Data-driven digital twinning adaptive modeling method for integrated energy system

Abstract

The invention relates to a digital twin adaptive modeling method for a data-driven integrated energy system. The method is suitable for the field of digital twinning modeling of the integrated energy system. According to the technical scheme, the data-driven digital twin adaptive modeling method for the integrated energy system comprises the steps that a training data set is constructed based on actual operation data of the integrated energy system, and all samples in the training data set have input and output parameters of equipment related to the integrated energy system; dividing the training data set into a plurality of sub-data sets representing different operation conditions by adopting a k-means clustering algorithm; adopting a generative adversarial network to perform data enhancement on the sub-data sets under different operation conditions so as to enable the sample number difference between the sub-data sets under the operation conditions to be within a preset range; and training a digital twinborn model of the integrated energy system based on the data enhanced training data set.

Description

Technical Field

[0001] The present invention relates to a data-driven digital twin adaptive modeling method for integrated energy systems, which is applicable to the field of digital twin modeling of integrated energy systems. Background Art

[0002] Integrated energy systems can meet the energy consumption needs of users in a region for electricity, heat, cold, etc. through the coordinated optimization and efficient complementarity of various forms of energy flows, thereby improving the overall energy utilization efficiency. Its popularization and application are of great significance for improving the consumption capacity of renewable energy and can significantly enhance the economy and reliability of traditional energy systems.

[0003] Integrated energy systems have characteristics such as complex system forms, large scales, and multi-physical field coupling. Digital twin technology can map the dynamic change process in its entire life cycle by constructing a high-fidelity virtual mirror of the integrated energy system in the digital space, and is widely regarded as the technical foundation for realizing application requirements such as comprehensive perception, optimal design, collaborative operation, and reliability assessment of integrated energy systems, and is a research hotspot in this field.

[0004] Data-driven methods have advantages such as strong flexibility, wide applicability, strong learning ability, and independence from domain knowledge, and are widely used in the field of digital twin modeling of integrated energy systems. The accuracy of data-driven models depends severely on the quality of training data. On the one hand, actual training data usually has the problem of data imbalance, and operation data is scarce under some operating conditions, resulting in a decrease in model accuracy under such conditions; on the other hand, when facing new operating conditions not covered by the training data set, the accuracy is also difficult to meet the actual application requirements. How to improve the model accuracy of data-driven methods in the digital twin modeling of integrated energy systems still needs further research. Summary of the Invention

[0005] The technical problem to be solved by the present invention is: in view of the above problems, to provide a data-driven digital twin adaptive modeling method for integrated energy systems.

[0006] The technical solution adopted by the present invention is: a data-driven digital twin adaptive modeling method for integrated energy systems, including: Constructing a training data set based on the actual operation data of the integrated energy system, and each sample in the training data set has the input and output parameters of the equipment involved in the integrated energy system; Using the k-means clustering algorithm to divide the training data set into multiple sub-data sets representing different operating conditions; Using a generative adversarial network to perform data augmentation on the sub-data sets of different operating conditions so that the difference in the number of samples between the sub-data sets of each operating condition is within a preset range; Train a digital twin model of the integrated energy system based on the data-augmented training dataset.

[0007] It also includes: At every preset time, calculate the deviation between the digital twin model and the actual operation result; If the deviation is greater than the preset threshold, form an updated dataset based on the actual operation data collected during the period between the current deviation calculation and the previous deviation calculation; Based on the updated dataset, adaptively update the digital twin model using the elastic weight consolidation technique.

[0008] The step of using the k-means clustering algorithm to divide the training dataset into multiple sub-datasets representing different operation conditions includes: Based on the training dataset, generate an input parameter dataset, where each sample in the input parameter dataset has the input parameters of the equipment involved in the integrated energy system; Based on the number K of operation conditions corresponding to the actual operation data of the integrated energy system in the training dataset, use the k-means clustering algorithm to divide the input parameter dataset into K input parameter sub-datasets representing different operation conditions; Based on the input parameter sub-datasets, divide the training dataset into multiple sub-datasets representing different operation conditions.

[0009] The step of using the generative adversarial network to perform data augmentation on the sub-datasets of different operation conditions so that the difference in the number of samples between the sub-datasets of each operation condition is within the preset range includes: Take the samples in the sub-dataset as real samples; add random noise to the samples in the sub-dataset to generate fake samples; Based on the real samples and the fake samples, train the generator and the discriminator in the generative adversarial network; Use the trained generator to generate new samples containing the input and output parameters of the equipment involved in the integrated energy system, and supplement them to the corresponding sub-datasets until the difference in the number of samples between the sub-datasets is within the preset range; Based on the sub-datasets after data augmentation, generate a training dataset after data augmentation.

[0010] The step of training the digital twin model of the integrated energy system based on the data-augmented training dataset includes: Use an artificial neural network to train the digital twin model.

[0011] The step of adaptively updating the digital twin model using the elastic weight consolidation technique based on the updated dataset includes: The loss function of the update process As shown in the following formula: Where, represents the loss function adopted in the online update process represents the neural network parameters in the online update process represents the neural network parameters in the offline training model represents the trade-off factor represents the parameter importance weight corresponding in the Fisher information matrix, obtained by for the model parameter the second derivative of, as shown in the following formula: wherein, represents the online data set represents a single sample in the online data set. The change of the model parameter has a greater impact on the loss function , the greater the importance weight , thus restricting the change during update.

[0012] A data-driven digital twin adaptive modeling device for an integrated energy system, comprising: A data set construction module, configured to construct a training data set based on the actual operation data of the integrated energy system, and each sample in the training data set has the input and output parameters of the devices involved in the integrated energy system; A data set classification module, which uses the k-means clustering algorithm to divide the training data set into multiple sub-data sets representing different operating conditions; A data augmentation module, configured to use a generative adversarial network to augment the sub-data sets of different operating conditions, so that the difference in the number of samples between the sub-data sets of each operating condition is within a preset range; A model training module, configured to train a digital twin model of the integrated energy system based on the training data set after data augmentation.

[0013] It further includes: A deviation calculation module, configured to calculate the deviation between the digital twin model and the actual operation result every preset time; A data acquisition module, configured to, if the deviation is greater than a preset threshold, form an update data set based on the actual operation data collected during the time period between the current deviation calculation and the previous deviation calculation; A model update module, configured to adaptively update the digital twin model based on the update data set by using the elastic weight preservation technique.

[0014] A storage medium, on which a computer program executable by a processor is stored, and when the computer program is executed, the steps of the data-driven digital twin adaptive modeling method for an integrated energy system are implemented.

[0015] A digital twin adaptive modeling device has a memory and a processor. A computer program that can be executed by the processor is stored on the memory. When the computer program is executed, the steps of the data-driven digital twin adaptive modeling method for integrated energy systems are implemented.

[0016] The beneficial effects of the present invention are as follows: The present invention uses a clustering algorithm and a data augmentation strategy to automatically identify and augment data for operating conditions with insufficient sample sizes, which can effectively improve the overall data quality of the model training set and ensure that the digital twin model still has high accuracy under operating conditions with insufficient sample sizes.

[0017] The present invention regularly calculates the deviation between the digital twin model and the actual operating results, and uses an elastic weight consolidation technique to adaptively learn and online update the model for new operating conditions encountered during real-time operation, which can effectively ensure the accuracy of the digital twin model during the full life cycle operation of the integrated energy system. Description of the Drawings

[0018] Figure 1 It is a schematic diagram of the integrated energy system in the embodiment of the present invention.

[0019] Figure 2 It is a flowchart of the digital twin modeling method for the integrated energy system in the embodiment of the present invention.

[0020] Figure 3 It is a comparison of the accuracy results of the offline training of the digital twin model in the embodiment of the present invention.

[0021] Figure 4 It is a comparison of the accuracy results of the online application of the digital twin model in the embodiment of the present invention. Detailed Embodiments

[0022] To better understand the technical solutions of the present application, the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0023] It should be clear that the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts belong to the scope of protection of the present application.

[0024] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "the" and "said" used in the embodiments of the present application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0025] Embodiment 1: As Figure 1As shown in the figure, the power generation side equipment of the integrated energy system in this embodiment includes a gas turbine unit, a photovoltaic power generation unit, and a storage battery, and external electric energy can be input through the superior power grid at the same time; the heating side equipment includes a gas boiler and an electric heat pump. Through the coordinated complementarity of energy between the power generation side equipment and the heating side equipment, electric and heat loads are jointly provided to the end.

[0026] As Figure 2 shown in the figure, this embodiment is a data-driven digital twin adaptive modeling method for an integrated energy system, which specifically includes the following steps: S100. Construct a training data set based on the actual operation data of the integrated energy system. Each sample in this training data set has the input and output parameters of the equipment involved in this integrated energy system.

[0027] In this embodiment, the input parameter set and output parameter set of the digital twin model are determined according to the actual form of the integrated energy system, and an offline training data set of the digital twin model is constructed based on the actual operation data.

[0028] In this example, the types of equipment involved in the integrated energy system include a gas turbine unit, a photovoltaic power generation unit, a storage battery, a gas boiler, and an electric heat pump. The input parameters and output parameters of various equipment are shown in the following table:

[0029] S200. Use the k-means clustering algorithm to divide the training data set into multiple sub-data sets representing different operating conditions. The operating conditions are composed of the input parameter sets of each integrated energy equipment. The distance between the input parameter vectors of the samples in the same sub-data set (operating condition) is closer than that of other sub-data sets. The purpose of this step is to screen out multiple main operating conditions from the training data set through the clustering algorithm, providing a basis for subsequent data augmentation for operating conditions with a small number of samples.

[0030] S210. Generate an input parameter data set based on the training data set. Each sample in this input parameter data set has the input parameters of the equipment involved in this integrated energy system.

[0031] Preprocess the input parameter set of each sample in the training data set, remove the outliers and missing values in the data, and perform data normalization to obtain the preprocessed input parameter data set : Among them, is the total number of samples. In this embodiment, the training data set is the data of a certain year of the system, the sampling frequency is 1h, and there are a total of 8760. Each sample is shown in the following formula:

[0032] In this embodiment, the number of input parameters of the gas turbine unit, photovoltaic power generation unit, battery, gas boiler, and heat pump model are 3, 3, 2, 3, and 3 respectively. Therefore, the sample dimension is .

[0033] S220. Based on the number of operating conditions K corresponding to the actual operating data of the integrated energy system in the training dataset, the k-means clustering algorithm is used to divide the input parameter dataset into K input parameter subsets representing different operating conditions.

[0034] S221. In this example, set the maximum number of iterations of the algorithm , and the number of cluster centers (number of operating conditions) ; S222. Randomly select K samples from the input parameter dataset as the initial cluster centers ; S223. For each sample in the input parameter dataset , calculate its distance to each cluster center and assign it to the category corresponding to the cluster center with the minimum distance ; S224. For each category , update its cluster center , as shown in the following formula: .

[0035] S225. If each cluster center no longer changes or the number of algorithm iterations reaches the upper limit M , the algorithm stops. At this time, the data in each category respectively corresponds to a subset of data; otherwise, repeat S223 to S225.

[0036] S230. Based on the input parameter subsets representing different operating conditions, divide the training dataset into multiple subsets representing different operating conditions. Among the finally obtained five subsets of data, the data representing operating conditions one to five respectively, and the number of samples are 2463, 2884, 485, 2580, and 348 in sequence.

[0037] S300. Use a generative adversarial network to perform data augmentation on the subsets of data for different operating conditions, so that the difference in the number of samples between the subsets of data for each operating condition is within a preset range. In this embodiment, the number of samples for operating conditions three and five is significantly less than that of the other three conditions. Therefore, a generative adversarial network is used to supplement the data until the number reaches the same as that of the other three conditions.

[0038] S310. Initialize the network parameters of the generator and discriminator using a Gaussian distribution and , set the learning rates of the generator and discriminator networks and , the number of training epochs , and the batch size .

[0039] S320. Select the samples of operating condition three and the WUZI dataset as real samples respectively , and add random Gaussian noise to the generator on this basis to generate a series of fake samples , as shown in the following formula:

[0040] S330. Train the generator and discriminator in the generative adversarial network based on the real samples and fake samples.

[0041] The goal of the discriminator is to maximize the discrimination ability for real samples, making the output of real samples close to 1 and the output of fake samples close to 0. Based on the input real samples and fake samples, the discriminator uses the cross-entropy loss function to train the network parameters, as shown in the following formula: where is the sample and

[0042] is the probability of the real sample.

[0043] In each round of training, first update the network parameters through the discriminator , and then update the network parameters through the generator , and alternately train until the discrimination accuracy of the discriminator for real samples and fake samples approaches 0.5 or reaches the preset number of training epochs, then the iteration stops and proceeds to step S340.

[0044] S340. Use the trained generator to add random noise to each sample of the dataset to be enhanced, and input it into the generator to generate new sample data containing the input and output parameters of the devices involved in the integrated energy system, and supplement it to the corresponding sub-datasets until the difference in the number of samples between the sub-datasets is within the preset range.

[0045] S400. Based on the training dataset enhanced by step S300, use an artificial neural network to train the digital twin model of the actual integrated energy system. The specific steps include data preprocessing, hyperparameter optimization, model training, and performance evaluation. The specific steps are as follows: S410. For the missing values in the training dataset, use the data of the two time periods before and after the missing data for mean imputation; S420. Use z-score normalization to normalize the training set data as shown in the following formula: where, and are the mean and standard deviation of this type of input respectively.

[0046] S430. Set the number of hidden layers , the number of neurons in each layer , the learning rate and the batch size ; S440. In each iteration, randomly sample from the training set according to the batch size , and use the Adam optimization algorithm to train the model to update the parameters. The loss function uses the mean squared error MSE as shown in the following formula: where, and are the true value and the predicted value output by the model respectively. If the root mean square error is less than the preset value, stop the iteration; otherwise, repeat the training until the iteration number reaches the upper limit.

[0047] S500. According to the deviation between the model output and the actual operation result, use the elastic weight consolidation technique to adaptively update the model.

[0048] S510. Every time , calculate the deviation between the digital twin model and the actual operation result of the system.

[0049] S520. Judge whether the digital twin model needs to be updated. If the deviation is greater than the preset threshold, return to step S510; otherwise, enter step S530.

[0050] S530. Based on the current deviation, form an update dataset for online model update from the actual operation data collected during the time period between the current deviation calculation and the previous deviation calculation (within the T time period before the current deviation calculation time).

[0051] S540. Based on the update dataset, use the elastic weight consolidation technique to adaptively update the digital twin model. The loss function in the update process is as shown in the following formula: where, represents the loss function used in the online update process. In this embodiment, the mean squared error MSE is used as the loss function. Represent the neural network parameters for the online update process Represent the neural network parameters in the offline training model Represent the trade - off factor Represent the corresponding parameter importance weights in the Fisher information matrix, obtained by For the model parameters The second - order derivative, as shown in the following formula: Wherein, Represent the online dataset Represent a single sample in the online dataset. The change of the model parameter has a greater impact on the loss function , the greater the importance weight , thus restricting the change of during the update.

[0052] Iteratively execute forward propagation, calculate the loss, backpropagation, and update the parameters until the algorithm converges to achieve the online update of the model. Then, return to step S510.

[0053] To verify the modeling accuracy of the method in this embodiment, a traditional data - driven modeling method is set for comparative verification. The traditional data - driven modeling method does not perform operating condition identification and data augmentation during the offline training process, and the digital twin model does not perform online updates during the real - time application process. In the verification session, an electro - thermal pump model is selected as a typical example for result comparison.

[0054] The accuracy results of model offline training using different methods are as Figure 3 shown. It can be seen from the figure that although the overall accuracy of the traditional data - driven modeling method on the entire dataset is not much different from that of the method in this embodiment, the accuracy in operating conditions three and five is significantly lower than that of the method in this embodiment. The main reason is that the data volume of these two types of operating conditions is relatively limited (485 and 348 respectively), and the traditional method lacks corresponding data samples during training, resulting in a decrease in the model accuracy under data - scarce operating conditions. In contrast, the method in this embodiment complements the data of the operating conditions with insufficient data volume through operating condition identification and data augmentation, significantly improving the accuracy of the digital twin model under various operating conditions.

[0055] The comparison of the accuracy changes of model online application using different methods is as Figure 4As shown in the figure. In the process of online application, the traditional data-driven modeling method always keeps the model unchanged. Therefore, when facing operating conditions not covered in the offline training set, the model accuracy is insufficient, and the model accuracy will gradually decrease with the change of operating time. In contrast, the method of this embodiment continuously conducts adaptive learning and parameter update on the model through the online update technology, and the model accuracy can always be maintained at a high level to meet the actual application requirements.

[0056] This embodiment can ensure the accuracy of the digital twin model of the integrated energy system during the whole life cycle operation through data augmentation and online learning, and can provide a reliable model foundation for subsequent application requirements such as optimal design, operation scheduling, and reliability analysis, which has great practical value for promoting the popularization of digital twin technology in the integrated energy field.

[0057] Embodiment 2: This embodiment is a data-driven digital twin adaptive modeling device for an integrated energy system, including: a data set construction module, a data set classification module, a data augmentation module, a model training module, and an adaptive update module, wherein the adaptive update module includes a deviation calculation module, a data acquisition module, and a model update module.

[0058] In this example, the data set construction module is used to construct a training data set based on the actual operation data of the integrated energy system, and each sample in the training data set has the input and output parameters of the equipment involved in the integrated energy system; the data set classification module uses the k-means clustering algorithm to divide the training data set into multiple sub-data sets representing different operating conditions; the data augmentation module is used to perform data augmentation on the sub-data sets of different operating conditions by using a generative adversarial network so that the difference in the number of samples between the sub-data sets of each operating condition is within a preset range; the model training module is used to train the digital twin model of the integrated energy system based on the training data set after data augmentation.

[0059] In this embodiment, the deviation calculation module is used to calculate the deviation between the digital twin model and the actual operation result every preset time; the data acquisition module is used to form an update data set based on the actual operation data collected during the time period between the current deviation calculation and the previous deviation calculation if the deviation is greater than a preset threshold; the model update module is used to perform adaptive update on the digital twin model based on the update data set by using the elastic weight consolidation technique.

[0060] Embodiment 3: This embodiment is a storage medium, on which a computer program executable by a processor is stored, and when the computer program is executed, the steps of the data-driven digital twin adaptive modeling method for an integrated energy system in Embodiment 1 are implemented.

[0061] Embodiment 4: This embodiment is a digital twin adaptive modeling device, which has a memory and a processor. A computer program capable of being executed by the processor is stored on the memory. When the computer program is executed, the steps of the data-driven digital twin adaptive modeling method for an integrated energy system in Embodiment 1 are implemented.

Claims

1. A data-driven integrated energy system digital twin adaptive modeling method, characterized in that: include: Constructing a training data set based on actual operation data of the integrated energy system, wherein each sample in the training data set has input and output parameters of the equipment involved in the integrated energy system; The k-means clustering algorithm is used to divide the training data set into multiple sub-data sets representing different operating conditions; Generative adversarial networks are used to enhance the sub-datasets of different operating conditions so that the difference in the number of samples between the sub-datasets of each operating condition is within a preset range; Train the digital twin model of the integrated energy system based on the data-enhanced training dataset.

2. The data-driven integrated energy system digital twin adaptive modeling method according to claim 1 is characterized in that: Also includes: At preset intervals, the deviation between the digital twin model and the actual operation results is calculated; If the deviation is greater than a preset threshold, an updated data set is formed based on the actual operating data collected during the time period between the current deviation calculation and the previous deviation calculation; Based on the updated data set, the elastic weight preservation technology is used to adaptively update the digital twin model.

3. The data-driven integrated energy system digital twin adaptive modeling method according to claim 1 is characterized in that: The k-means clustering algorithm is used to divide the training data set into multiple sub-data sets representing different operating conditions, including: Based on the training data set, an input parameter data set is generated, wherein each sample in the input parameter data set has input parameters of the equipment involved in the integrated energy system; Based on the number of operating conditions K corresponding to the actual operating data of the integrated energy system in the training data set, the k-means clustering algorithm is used to divide the input parameter data set into K input parameter sub-data sets representing different operating conditions; Based on the input parameter sub-dataset, the training data set is divided into multiple sub-datasets representing different operating conditions.

4. The data-driven integrated energy system digital twin adaptive modeling method according to claim 1 is characterized in that: The method of using a generative adversarial network to perform data enhancement on sub-datasets of different operating conditions so that the difference in the number of samples between the sub-datasets of each operating condition is within a preset range includes: Take the samples in the sub-dataset as real samples; add random noise to the samples in the sub-dataset to generate fake samples; Train the generator and discriminator in the generative adversarial network based on real samples and fake samples; Generate new samples containing input and output parameters of the equipment involved in the integrated energy system using the trained generator, and add them to the corresponding sub-datasets until the difference in the number of samples between the sub-datasets is within a preset range; Based on the data-enhanced sub-dataset, a data-enhanced training dataset is generated.

5. The data-driven integrated energy system digital twin adaptive modeling method according to claim 1 is characterized in that: The method of training a digital twin model of a comprehensive energy system based on a data-enhanced training data set includes: Artificial neural network is used to train the digital twin model.

6. The data-driven integrated energy system digital twin adaptive modeling method according to claim 2 is characterized in that: The method of adaptively updating the digital twin model based on the updated data set and using the elastic weight retention technology includes: Loss function of the update process As shown below: in, represents the loss function used in the online update process, represents the neural network parameters of the online update process, represents the neural network parameters in the offline training model, represents the trade-off factor, Represents the corresponding parameter importance weight in the Fisher information matrix, by calculating the loss function For model parameters The second-order derivative of is obtained as shown below: in, represents an online dataset, Represents a single sample in the online dataset.

7. A data-driven integrated energy system digital twin adaptive modeling device, characterized in that: include: A data set construction module is used to construct a training data set based on the actual operation data of the integrated energy system, wherein each sample in the training data set has input and output parameters of the equipment involved in the integrated energy system; The data set classification module uses the k-means clustering algorithm to divide the training data set into multiple sub-data sets representing different operating conditions; A data enhancement module is used to enhance the sub-datasets of different operating conditions by using a generative adversarial network, so that the difference in the number of samples between the sub-datasets of each operating condition is within a preset range; The model training module is used to train the digital twin model of the integrated energy system based on the data-enhanced training data set.

8. The data-driven integrated energy system digital twin adaptive modeling device according to claim 7, characterized in that: Also includes: The deviation calculation module is used to calculate the deviation between the digital twin model and the actual operation result at preset time intervals; A data collection module, for forming an updated data set based on actual operating data collected during the time period between the current deviation calculation and the previous deviation calculation if the deviation is greater than a preset threshold; The model update module is used to adaptively update the digital twin model based on the updated data set using elastic weight retention technology.

9. A storage medium having stored thereon a computer program executable by a processor, characterized in that: When the computer program is executed, the steps of the data-driven integrated energy system digital twin adaptive modeling method according to any one of claims 1 to 6 are implemented.

10. A digital twin adaptive modeling device, comprising a memory and a processor, wherein the memory stores a computer program executable by the processor, characterized in that: When the computer program is executed, the steps of the data-driven integrated energy system digital twin adaptive modeling method according to any one of claims 1 to 6 are implemented.

Images