that: The method comprises: Acquire a test data set of a target power device; the test data set includes a plurality of test samples with operation index labels; one of the test samples includes a plurality of historical operation indexes of the target power device at the same time point; the operation index label of the test sample is the actual operation index corresponding to the test sample; Based on the test data set, a prediction data set of the target power equipment is obtained through a pre-built preliminary multi-feedforward prediction model; the test data set and the prediction data set are both time series; the time length of the prediction data set is less than or equal to the time length of the test operation indicator; Calculating the error between the predicted data set and the running indicator label of the test sample to obtain a prediction error; Calculating the loss function value between the predicted data set and the running indicator label of the test sample using a time-varying loss function; The preliminary multi-feedforward prediction model is optimized based on the loss function value and the prediction error.
2. The method according to claim 1, characterized in that The time-varying loss function is: L=α·L Huber +β·L Volatility +γ·L Reg ; Wherein, L is the time-varying loss function; L Huber is the Huber loss function; L Volatility is the volatility loss function; L Reg is the regularization function; α, β, γ are the weights of each function; y is the actual operation index; is the prediction data set; δ is the hyperparameter of Huber loss; Any time point t for the test run indicator i The prediction dataset; for The corresponding actual operation index; t is the time length of the predicted data set; m is the number of samples used when constructing the preliminary multi-feedforward prediction model; At any time point t k Samples of For sample W is the weight parameter connecting different layers in the preliminary multi-feedforward prediction model; is the square of the Euclidean norm of W; λ is the hyperparameter of the regularization term.
3. The method according to claim 2, characterized in that The optimizing the preliminary multi-feedforward prediction model based on the loss function value and the prediction error comprises: Determine adjustment parameters and adjustment directions and adjustment ranges of the adjustment parameters according to the current prediction error and loss function value; the adjustment parameters include one or more of various hyperparameters and weights of various functions; Based on the adjustment parameters and the adjustment direction and adjustment range of the adjustment parameters, the parameters of the preliminary multi-feedforward prediction model are updated using an optimization algorithm such as gradient descent.
4. The method according to claim 3, characterized in that The method further comprises: Acquire a verification data set, and obtain verification operation indicators on the verification data set through the updated preliminary multi-feedback prediction model; the verification data set includes multiple verification samples with operation indicator labels; one verification sample includes multiple historical operation indicators of the target power equipment at the same time point; the operation indicator label of the verification sample is the actual operation indicator corresponding to the verification sample; Calculating the error between the verification running indicator and the running indicator label of the verification data set to obtain a verification error; If the verification error is less than or equal to the error threshold, the updated parameters of the preliminary multi-feedforward prediction model are solidified to obtain the target multi-feedforward prediction model.
5. The method according to claim 1, characterized in that The construction process of the pre-built preliminary multi-feedforward prediction model includes: Acquire a training data set; the training data set includes a plurality of training samples with operation index labels; one of the training samples includes a plurality of historical operation indexes of the target power equipment at the same time point; the operation index label of the training sample is the actual operation index corresponding to the training sample; The training data set is input into a neural network structure for forward propagation to obtain the preliminary multi-feedforward prediction model.
6. An optimization device for a multi-feedforward prediction model, characterized in that: The device comprises: A first acquisition unit is used to acquire a test data set of a target power device; the test data set includes a plurality of test samples with operation index labels; one of the test samples includes a plurality of historical operation indexes of the target power device at the same time point; the operation index label of the test sample is an actual operation index corresponding to the test sample; A second acquisition unit is used to obtain a prediction data set of the target power equipment through a pre-built preliminary multi-feedforward prediction model based on the test data set; the test data set and the prediction data set are both time series; the time length of the prediction data set is less than or equal to the time length of the test operation indicator; A first calculation unit, used to calculate the error between the predicted data set and the running indicator label of the test sample to obtain a prediction error; A second calculation unit, used to calculate the loss function value between the prediction data set and the running indicator label of the test sample by using a time-varying loss function; An optimization unit is used to optimize the preliminary multi-feedforward prediction model based on the loss function value and the prediction error.
7. The device according to claim 6, characterized in that The optimization unit specifically includes: A determination unit, used to determine an adjustment parameter and an adjustment direction and an adjustment range of the adjustment parameter according to a current prediction error and a loss function value; the adjustment parameter includes one or more of various hyperparameters and weights of various functions; A parameter updating unit is used to update the parameters of the preliminary multi-feedforward prediction model based on the adjustment parameters and the adjustment direction and adjustment range of the adjustment parameters using an optimization algorithm such as gradient descent.
8. The device according to claim 7, characterized in that The device also includes: A third acquisition unit is used to acquire a verification data set, and obtain a verification operation index on the verification data set through the updated preliminary multi-feedforward prediction model; the verification data set includes a plurality of verification samples with operation index labels; one of the verification samples includes a plurality of historical operation indexes of the target power equipment at the same time point; A third calculation unit, used to calculate the error between the verification operation index and the operation index label of the verification data set to obtain a verification error; A parameter solidification unit is used to solidify the updated parameters of the preliminary multi-feedforward prediction model to obtain a target multi-feedforward prediction model if the verification error is less than or equal to an error threshold.
9. An optimization device for a multi-feedforward prediction model, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the optimization method for the multi-feedforward prediction model as described in any one of claims 1 to 5 is implemented.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed on a terminal device, the terminal device executes the optimization method for the multi-feedforward prediction model as described in any one of claims 1 to 5.
