Cross-building energy consumption collaborative prediction method and device based on improved KAN network

Abstract

The invention is suitable for the field of building energy consumption, and provides a cross-building energy consumption collaborative prediction method and device based on an improved KAN network, and the method comprises the steps: obtaining the current building information of a to-be-predicted building and a target energy consumption prediction model, and enabling the target energy consumption prediction model to be obtained according to the knowledge sharing of the historical building information of a plurality of original buildings, an improved KAN network model is trained, and an activation function of the improved KAN network model is a Mexico cap wavelet function; and inputting the current building information into the target energy consumption prediction model to obtain target energy consumption of the to-be-predicted building output by the target energy consumption prediction model. According to the method, while cross-building knowledge sharing is realized, the expression ability of the target energy consumption prediction model in the aspect of capturing the nonlinear relationship and the multi-scale characteristics in the building energy consumption data is enhanced, so that the prediction precision of the building energy consumption is improved.

Description

Technical Field

[0001] This application belongs to the technical field of building energy consumption, and particularly relates to a cross-building energy consumption collaborative prediction method and device based on an improved KAN network. Background Art

[0002] With the rapid development of Internet of Things technology and intelligent building management systems, a large amount of building operation data has been continuously collected, including multi-dimensional information such as energy consumption data, user behavior patterns, and indoor and outdoor environmental parameters. These rich data resources provide a solid foundation for building energy consumption prediction. Among them, data-driven deep learning methods have shown good application prospects in the field of building energy consumption prediction.

[0003] However, the building energy consumption prediction methods in the prior art mainly model based on single-building data, and are often limited by objective factors such as sensor deployment time and equipment failures, resulting in insufficient data volume or incomplete data coverage conditions. At the same time, the deep learning models used in the prior art are mainly traditional neural network structures, and have limited expression ability in capturing complex non-linear relationships and multi-scale features of building energy consumption data, seriously reducing the prediction accuracy of building energy consumption. Summary of the Invention

[0004] The embodiments of this application provide a cross-building energy consumption collaborative prediction method and device based on an improved KAN network, which can enhance the expression ability of the model in capturing complex non-linear relationships and multi-scale features in building energy consumption data while realizing cross-building knowledge sharing, thereby improving the prediction accuracy of building energy consumption.

[0005] In the first aspect, the embodiments of this application provide a cross-building energy consumption collaborative prediction method based on an improved KAN network, including: Obtain the current building information of the building to be predicted and the target energy consumption prediction model, where the target energy consumption prediction model is determined by training an improved Kolmogorov-Arnold KAN network model based on the knowledge sharing of the historical building information of multiple original buildings, and the activation function of the improved KAN network model is the Mexican hat wavelet function; input the current building information into the target energy consumption prediction model to obtain the target energy consumption of the building to be predicted output by the target energy consumption prediction model.

[0006] In the second aspect, the embodiments of this application provide a cross-building energy consumption collaborative prediction device based on an improved KAN network, including: An obtaining module, configured to obtain the current building information of the building to be predicted and the target energy consumption prediction model, where the target energy consumption prediction model is determined by training an improved Kolmogorov-Arnold KAN network model based on the knowledge sharing of the historical building information of multiple original buildings, and the activation function of the improved KAN network model is the Mexican hat wavelet function; An input module for inputting current building information into a target energy consumption prediction model to obtain the target energy consumption of a building to be predicted output by the target energy consumption prediction model.

[0007] In a third aspect, an embodiment of the present application provides a terminal device, including 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 method according to any one of the first aspects is implemented.

[0008] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the method according to any one of the first aspects is implemented.

[0009] In a fifth aspect, an embodiment of the present application provides a computer program product, and when the computer program product runs on a terminal device, the terminal device is enabled to execute the method according to any one of the first aspects above.

[0010] An embodiment of the present application provides a cross-building energy consumption collaborative prediction method and device based on an improved KAN network. The method includes: obtaining the current building information of a building to be predicted and a target energy consumption prediction model, where the target energy consumption prediction model is determined by training an improved Kolmogorov-Arnold KAN network model based on the knowledge sharing of the historical building information of multiple original buildings, and the activation function of the improved KAN network model is a Mexican hat wavelet function; inputting the current building information into the target energy consumption prediction model to obtain the target energy consumption of the building to be predicted output by the target energy consumption prediction model. By using the above technical solution, the target energy consumption prediction model is determined by training an improved KAN network model according to the building information across buildings, and the activation function of the improved KAN network model is a Mexican hat wavelet function, which can enhance the expression ability of the target energy consumption prediction model in capturing the non-linear relationship and multi-scale features in building energy consumption data while realizing cross-building knowledge sharing, thereby improving the prediction accuracy of building energy consumption. Description of the Drawings

[0011] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained without creative efforts based on these drawings.

[0012] Figure 1 It is a flowchart of a cross-building energy consumption collaborative prediction method based on an improved KAN network provided by an embodiment of the present application; Figure 2It is a schematic structural diagram of an improved KAN network model provided by an embodiment of the present application; Figure 3 It is a schematic flowchart of a cross-building energy consumption collaborative prediction method based on an improved KAN network provided by another embodiment of the present application; Figure 4 It is a schematic flowchart of a clustering analysis provided by an embodiment of the present application; Figure 5 It is a schematic flowchart of a cross-building energy consumption collaborative prediction method based on an improved KAN network provided by another embodiment of the present application; Figure 6 It is a schematic flowchart of a federated learning provided by an embodiment of the present application; Figure 7 It is a schematic architecture diagram of a cross-building energy consumption collaborative prediction method based on an improved KAN network provided by another embodiment of the present application; Figure 8 It is a structural block diagram of a cross-building energy consumption collaborative prediction device based on an improved KAN network provided by an embodiment of the present application; Figure 9 It is a schematic structural diagram of a terminal device provided by an embodiment of the present application. Detailed implementation manners

[0013] In the following description, for the purpose of illustration rather than limitation, specific details such as specific system architectures, technologies, etc. are presented to thoroughly understand the embodiments of the present application. However, those skilled in the art should clearly understand that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from hindering the description of the present application.

[0014] It should be understood that when used in the specification and appended claims of the present application, the term "comprising" indicates the presence of the described features, wholes, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0015] It should also be understood that the term " / and" as used in the specification and appended claims of the present application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0016] As used in the specification and the appended claims of this application, the term "if" may be construed as "when" or "once" or "in response to determining" or "in response to detecting" depending on the context. Similarly, the phrases "if determined" or "if [the described condition or event] is detected" may be construed as meaning "once determined" or "in response to determining" or "once [the described condition or event] is detected" or "in response to detecting [the described condition or event]" depending on the context.

[0017] In addition, in the description of the specification and the appended claims of this application, the terms "first", "second", "third", etc. are only used for differentiating descriptions and cannot be construed as indicating or implying relative importance.

[0018] The reference to "one embodiment" or "some embodiments" or the like described in the specification of this application means that a specific feature, structure, or characteristic described in connection with that embodiment is included in one or more embodiments of this application. Thus, statements such as "in one embodiment", "in some embodiments", "in other some embodiments", "in still other embodiments", etc. that appear in different places in this specification do not necessarily all refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "comprising", "including", "having", and their variants all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0019] It should be noted that the information collection process (such as the face image collection process, the building information collection process, etc.) / feature extraction process involved in this application is executed with the user's knowledge and permission, that is, the information collection process / feature extraction process complies with the requirements of laws and regulations and does not belong to acts that harm the public interest.

[0020] It can be considered that building energy consumption prediction, as a core component of the building energy conservation management and control system, is of great strategic significance for realizing optimal energy utilization, reducing operating costs, and reducing carbon emissions.

[0021] The building energy consumption prediction methods in the prior art mainly perform modeling based on single-building data, and this method has many important limitations. First, the data collection of individual buildings is often restricted by objective factors such as sensor deployment time and equipment failures, resulting in insufficient data volume or incomplete data coverage of working conditions, making it difficult to support high-precision model training. Second, the model trained only using single-building data is difficult to handle the situation of newly built buildings or changes in operating modes, and its generalization ability is severely restricted. In addition, this method fails to fully utilize the common knowledge that may exist between different buildings, resulting in a waste of valuable data resources and a relatively low overall data utilization efficiency.

[0022] To address the above problems, two new cross-building machine learning paradigms, namely transfer learning and federated learning, have emerged. However, both methods still have deficiencies at present. On the one hand, there are certain limitations in the source domain building screening mechanism of existing transfer learning. Currently, there is a lack of a quantitative standard for source domain selection, and it overly relies on subjective experience screening, making it difficult to ensure the optimal similarity of source domain data. The existing standard federated learning framework, on the other hand, assumes homogeneous data distribution among all participating parties. However, in actual scenarios, significant differences in building functions, scales, and climate conditions lead to highly heterogeneous data distributions.

[0023] On the other hand, the deep learning models used in existing building energy consumption prediction tasks are mainly traditional neural network structures and have limitations in many aspects. For example, first, these deep learning models have limited expressive ability in capturing complex non-linear relationships and multi-scale features in building energy consumption data. Second, in the case of limited data volume, the performance of traditional deep learning models drops significantly, making it difficult to effectively learn. Subsequently, traditional black-box models lack interpretability and it is difficult to understand the basis for model decisions, which is an obvious defect in energy management decision support. Finally, the performance of traditional deep learning models is unstable when migrating across buildings, making it difficult to flexibly adapt to the data characteristics and distribution laws of different types of buildings.

[0024] In summary, the existing building energy consumption prediction methods for cross-building machine learning are not perfect, and the deep learning model architecture has certain limitations. Based on this, in view of the key technical bottlenecks such as data island problems, imperfect cross-domain machine learning paradigms, and model architecture limitations in the existing technology, the embodiments of this application provide a cross-building energy consumption collaborative prediction method based on an improved KAN (Kolmogorov Arnold Network) network model. By introducing the improved KAN model as the basic model architecture and integrating two cross-domain machine learning paradigms of transfer learning and federated learning, a complete cross-building energy consumption prediction technology system considering building similarity is constructed, improving the accuracy of energy consumption prediction for individual buildings.

[0025] Figure 1 FIG. is a schematic flowchart of a cross-building energy consumption collaborative prediction method based on an improved KAN network provided by an embodiment of this application. By way of example and not limitation, this method can be applied to a terminal device. The embodiments of this application do not impose any restrictions on the specific type of the terminal device. For example, the terminal device can be a terminal computer, or a server, etc. As Figure 1 shown, the method includes: S101. Obtain the current building information of the building to be predicted and the target energy consumption prediction model.

[0026] Among them, the target energy consumption prediction model is determined by training and improving the Kolmogorov-Arnold KAN network model based on the knowledge sharing of historical building information of multiple original buildings. The activation function of the improved KAN network model is the Mexican hat wavelet function.

[0027] The building to be predicted can refer to the building for which energy consumption prediction is to be carried out. The current building information can be considered as the building information currently related to the building to be predicted. The type of building information is not limited. For example, it can involve relevant data such as indoor and outdoor environmental parameters and time variables.

[0028] Optionally, the current building information at least includes current time data, current environmental data, and building area data. Among them, the current time data, for example, includes the current month, the current day type (i.e., Monday to Sunday), and the current hour number (i.e., 0 to 23); the current environmental data can, for example, include data such as dry bulb temperature, wet bulb temperature, and wind speed; the building area data is used to represent the total building area of the building to be predicted.

[0029] The target energy consumption prediction model can be understood as the energy consumption prediction model corresponding to the building to be predicted and can be used to predict the energy consumption of the building to be predicted. The target energy consumption prediction model can be considered to be determined by training and improving the Kolmogorov-Arnold KAN network model based on the knowledge sharing of historical building information of multiple original buildings. Among them, the multiple original buildings can include the building to be predicted, or other buildings different from the building to be predicted. Or the building to be predicted may not be one of the multiple original buildings, as long as there are similar buildings with a similar category relationship to the building to be predicted among the multiple original buildings. This embodiment does not make a limitation in this regard. The historical building information is the building information related to the history of the original building. For example, it can at least include historical time data, historical environmental data, building area data, and historical energy consumption data.

[0030] Furthermore, in this embodiment, an improved KAN network model is used for prediction. Among them, the KAN network model can be considered as a new neural network architecture based on the Kolmogorov-Arnold representation theorem, with a learnable activation function and high-efficient function approximation ability. However, due to the different activation functions in the network structure of the KAN network model itself, it is difficult to be directly used in transfer learning and federated learning. In this embodiment, the Mexican hat wavelet function is innovatively selected as the benchmark activation function of the KAN network model, and the improved network model is called the Improved-KAN model (abbreviated as IKAN). On this basis, the ability of the network model to capture multi-level features of data can be improved, the model structure can be simplified, and the prediction accuracy can be improved.

[0031] Figure 2 is a schematic structural diagram of an improved KAN network model provided by an embodiment of the present application, asFigure 2 As shown, the improved KAN network model can be composed of multiple activation functions. The benchmark for each activation function can be the Mexican hat wavelet function, and the parameters of the Mexican hat wavelet functions corresponding to different activation functions are different. The Mexican hat wavelet function can be used as an activation function to control the learnable hyperparameters of the function height and width. At the same time, the Mexican hat wavelet function enables the IKAN model to analyze data at different scales, thereby effectively capturing and representing the multi-level features of the data. The simple expression of the IKAN model can be the weighted sum of multiple activation functions, such as , are multiple learnable non-linear activation functions (i.e., Mexican hat wavelet functions). Among them, the symbol " " can be understood as the weighted sum of multiple learnable non-linear activation functions, that is can be further expressed as , are respectively 's weights. Each non-linear activation function can be represented by , where p is the number of input variables and q is the number of output variables. Here, multiple learnable non-linear activation functions can be represented as multiple , and each Among them, is an independent variable, representing time or position, which determines the evaluation point of the wavelet on the time axis; is the scale parameter, which controls the width of the wavelet. For example, a larger makes the wavelet wider and is suitable for capturing low-frequency components; a smaller makes the wavelet narrower and is suitable for high-frequency analysis; w is a learnable hyperparameter used to control the height and width of the Mexican hat wavelet function.

[0032] This embodiment does not limit the specific means of obtaining the target energy consumption prediction model. For example, the terminal device can pre-train the target energy consumption prediction model and directly obtain it from within the terminal device every time energy consumption prediction is required, or it can be obtained by the terminal device through interaction and communication with other devices. Or, this embodiment can also comprehensively consider various factors, select different cross-building machine learning paradigms, and realize the real-time construction of the target energy consumption prediction model of the building to be predicted based on the knowledge sharing of the historical building information of multiple original buildings. On the one hand, selecting a cross-building machine learning paradigm can focus on key factors such as privacy constraints, data heterogeneity, and communication costs of the target scenario. On the other hand, selecting a cross-building machine learning paradigm can combine the characteristics of different cross-domain machine learning paradigms. For example, the target energy consumption prediction model of the target domain is finally trained by transfer learning, while in federated learning, each participating building trains a target energy consumption prediction model respectively. Based on the above factors, a reasonable cross-domain machine learning paradigm can be selected to construct the target energy consumption prediction model of the building to be predicted.

[0033] Optionally, the cross-domain machine learning paradigm can include transfer learning and federated learning. Among them, transfer learning aims to transfer the knowledge learned from the source domain to promote the performance of the data-driven model in the target domain, and its model parameter adjustment focuses on the target domain. The transfer learning method is applicable to the situation where there is limited data in the target domain but a large amount of relevant source domain data. The main implementation steps of transfer learning are as follows: First, it is necessary to select source domain data related to the target domain; then, train a model on the source domain data or directly obtain a pre-trained model, which has learned the feature representations and patterns in the source domain; then decide how to adjust the pre-trained model according to the characteristics of the target domain; then perform knowledge transfer, apply the adjusted model to the target domain, and use the knowledge learned from the source domain to help the model in the target domain learn; finally, perform performance evaluation and optimization to evaluate the performance of the model on the target domain data. Compared with the traditional machine learning paradigm, transfer learning can make full use of cross-domain data resources to build a prediction model, significantly reducing the data requirements and computational costs required to develop a data-driven model for a new task.

[0034] Federated learning can be a cross - domain distributed learning method that allows multiple participants to collaborate in training a model while maintaining data privacy. This method does not require centralized data storage. Instead, it optimizes the model through iterative cycles of local client training and server parameter aggregation, enabling joint model training while protecting privacy. Federated learning can be described as a three - step learning process: (1) Initialize a global model on the central server and distribute it to all participating clients; (2) The clients perform model training based on local data; (3) Upload the local model information to the central server for global model update through privacy - preserving methods, such as using the average of gradients or model parameters. This process ensures data privacy through privacy - preserving protocols and achieves model convergence through multiple rounds of parameter interaction. Compared with traditional deep learning, federated learning has significant advantages in solving the data silo problem, reducing the risk of privacy leakage, and improving cross - domain collaboration efficiency, especially suitable for multi - participant collaborative modeling scenarios such as building energy consumption prediction.

[0035] S102. Input the current building information into the target energy consumption prediction model to obtain the target energy consumption of the building to be predicted output by the target energy consumption prediction model.

[0036] A cross - building energy consumption collaborative prediction method based on an improved KAN network provided in this embodiment is determined by training an improved KAN network model according to the building information of cross - buildings through the target energy consumption prediction model. The activation function of the improved KAN network model is the Mexican hat wavelet function, which can enhance the expression ability of the target energy consumption prediction model in capturing non - linear relationships and multi - scale features in building energy consumption data while realizing cross - building knowledge sharing, thereby improving the prediction accuracy of building energy consumption.

[0037] Figure 3 It is a schematic flowchart of a cross - building energy consumption collaborative prediction method based on an improved KAN network provided in another embodiment of the present application. In this embodiment, the target energy consumption prediction model for obtaining the building to be predicted is further optimized as follows: Obtain the historical building information of multiple original buildings, and the multiple original buildings at least include the building to be predicted; Screen the source - domain building information corresponding to the building to be predicted from the historical building information of the multiple original buildings. The source - domain building information is the historical building information with the highest similarity to the historical building information of the building to be predicted among the historical building information of the multiple original buildings; Use transfer learning to determine the target energy consumption prediction model of the building to be predicted based on the source - domain building information. As Figure 3 shown, the method includes: S201. Obtain the current building information of the building to be predicted and the historical building information of multiple original buildings.

[0038] Among them, the multiple original buildings at least include the building to be predicted.

[0039] S202. Screen the source domain building information corresponding to the building to be predicted from the historical building information of multiple original buildings.

[0040] The source domain building information is the historical building information with the highest similarity to the historical building information of the building to be predicted among the historical building information of multiple original buildings.

[0041] It can be considered that traditional transfer learning generally directly uses the collected historical building data as the source domain data. In this embodiment, however, the historical building data is screened to select the historical building data with high similarity to the target domain data as the source domain data, that is, the historical building information with the highest similarity to the historical building information of the building to be predicted among the historical building information of multiple original buildings is used as the source domain building information. Among them, the means for determining the source domain building information in this embodiment is not limited. For example, a pre-set neural network model can be used to directly output the source domain building information corresponding to the building to be predicted by inputting the historical building information of multiple original buildings into the neural network model; or clustering analysis can be performed on the historical building information of multiple original buildings to obtain similar buildings among multiple original buildings that have a similar category relationship with the building to be predicted; and then the historical building information of the similar buildings is determined as the source domain building information corresponding to the building to be predicted. Among them, the process of clustering analysis is not limited. For example, an autoencoder or other screening methods can be used as the processing means for clustering analysis, aiming to screen out other data with high similarity to the historical building information of the building to be predicted and determine it as the source domain building information.

[0042] As a feasible implementation method, performing clustering analysis on the historical building information of multiple original buildings to obtain similar buildings among multiple original buildings that have a similar category relationship with the building to be predicted includes: For the historical building information of each original building, perform data compression processing on the historical building information of each original building to obtain the target building features of the historical building information of each original building; Calculate the Euclidean distance between the target building features to obtain a distance matrix corresponding to multiple original buildings; Perform category processing on the distance matrix to obtain similar buildings among multiple original buildings that have a similar category relationship with the building to be predicted.

[0043] Figure 4 It is a schematic flow diagram of a clustering analysis provided by an embodiment of the present application, such as Figure 4As shown, the clustering analysis process can be implemented using an autoencoder model. Its operating mechanism mainly involves the neural network learning the latent representation of the data, extracting key features while reducing the dimensionality, and effectively screening out the source domain data with high similarity to the target domain. Among them, the autoencoder model can consist of an encoder and a decoder. The encoder is responsible for compressing the input data to a low-dimensional bottleneck layer, and the decoder is dedicated to reconstructing the original input data based on the latent features. Among them, during the training phase, by continuously minimizing the reconstruction loss, the autoencoder model forces the bottleneck layer to retain the key information of the data. After training is completed, the latent features output by the encoder are used to construct a distance matrix, and a hierarchical clustering strategy is adopted for class division.

[0044] Taking the historical building information of the original buildings, including the local data of Building #A, Building #B, and Building #C, as an example, for the historical building information of each original building, an encoder can be used to perform data compression processing on the historical building information of each original building. For example, when the local data includes time variables, indoor and outdoor environment, and energy consumption data, etc., the encoder can compress the high-dimensional local data into 1-dimensional target building features. The target building features can be seasonal data, or time feature data encoding weekdays and rest days, etc.; the decoder can calculate the Euclidean distances between each target building feature (such as Output #1, Output #2, and Output #3) to obtain a distance matrix corresponding to multiple original buildings; finally, a hierarchical clustering strategy is used to perform class division on the distance matrix, and it can be obtained that Building #A and Building #B belong to Category #1, and Building #C belongs to Category #2. Assuming that the building to be predicted is Building #A, then through the autoencoder clustering analysis, it can be determined that the similar building with a similar category relationship to Building #A is Building #B.

[0045] S203. Adopt the transfer learning method to determine the target energy consumption prediction model of the building to be predicted based on the source domain building information.

[0046] In this step, the transfer learning method can be adopted to directly transfer all the parameters of the pre-trained model trained on the source domain data to the target task as the initial state of the target domain data model training. For example, an improved KAN network model can be trained based on the source domain building information to obtain a source domain energy consumption prediction model; all the weight parameters of all layers in the source domain energy consumption prediction model are completely retained and further trained on the target domain data, and these parameters are fine-tuned to adapt to the specific requirements of the target task, that is, knowledge transfer processing is performed on the source domain energy consumption prediction model based on the historical building information of the building to be predicted to obtain the target energy consumption prediction model of the building to be predicted. On this basis, the general feature representation already learned in the source domain model can be utilized to accelerate the training process of the target domain and improve the model performance.

[0047] S204. Input the current building information into the target energy consumption prediction model to obtain the target energy consumption of the building to be predicted output by the target energy consumption prediction model.

[0048] A cross-building energy consumption collaborative prediction method based on an improved KAN network provided in this embodiment screens the source domain building information corresponding to the building to be predicted from the historical building information of multiple original buildings. The source domain building information is the historical building information with the highest similarity to the historical building information of the building to be predicted among the historical building information of multiple original buildings. Then, by using the transfer learning method, determining the target energy consumption prediction model of the building to be predicted based on the source domain building information can improve the efficiency of source domain knowledge transfer, solve the problems of insufficient data coverage and poor generalization ability, and further improve the prediction accuracy of building energy consumption.

[0049] Figure 5 It is a schematic flowchart of a cross-building energy consumption collaborative prediction method based on an improved KAN network provided in another embodiment of the present application. In this embodiment, obtaining the target energy consumption prediction model of the building to be predicted is further optimized as follows: training the improved KAN network model based on the historical building information of the building to be predicted to obtain the current energy consumption prediction model; in each learning round of federated learning, dynamically obtaining the current random data sent by the server, and inputting the current random data into the current energy consumption prediction model to obtain the current energy consumption prediction value; dynamically sending the current energy consumption prediction value to the server so that the server performs clustering processing on the current energy consumption prediction values corresponding to multiple original buildings to obtain similar buildings having a similar category relationship with the building to be predicted; receiving the target model parameters returned by the server based on the current model parameters of the current energy consumption prediction model of the target building, where the target building includes the building to be predicted and the similar buildings; dynamically updating the current energy consumption prediction model of the building to be predicted based on the target model parameters until the updated current energy consumption prediction model converges; determining the updated current energy consumption prediction model that has converged as the target energy consumption prediction model of the building to be predicted. As Figure 5 shown, the method includes: S301. Obtain the current building information of the building to be predicted, and train the improved KAN network model based on the historical building information of the building to be predicted to obtain the current energy consumption prediction model.

[0050] S302. In each learning round of federated learning, dynamically obtain the current random data sent by the server, and input the current random data into the current energy consumption prediction model to obtain the current energy consumption prediction value.

[0051] S303. Dynamically send the current energy consumption prediction value to the server so that the server performs clustering processing on the current energy consumption prediction values corresponding to multiple original buildings to obtain similar buildings having a similar category relationship with the building to be predicted.

[0052] S304. Receive the target model parameters returned by the server based on the current model parameters of the current energy consumption prediction model of the target building, where the target building includes the building to be predicted and similar buildings.

[0053] S305. Dynamically update the current energy consumption prediction model of the building to be predicted based on the target model parameters until the updated current energy consumption prediction model converges.

[0054] S306. Determine the updated current energy consumption prediction model that has converged as the target energy consumption prediction model of the building to be predicted.

[0055] S307. Input the current building information into the target energy consumption prediction model to obtain the target energy consumption of the building to be predicted output by the target energy consumption prediction model.

[0056] Among them, the current random data can be a set of data randomly generated by the server and used for clustering the current energy consumption prediction values corresponding to multiple original buildings. The current energy consumption prediction value is the energy consumption prediction value output after inputting the current random data into the current energy consumption prediction model.

[0057] In the process of specifically training the model, after obtaining the current energy consumption prediction model in each round of training, the model parameters of the current energy consumption prediction model can be dynamically updated to obtain the updated current energy consumption prediction model, and then the next round of local model training can be carried out until the updated current energy consumption prediction model converges. Finally, the updated current energy consumption prediction model that has converged can be determined as the target energy consumption prediction model of the building to be predicted.

[0058] Specifically, the update of the model parameters can be based on the results of the server's clustering of multiple original buildings. Exemplarily, each client can dynamically obtain the current random data sent by the server and dynamically send the current energy consumption prediction values output by their respective current energy consumption prediction models based on the current random data to the server. Then, after receiving the current energy consumption prediction values corresponding to multiple original buildings, the server can perform clustering processing on the current energy consumption prediction values corresponding to multiple original buildings to obtain groups of similar buildings, and thus obtain similar buildings that have a similar category relationship with the building to be predicted. The server can determine the target model parameters based on the current model parameters of each current energy consumption prediction model corresponding to the similar building groups and return the target model parameters to the clients in the similar building groups. Therefore, the clients in the similar building groups can use the federated learning method to receive the target model parameters and dynamically update their current energy consumption prediction models. Taking the client corresponding to the building to be predicted as an example, this client can dynamically update the current energy consumption prediction model of the building to be predicted based on the target model parameters. That is, after each time this client receives the current random data sent by the server, it can implement the dynamic update of the model parameters of the current energy consumption prediction model according to the above steps until the updated current energy consumption prediction model converges. Then, the updated current energy consumption prediction model that has converged can be determined as the target energy consumption prediction model of the building to be predicted.

[0059] Furthermore, the means for the server to calculate the target model parameters are not limited. For example, it can simply select one of the current model parameters as the target model parameter by comparing the current model parameters of each current energy consumption prediction model corresponding to the similar building groups. For example, it can select the largest current model parameter as the target model parameter from the current model parameter corresponding to the building to be predicted and the current model parameters corresponding to the similar buildings that have a similar category relationship with the building to be predicted. Or, it can also calculate the target model parameters by performing certain calculations on the current model parameters of each current energy consumption prediction model corresponding to the similar building groups.

[0060] Optionally, the target model parameters are the model parameters obtained by the server after performing a mean processing on the current model parameter corresponding to the building to be predicted and the current model parameters corresponding to the similar buildings. On this basis, the aggregation method of the mean processing does not depend on a complex weight allocation strategy and can balance the contributions of each client to the global model only through arithmetic averaging. This method is not only easy to implement and understand but also can, to a certain extent, avoid the deviation of the global model parameters caused by abnormal data of individual clients or excessive differences in model training.

[0061] Figure 6 is a schematic flowchart of a federated learning provided by an embodiment of the present application, as Figure 6As shown, in the initial stage of federated learning, the server can initialize the global model and distribute it to all clients for local training. Taking the current existing Building #A, Building #B, and Building #C as examples, each client can be responsible for the model training of one of the buildings, that is, each client can perform local training based on the local data of its respective building to obtain the local model of the current round (that is, train the improved KAN network model based on the historical building information of the building to be predicted to obtain the current energy consumption prediction model); then, the server can generate a set of random but fixed input data (that is, the current random data) and send it to each client so that each client can generate model prediction values (that is, the current energy consumption prediction values) according to its respective local model.

[0062] Using these model prediction values as the basis for clustering analysis, a clustering result can be obtained for personalized federated learning. The core principle is that similar parameter models will produce similar outputs under the same input, which enables the classification results of similar buildings in each round of federated learning to be dynamically adjusted, thereby achieving more accurate and adaptable personalized federated learning. For example, in the clustering analysis process of the current round, a distance matrix can be generated based on the model prediction values output by the model (such as Output #1, Output #2, and Output #3), and then a hierarchical clustering result can be obtained according to the distance matrix, that is, Building #A and Building #B belong to Category #1, and Building #C belongs to Category #2. Then the server can perform mean processing on the current model parameters of the buildings corresponding to the same category to obtain the target model parameters, so as to update the current energy consumption prediction model of the buildings corresponding to the same category. In the clustering analysis process of the next round, the hierarchical clustering result of similar buildings can be updated according to the new model prediction values, so that in each round of federated learning, the current energy consumption prediction model of similar buildings can be dynamically adjusted according to different hierarchical clustering results.

[0063] It can be considered that, within the framework of federated learning, the parameters aggregated by the IKAN model proposed in this embodiment are not simple weights, but the parameter changes of the Mexican hat wavelet function, that is, the current model parameters. Further, for the parameter aggregation of the IKAN model, this embodiment adopts the method of average aggregation. Specifically, the operation process of this method is as follows: at the end of each training round, each client sends the Mexican hat wavelet function parameters of the IKAN model obtained by local training to the central server; after receiving the current model parameters of each client, the central server calculates the average value of the current model parameters corresponding to the hierarchical clustering results of similar buildings, and uses this as the aggregated global model parameters (that is, the target model parameters). Subsequently, this new global model parameter will be distributed to all clients corresponding to similar buildings. After each client updates the current model parameters of the current energy consumption prediction model based on the target model parameters, it can perform the next round of local training until the updated current energy consumption prediction model converges.

[0064] More specifically, the number of central servers can be one or more. For example, in the clustering analysis process of each round, different central servers can be divided according to different hierarchical clustering results. For example, the average value between the current model parameters of buildings of the same category can be calculated by the same central server, and the central server distributes the calculated target model parameters to the clients corresponding to the buildings of this category; while the average value between the current model parameters of buildings of different categories is calculated by different central servers.

[0065] It should be noted that in standard federated learning, the participating buildings are fixed in distributed training, and the grouping of similar buildings cannot be dynamically adjusted, so it is difficult to effectively utilize the similarity between buildings to improve the model performance. By adopting the federated learning strategy based on dynamic model parameters in this embodiment, while protecting data privacy, it can dynamically identify and utilize the similarity between buildings, not only improving the adaptability of the model to the data distribution of each client, but also enhancing the flexibility and accuracy of the federated learning framework, enabling it to better cope with the dynamic changes and heterogeneous requirements of building data, and realizing dynamic personalized federated learning.

[0066] Figure 7 It is a schematic diagram of the architecture of a cross-building energy consumption collaborative prediction method based on an improved KAN network provided by another embodiment of the present application. Refer to Figure 7 In the data acquisition stage, step S1 can acquire multi-source building operation data. In this embodiment, 5 building communities with no more than 20 buildings can be selected. As shown in Table 1, it is a summary of the characteristics of different building communities. The number of buildings in each building community ranges from 5 to 19, the single building area ranges from 943.9㎡ to 70999.9㎡, and the main building types are for educational use and office use.

[0067] The operation data of each building is collected once per hour, with a time span of about two years. The operation data covers key information such as building type and building area. At the same time, it involves time variables, energy consumption data, and indoor and outdoor environmental parameters, such as dry bulb temperature, wet bulb temperature, wind speed, etc., and relevant data such as time variables will also be included. As shown in Table 2 and the continued Table 2, several key numerical variables included in different building communities are summarized. The indoor and outdoor environmental parameters can include dry bulb temperature, wet bulb temperature, and wind speed. It can be found that the outdoor environmental conditions of each building community are highly similar.

[0068] Table 1 Summary of the characteristics of different building communities

[0069] Table 2 Key numerical variables included in different building communities

[0070] Continued Table 2 Key numerical variables included in different building communities

[0071] Step S2 can select a strategy according to the application scenario requirements, that is, select a suitable cross-domain machine learning paradigm, including a transfer learning strategy or a federated learning strategy. Since the application scenario of this embodiment meets the requirements of the two cross-domain machine learning paradigms, the transfer learning and federated learning paradigms can be respectively used to construct a cross-building energy consumption prediction model.

[0072] Step S3 can be divided into steps A3 and B3. Step A3 can include using an autoencoder to screen source domain data. That is, in the collected multi-source building data, select a building as the local data set, that is, determine the target domain data. Subsequently, an autoencoder can be used as a clustering processing means to screen out other data with a high similarity to the target domain data and determine it as the source domain data. Finally, all the model parameters trained on the source domain data can be migrated and fine-tuned on the target domain to obtain an energy consumption prediction model for the target domain (i.e., the target energy consumption prediction model). Step B3 can include a strategy based on dynamic model parameters. That is, in this embodiment, federated learning can adopt a strategy based on dynamic model parameters. This strategy can achieve dynamic personalization, allow similar building groups to change over time, better adapt to the dynamic and complex requirements of data characteristics, and can operate effectively under computational resource constraints.

[0073] Among them, whether transfer learning or federated learning is selected as the cross-building machine learning paradigm, when implementing the building energy consumption prediction task, input variables of the prediction model need to be selected. In this embodiment, since building operation data usually has obvious seasonality, three time variables can be selected as inputs, namely month, day type (i.e., Monday to Sunday), and hour number (i.e., 0 to 23) as representative variables to reflect the influence of implicit information (such as unmeasurable factors like the number of residents). And environmental variables can describe the seasonal changes of outdoor conditions, which will significantly affect the operation mode of buildings and their energy consumption. In addition, considering that the difference in the scale of the same type of building may lead to completely different operation modes, the total building floor area is also used for predictive modeling. In the data preprocessing stage, traditional standardization methods can be used to normalize numerical input variables to construct an improved KAN network prediction model (i.e., the target energy consumption prediction model) based on the preprocessed input variables.

[0074] Step S4 can then implement the building energy consumption prediction task according to the constructed improved KAN network prediction model. For example, an improved KAN network prediction model can be used to predict the hourly electricity consumption of a building based on data such as time, environmental variables, and building characteristics.

[0075] In addition, in this embodiment, a multi-layer perceptron (MLP) prediction can also be used as a control group to evaluate the performance advantages of the proposed transfer learning and federated learning methods. An MLP can be a basic feedforward artificial neural network composed of multiple layers, including an input layer, hidden layers, and an output layer. Each layer consists of multiple neurons, and the neurons are connected by weights. The MLP is trained using a supervised learning technique called backpropagation, which optimizes the weights through the chain rule to minimize the prediction error.

[0076] Through the comparative analysis of the implementation results, it is found that compared with the traditional local MLP modeling prediction method, the proposed transfer learning method performs excellently in the energy consumption prediction task, and can reduce the average value of the energy consumption prediction error by 7.31%; while the federated learning method also shows significant advantages, and this method can reduce the average value of the energy consumption prediction error by 6.70%. Further, the prediction technology based on the IKAN model further improves the prediction accuracy on the basis of the former two, reaching 8.06% and 10.07% respectively. This shows that both the transfer learning and federated learning methods can effectively mine the knowledge in building data and achieve cross-building sharing, effectively solve the problem of building data silos, and realize cross-building knowledge transfer without sharing the original data, thereby improving the accuracy of single-building energy consumption prediction. At the same time, with its unique neural network architecture, the IKAN model is more efficient than the traditional MLP model in the transfer and federated learning methods in building energy consumption prediction, can adapt to the diverse characteristics of buildings of different scales and types, has strong generalization ability and practical value, provides more accurate prediction support for the building energy management system, and helps buildings to save energy and reduce emissions and achieve the carbon neutrality goal.

[0077] From the above description, it can be found that the core idea of the cross-building energy consumption collaborative prediction method based on the improved KAN network in this embodiment mainly includes: on the one hand, a transfer learning strategy based on data similarity is designed. Different from the existing transfer learning that relies on experience to select the source domain building, in this embodiment, through the clustering method of the autoencoder model, the source domain data with high similarity to the target building is systematically screened, improving the efficiency of source domain knowledge transfer and solving the problems of insufficient data coverage and poor generalization ability.

[0078] On the other hand, aiming at the heterogeneity of building data, a federated learning strategy based on dynamic model parameters is proposed. By dynamically adjusting the grouping of similar buildings according to the model prediction results, the problem of data heterogeneity caused by the differences in building operation rules is effectively solved, enabling the model to adapt to the data distribution differences of different types of buildings, improving the dynamic adjustment ability of the model, and providing an efficient and scalable solution for building energy conservation management.

[0079] Meanwhile, aiming at the problem of limited expression ability of traditional neural network models, an improved KAN model based on the Mexican hat wavelet function was designed, which enhanced the model's ability to capture multi-scale non-linear features and its performance in small-sample scenarios. At the same time, the network structure was simplified, making the model more interpretable and significantly improving the accuracy of energy consumption prediction in small-sample scenarios.

[0080] Therefore, the cross-building energy consumption collaborative prediction method based on the improved KAN network in this embodiment aims to solve the technical bottlenecks of data island problems and insufficient model generalization ability in the field of building energy consumption prediction. Through an innovative machine learning paradigm, it realizes knowledge sharing and collaborative prediction across buildings, and uses the improved KAN model as the neural network architecture to significantly improve the energy consumption prediction accuracy of individual buildings while protecting data privacy, providing effective technical support for building energy conservation and carbon neutrality goals. Therefore, this embodiment can be applied to various building systems that require energy consumption prediction, especially suitable for scenarios where the data of a single building is limited but the data of similar building types is rich.

[0081] Corresponding to the cross-building energy consumption collaborative prediction method based on the improved KAN network in the above embodiment, Figure 8 is a structural block diagram of a cross-building energy consumption collaborative prediction device provided by an embodiment of the present application. For the sake of convenience of description, only the parts related to the embodiment of the present application are shown.

[0082] Referring to Figure 8 , the device includes: An acquisition module 401, configured to acquire the current building information of the building to be predicted and the target energy consumption prediction model, where the target energy consumption prediction model is determined by training an improved Kolmogorov-Arnold KAN network model based on the knowledge sharing of the historical building information of multiple original buildings, and the activation function of the improved KAN network model is the Mexican hat wavelet function; An input module 402, configured to input the current building information into the target energy consumption prediction model to obtain the target energy consumption of the building to be predicted output by the target energy consumption prediction model.

[0083] A cross-building energy consumption collaborative prediction device based on an improved KAN network provided in this embodiment obtains the current building information of the building to be predicted and the target energy consumption prediction model through an acquisition module. Among them, the target energy consumption prediction model is determined by training an improved Kolmogorov-Arnold KAN network model based on the knowledge sharing of the historical building information of multiple original buildings. The activation function of the improved KAN network model is the Mexican hat wavelet function. The current building information is input into the target energy consumption prediction model through an input module, and the target energy consumption of the building to be predicted output by the target energy consumption prediction model is obtained. Using this device, the target energy consumption prediction model is determined by training an improved KAN network model based on the building information across buildings. The activation function of the improved KAN network model is the Mexican hat wavelet function, which can enhance the expression ability of the target energy consumption prediction model in capturing the non-linear relationship and multi-scale features in building energy consumption data while realizing cross-building knowledge sharing, thereby improving the prediction accuracy of building energy consumption.

[0084] Optionally, the acquisition module includes: An acquisition unit for acquiring the historical building information of multiple original buildings, and the multiple original buildings at least include the building to be predicted; An information screening unit for screening the source domain building information corresponding to the building to be predicted from the historical building information of multiple original buildings. The source domain building information is the historical building information with the highest similarity to the historical building information of the building to be predicted among the historical building information of multiple original buildings; A model determination unit for determining the target energy consumption prediction model of the building to be predicted based on the source domain building information by using the transfer learning method.

[0085] Optionally, the information screening unit includes: A clustering analysis subunit for performing clustering analysis on the historical building information of multiple original buildings to obtain similar buildings with a similar category relationship with the building to be predicted among the multiple original buildings; A determination subunit for determining the historical building information of the similar buildings as the source domain building information corresponding to the building to be predicted.

[0086] Optionally, the clustering analysis subunit is specifically used for: For the historical building information of each original building, performing data compression processing on the historical building information of each original building to obtain the target building features of the historical building information of each original building; Calculating the Euclidean distance between the target building features to obtain a distance matrix corresponding to multiple original buildings; Performing category processing on the distance matrix to obtain similar buildings with a similar category relationship with the building to be predicted among the multiple original buildings.

[0087] Optionally, the model determination unit is specifically used for: Train an improved KAN network model based on the building information in the source domain to obtain a source domain energy consumption prediction model; Perform knowledge transfer processing on the source domain energy consumption prediction model based on the historical building information of the building to be predicted to obtain a target energy consumption prediction model for the building to be predicted.

[0088] Optionally, the acquisition module is specifically configured to: Train an improved KAN network model based on the historical building information of the building to be predicted to obtain a current energy consumption prediction model; In each learning round of federated learning, dynamically obtain the current random data sent by the server, and input the current random data into the current energy consumption prediction model to obtain a current energy consumption prediction value; Dynamically send the current energy consumption prediction value to the server so that the server can perform clustering processing on the current energy consumption prediction values corresponding to multiple original buildings to obtain similar buildings that have a similar category relationship with the building to be predicted; Receive the target model parameters returned by the server based on the current model parameters of the current energy consumption prediction model of the target building, where the target building includes the building to be predicted and the similar buildings; Dynamically update the current energy consumption prediction model of the building to be predicted based on the target model parameters until the updated current energy consumption prediction model converges; Determine the updated current energy consumption prediction model that has converged as the target energy consumption prediction model of the building to be predicted.

[0089] Optionally, the target model parameters are the model parameters obtained by the server through averaging the current model parameters corresponding to the building to be predicted and the current model parameters corresponding to the similar buildings.

[0090] Optionally, the current building information at least includes current time data, current environmental data, and building area data.

[0091] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units, due to being based on the same concept as the method embodiment of the present application, for their specific functions and the technical effects brought, please refer to the method embodiment part specifically, and will not be elaborated here.

[0092] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction and do not limit the protection scope of this application. The specific working processes of the units and modules in the above system can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated here.

[0093] An embodiment of this application also provides a terminal device. Figure 9 FIG. is a schematic structural diagram of a terminal device provided by an embodiment of this application. As Figure 9 shown, the terminal device includes: at least one processor 501, a memory 502, an input device 503, an output device 504, and a computer program stored in the memory 502 and executable on at least one processor 501. When the processor 501 executes the computer program, the steps in any of the foregoing method embodiments are implemented.

[0094] The input device 503 can be used to receive input digital or character information, and generate key signal inputs related to the user settings and function controls of the terminal device. The output device 504 can include display devices such as a display screen.

[0095] An embodiment of this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program. When the computer program is executed by the processor 501, the steps in any of the foregoing method embodiments can be implemented.

[0096] An embodiment of this application provides a computer program product. When the computer program product runs on the terminal device, the terminal device can be caused to execute the steps in any of the foregoing method embodiments.

[0097] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, to implement all or part of the processes in the above method embodiments of this application, a computer program can be used to instruct the relevant hardware to complete. The computer program can be stored in the computer-readable storage medium. When the computer program is executed by the processor 501, the steps of the above method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable storage medium can at least include: any entity or device that can carry the computer program code to the device / terminal device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium. For example, USB flash drive, mobile hard disk, magnetic disk or optical disc, etc. In some jurisdictions, according to legislation and patent practice, the computer-readable storage medium cannot be an electrical carrier signal and a telecommunication signal.

[0098] In the above embodiments, the descriptions of the respective embodiments have their own focuses. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0099] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0100] In the embodiments provided in this application, it should be understood that the disclosed device / terminal device and method can be implemented in other ways. For example, the device / terminal device embodiments described above are only illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical, mechanical or other form.

[0101] The unit described as a separation component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place or may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0102] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A cross-building energy consumption collaborative prediction method based on an improved KAN network, characterized in that, Including: Obtain the current building information of the building to be predicted and the target energy consumption prediction model. Among them, the target energy consumption prediction model is determined by training and improving the Kolmogorov - Arnold (KAN) network model based on the knowledge sharing of the historical building information of multiple original buildings. The activation function of the improved KAN network model is the Mexican hat wavelet function; Input the current building information into the target energy consumption prediction model to obtain the target energy consumption of the building to be predicted output by the target energy consumption prediction model.

2. The cross-building energy consumption collaborative prediction method based on the improved KAN network according to claim 1, characterized in that, Obtaining the target energy consumption prediction model of the building to be predicted includes: Obtain the historical building information of multiple original buildings, and the multiple original buildings at least include the building to be predicted; Screen the source domain building information corresponding to the building to be predicted from the historical building information of the multiple original buildings. The source domain building information is the historical building information with the highest similarity to the historical building information of the building to be predicted among the historical building information of the multiple original buildings; Adopt the transfer learning method to determine the target energy consumption prediction model of the building to be predicted based on the source domain building information.

3. The cross-building energy consumption collaborative prediction method based on the improved KAN network according to claim 2, wherein, The screening of the source domain building information corresponding to the building to be predicted from the historical building information of the multiple original buildings includes: Conduct cluster analysis on the historical building information of the multiple original buildings to obtain similar buildings among the multiple original buildings that have a similar category relationship with the building to be predicted; Determine the historical building information of the similar buildings as the source domain building information corresponding to the building to be predicted.

4. The cross-building energy consumption collaborative prediction method based on the improved KAN network according to claim 3, wherein, The conducting of cluster analysis on the historical building information of the multiple original buildings to obtain similar buildings among the multiple original buildings that have a similar category relationship with the building to be predicted includes: For the historical building information of each original building, conduct data compression processing on the historical building information of each original building to obtain the target building features of the historical building information of each original building; Calculate the Euclidean distance between the target building features to obtain the distance matrix corresponding to the multiple original buildings; Conduct category processing on the distance matrix to obtain similar buildings among the multiple original buildings that have a similar category relationship with the building to be predicted.

5. The cross-building energy consumption collaborative prediction method based on the improved KAN network according to claim 2, wherein, The adopting of the transfer learning method to determine the target energy consumption prediction model of the building to be predicted based on the source domain building information includes: Train the improved KAN network model based on the source domain building information to obtain the source domain energy consumption prediction model; Conduct knowledge transfer processing on the source domain energy consumption prediction model based on the historical building information of the building to be predicted to obtain the target energy consumption prediction model of the building to be predicted.

6. The cross-building energy consumption collaborative prediction method based on the improved KAN network according to claim 1, wherein, Obtaining the target energy consumption prediction model of the building to be predicted includes: Train the improved KAN network model based on the historical building information of the building to be predicted to obtain the current energy consumption prediction model; In each learning round of federated learning, dynamically obtain the current random data sent by the server, and input the current random data into the current energy consumption prediction model to obtain the current energy consumption prediction value; Dynamically send the current energy consumption prediction value to the server, so that after the server clusters the current energy consumption prediction values corresponding to multiple original buildings, similar buildings with a similar category relationship to the building to be predicted are obtained; Receive the target model parameters returned by the server based on the current model parameters of the current energy consumption prediction model of the target building, where the target building includes the building to be predicted and the similar buildings; Dynamically update the current energy consumption prediction model of the building to be predicted based on the target model parameters until the updated current energy consumption prediction model converges; Determine the updated current energy consumption prediction model that has converged as the target energy consumption prediction model of the building to be predicted.

7. The cross-building energy consumption collaborative prediction method based on the improved KAN network according to claim 6, wherein The target model parameters are the model parameters obtained by the server after averaging the current model parameters corresponding to the building to be predicted and the current model parameters corresponding to the similar buildings.

8. The cross-building energy consumption collaborative prediction method based on the improved KAN network according to any one of claims 1-7, characterized in that The current building information at least includes current time data, current environmental data, and building area data.

9. A cross-building energy consumption collaborative prediction device based on an improved KAN network, characterized in that, It includes: An acquisition module, configured to acquire the current building information and the target energy consumption prediction model of the building to be predicted, where the target energy consumption prediction model is determined by training and improving the Kolmogorov - Arnold KAN network model based on the knowledge sharing of the historical building information of multiple original buildings, and the activation function of the improved KAN network model is the Mexican hat wavelet function; An input module, configured to input the current building information into the target energy consumption prediction model to obtain the target energy consumption of the building to be predicted output by the target energy consumption prediction model.

10. A terminal device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, the terminal device implements the method according to any one of claims 1 - 8.

Images