Building energy consumption prediction system and method based on multi-source data fusion

Abstract

The invention relates to the technical field of data reasoning, and discloses a building energy consumption prediction system and method based on multi-source data fusion, the system comprises a data processing module, a feature weight distribution module, a feature fusion module, an energy consumption prediction module and an energy consumption optimization module, obtaining a standardized energy consumption data set of the building; determining the importance weight of the standardized energy consumption data set and the correlation weight of the standardized energy consumption data set; performing weighted fusion on the depth features extracted from the multi-source energy consumption heterogeneous data to obtain multi-source feature vectors, and performing splicing dimension reduction on the multi-source feature vectors to obtain fused feature vectors; extracting local time sequence mode characteristics and a long-term time sequence dependency relationship, and performing energy consumption prediction on the building to obtain a building energy consumption prediction value; matching an energy consumption abnormal mode in the pre-obtained building operation characteristic library, and generating an energy consumption optimization report of the building according to the high-energy-consumption area corresponding to the energy consumption abnormal mode; the accuracy of building energy consumption prediction can be improved.

Description

Technical Field

[0001] This invention relates to the field of data reasoning technology, and in particular to a building energy consumption prediction system and method based on multi-source data fusion. Background Technology

[0002] In the field of building energy consumption forecasting, multi-source heterogeneous energy consumption data (such as equipment operating energy consumption, ambient temperature and humidity, and building usage load) are the core inputs supporting predictive analysis. However, existing technologies have significant shortcomings in the multi-source data preprocessing stage. To address the common problems of missing values, inconsistent dimensions, and outlier data encountered during data collection, existing processing methods often employ simple linear interpolation or fixed threshold filtering strategies. These methods fail to adequately adapt to the temporal correlation and dynamic changes in energy consumption data, and struggle to accurately identify and remove outlier data points that deviate from actual operating conditions. This results in preprocessed datasets with insufficient accuracy and poor consistency, directly weakening the quality of the foundational data for subsequent energy consumption forecasting and making it difficult to meet the demands of high-precision prediction.

[0003] Existing building energy consumption prediction technologies also fall short in feature fusion and prediction optimization. In the feature processing stage, existing solutions often fail to fully consider the varying importance of different feature dimensions within the same data source to energy consumption, nor do they effectively quantify the complementarity and overlap of information between different data sources. They frequently employ equal weights or single fusion rules for feature integration, easily leading to the weakening of key feature information and the accumulation of redundant data. The resulting feature vectors cannot accurately reflect the core driving factors of energy consumption changes. In the prediction and optimization stages, most models can only capture local temporal fluctuations or long-term trends in a single dimension of energy consumption data, making it difficult to comprehensively depict the complex patterns of energy consumption changes. This results in significant discrepancies between prediction results and actual energy consumption. Furthermore, in subsequent energy consumption optimization, they cannot perform deep matching with building operation characteristic databases to accurately locate high-energy-consuming areas and inefficient equipment. The generated optimization schemes lack specificity and are unable to effectively reduce building energy consumption. Summary of the Invention

[0004] This invention provides a building energy consumption prediction system and method based on multi-source data fusion to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a building energy consumption prediction system based on multi-source data fusion, characterized in that the system includes a data processing module, a feature weight allocation module, a feature fusion module, an energy consumption prediction module, and an energy consumption optimization module, wherein:

[0006] The data processing module is used to fill in missing values ​​in the heterogeneous multi-source energy consumption data collected within the building, and to unify the heterogeneous multi-source energy consumption data after missing value filling to the same dimension, thereby obtaining the standardized energy consumption dataset of the building.

[0007] The feature weight allocation module is used to determine the importance weights of different feature dimensions in the same data source in the standardized energy consumption dataset, and the correlation weights between different data sources in the standardized energy consumption dataset.

[0008] The feature fusion module is used to perform weighted fusion of the deep features extracted from the multi-source energy consumption heterogeneous data based on the importance weight and the relevance weight, to obtain the multi-source feature vector of the building, and to concatenate and reduce the dimensionality of the multi-source feature vector to obtain the fused feature vector of the building.

[0009] The energy consumption prediction module is used to extract the local temporal pattern features and long-term temporal dependencies of the fused feature vector, and to predict the energy consumption of the building based on the local temporal pattern features and the long-term temporal dependencies, so as to obtain the predicted building energy consumption value for a specific time period in the future.

[0010] The energy consumption optimization module is used to match the predicted building energy consumption value with the energy consumption anomaly patterns in the pre-acquired building operation characteristic library, and generate an energy consumption optimization report for the building based on the high energy consumption area corresponding to the energy consumption anomaly pattern.

[0011] In a preferred embodiment, when the data processing module performs missing value imputation on the heterogeneous multi-source energy consumption data collected within the building and unifies the imputed heterogeneous multi-source energy consumption data to the same dimension to obtain the standardized energy consumption dataset of the building, it is specifically used for:

[0012] Identify missing timestamp data in heterogeneous multi-source energy consumption data;

[0013] Time series interpolation is performed on the missing timestamp data to obtain multi-source energy consumption heterogeneous data after missing values ​​are filled.

[0014] Remove the abnormal data points that deviate from the overall data distribution pattern in the multi-source energy consumption heterogeneous data after the missing value filling is completed, and obtain the cleaned multi-source energy consumption heterogeneous data;

[0015] Based on the statistical distribution parameters of the feature data in the cleaned multi-source energy consumption heterogeneous data, the feature data is converted into standardized data with the same unit of measurement to obtain the standardized energy consumption dataset of the building.

[0016] In a preferred embodiment, when the feature weight allocation module determines the importance weights of different feature dimensions within the same data source in the standardized energy consumption dataset, and the correlation weights between different data sources in the standardized energy consumption dataset, it is specifically used for:

[0017] Based on the correlation between the internal feature dimensions of the data source and the building energy consumption value in the standardized energy consumption dataset, an importance weight value is assigned to the internal feature dimensions of the data source.

[0018] Based on the data complementarity and information overlap between different data sources in the standardized energy consumption dataset, the correlation weight values ​​between the different data sources are determined.

[0019] Establish a table mapping feature dimension importance weights to data source relevance weights;

[0020] The corresponding relationship table is used as a weight identifier and associated with the standardized energy consumption dataset for storage to obtain the energy consumption weight allocation scheme of the building.

[0021] In a preferred embodiment, when the feature fusion module performs weighted fusion of deep features extracted from the multi-source energy consumption heterogeneous data based on the importance weight and the relevance weight to obtain the multi-source feature vector of the building, it is specifically used for:

[0022] The deep features of the feature dimension within the same data source are weighted according to the importance weight to obtain the weighted feature representation of the building;

[0023] Based on the relevance weights, the weighted feature representations from the different data sources are fused across sources to obtain the preliminary fused features of the building.

[0024] The preliminary fusion features after dimension alignment are merged to form a high-dimensional fusion feature representation of the building.

[0025] Redundant information in the high-dimensional fusion feature representation is removed while key features are retained to obtain the multi-source feature vector of the building.

[0026] In a preferred embodiment, when the feature fusion module performs dimensionality reduction by concatenating the multi-source feature vectors to obtain the fused feature vector of the building, it is specifically used for:

[0027] The multi-source feature vectors from different data sources are arranged and combined in a preset order to obtain a preliminary concatenated vector of the multi-source feature vectors;

[0028] The features after feature information condensation are mapped to a low-dimensional feature space to obtain the dimensionality-reduced feature representation of the building.

[0029] The reduced-dimensional feature representation is standardized to obtain the fused feature vector of the building.

[0030] In a preferred embodiment, the energy consumption prediction module, when extracting local temporal pattern features and long-term temporal dependencies from the fused feature vector, is specifically used for:

[0031] The fused feature vector is divided into overlapping local time segments according to continuous temporal characteristics to obtain the local temporal window of the fused feature vector;

[0032] Capture the energy consumption fluctuation patterns and periodic characteristics within the short term in the local time window to obtain the local time pattern features of the fused feature vector;

[0033] Based on the feature evolution patterns between the local time windows, the long-term dependencies of the fused feature vector are captured.

[0034] In a preferred embodiment, the energy consumption prediction module performs energy consumption prediction on the building based on the local time-series pattern characteristics and the long-term time-series dependency to obtain the predicted building energy consumption value for a specific future time period, specifically for:

[0035] The local temporal pattern features are integrated with the long-term temporal dependencies to construct the temporal features of the building;

[0036] By analyzing the correspondence between historical energy consumption data and time-series characteristics, an energy consumption prediction model for the building is constructed.

[0037] The energy consumption of the time-series feature representation is predicted using the prediction model to generate the initial predicted building energy consumption value of the building.

[0038] By eliminating abnormal fluctuations in the initial building energy consumption forecast, the building energy consumption forecast for a specific future time period is obtained.

[0039] In a preferred embodiment, the mathematical expression of the energy consumption prediction model is as follows:

[0040] ;

[0041] In the formula, For a moment The predicted energy consumption value, The weights of the historical time-series features in the historical energy consumption data are given. The size of the local timing window. The index for the time-series feature. For the first Weights of each time-series feature, for Quantization values ​​of temporal characteristics at any given time. The weighting of the impact of the rate of change in energy consumption in the historical energy consumption data. For time factors, For the local timing window, Let be the influence function of the rate of change of energy consumption. For the target energy consumption value, For time integration variables, The weight of the building in the current time series state. Let be the feature mapping function of the building in the current time series state. for A set of temporal and temporal characteristics.

[0042] In a preferred embodiment, when the energy consumption optimization module performs the following steps: matching the predicted building energy consumption value with an anomaly pattern in the pre-acquired building operation characteristic database, and generating an energy consumption optimization report for the building based on the high-energy-consumption area corresponding to the anomaly pattern, the module is specifically used for:

[0043] The predicted building energy consumption is compared and analyzed with the standard energy consumption patterns in the pre-acquired building operation characteristic database to obtain the abnormal fluctuation range of the building's energy consumption.

[0044] Based on the abnormal energy consumption fluctuation range, the abnormal energy consumption patterns stored in the pre-acquired building operation characteristic library are matched to obtain the abnormal pattern type of the building and the corresponding high energy consumption area.

[0045] Based on the equipment operating status and energy consumption characteristics of the high-energy-consumption area, locate the specific equipment with low energy efficiency;

[0046] Based on the energy consumption data of the specific equipment and the energy-saving measures knowledge base corresponding to the abnormal mode types in the pre-acquired building operation characteristic library, an energy consumption optimization report for the building is generated.

[0047] To address the aforementioned problems, this invention also provides a building energy consumption prediction method based on multi-source data fusion, the method comprising:

[0048] S1. Impute missing values ​​in the heterogeneous multi-source energy consumption data collected within the building, and unify the heterogeneous multi-source energy consumption data after missing value imputation to the same dimension to obtain the standardized energy consumption dataset of the building.

[0049] S2. Determine the importance weights of different feature dimensions in the same data source within the standardized energy consumption dataset, and the correlation weights between different data sources within the standardized energy consumption dataset;

[0050] S3. Based on the importance weight and the relevance weight, the deep features extracted from the multi-source energy consumption heterogeneous data are weighted and fused to obtain the multi-source feature vector of the building, and the multi-source feature vector is spliced ​​and dimensionality reduced to obtain the fused feature vector of the building.

[0051] S4. Extract the local temporal pattern features and long-term temporal dependencies of the fused feature vector, and perform energy consumption prediction on the building based on the local temporal pattern features and the long-term temporal dependencies to obtain the predicted building energy consumption value for a specific time period in the future.

[0052] S5. Based on the predicted building energy consumption value, match the abnormal energy consumption patterns in the pre-acquired building operation characteristic library, and generate an energy consumption optimization report for the building based on the high energy consumption area corresponding to the abnormal energy consumption pattern.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. This invention significantly improves the quality of basic data and the effectiveness of core features for building energy consumption prediction through a systematic multi-source data processing and feature fusion design. The data processing module addresses heterogeneous multi-source energy consumption data by using time-series interpolation to fill missing values, remove outlier data points, and standardize units, ensuring high accuracy and consistency in the standardized energy consumption dataset. The feature weight allocation module accurately determines the importance weights of different feature dimensions within the same data source and the correlation weights between different data sources. The feature fusion module uses these weights to perform weighted fusion and dimensionality reduction on deep features, effectively preserving key driving information of energy consumption changes and eliminating redundant data, forming a fused feature vector that accurately characterizes energy consumption patterns, thus laying a high-quality data foundation for subsequent prediction stages.

[0055] 2. This invention achieves a dual improvement in energy consumption prediction accuracy and energy consumption management practicality, providing strong support for efficient building energy consumption control. The energy consumption prediction module extracts local temporal pattern features and long-term temporal dependencies from the fused feature vectors. After feature integration and prediction model calculations, and the elimination of abnormal fluctuations, it can accurately output the predicted energy consumption values ​​for a specific future time period. The energy consumption optimization module matches the predicted values ​​with energy consumption anomaly patterns in the building's operational characteristic library, accurately locating high-energy-consuming areas and specific energy-inefficient equipment. This generates targeted energy consumption optimization reports, helping buildings clarify energy-saving directions, effectively improving energy consumption management efficiency, and promoting the rational control and optimization of building energy consumption. Attached Figure Description

[0056] Figure 1 This is a system architecture diagram of a building energy consumption prediction system based on multi-source data fusion provided in an embodiment of the present invention;

[0057] Figure 2This is a flowchart illustrating a building energy consumption prediction method based on multi-source data fusion, provided in an embodiment of the present invention.

[0058] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments belong to some, but not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “said” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0061] Depending on the context, the word "if" or "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0062] Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0063] In practice, the server-side equipment deployed in a building energy consumption prediction system based on multi-source data fusion may consist of one or more devices. This system can be implemented as a business instance, a virtual machine, or hardware devices. For example, it can be implemented as a business instance deployed on one or more devices in a cloud node. Simply put, it can be understood as software deployed on a cloud node to provide multi-source data fusion-based building energy consumption prediction services to various user terminals. Alternatively, it can be implemented as a virtual machine deployed on one or more devices in a cloud node, with application software installed to manage various user terminals. Or, it can be implemented as a server composed of numerous identical or different types of hardware devices, with one or more devices configured to provide multi-source data fusion-based building energy consumption prediction services to various user terminals.

[0064] In terms of implementation, the building energy consumption prediction system based on multi-source data fusion and the user terminal are mutually compatible. That is, if the building energy consumption prediction system based on multi-source data fusion is implemented as an application installed on a cloud service platform, then the user terminal is implemented as a client that establishes a communication connection with the application; or if the building energy consumption prediction system based on multi-source data fusion is implemented as a website, then the user terminal is implemented as a webpage; or if the building energy consumption prediction system based on multi-source data fusion is implemented as a cloud service platform, then the user terminal is implemented as a mini-program in an instant messaging application.

[0065] like Figure 1 The figure shown is a system architecture diagram of a building energy consumption prediction system based on multi-source data fusion provided in an embodiment of the present invention.

[0066] The building energy consumption prediction system 100 based on multi-source data fusion described in this invention can be located on a cloud server. In terms of implementation, it can be used as one or more service devices, or as an application installed on the cloud (e.g., a mobile service operator's server, server cluster, etc.), or it can be developed as a website. Depending on the functions implemented, the building energy consumption prediction system 100 based on multi-source data fusion may include a data processing module 101, a feature weight allocation module 102, a feature fusion module 103, an energy consumption prediction module 104, and an energy consumption optimization module 105. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0067] In this embodiment of the invention, in the building energy consumption prediction system based on multi-source data fusion, each of the above modules can be implemented independently and can call other modules. Here, "calling" can be understood as one module connecting to multiple modules of another type and providing corresponding services to those connected modules. In the building energy consumption prediction system based on multi-source data fusion provided by this embodiment of the invention, the applicability of the system architecture can be adjusted by adding modules and directly calling them without modifying the program code, achieving cluster-based horizontal expansion to quickly and flexibly expand the building energy consumption prediction system based on multi-source data fusion. In practical applications, the above modules can be set in the same device or different devices, or they can be set in a virtual device, such as a service instance in a cloud server.

[0068] The following describes the components and workflow of a building energy consumption prediction system based on multi-source data fusion, using specific embodiments as examples:

[0069] The data processing module 101 is used to fill in missing values ​​in the heterogeneous multi-source energy consumption data collected in the building, and to unify the heterogeneous multi-source energy consumption data after missing value filling to the same dimension, so as to obtain the standardized energy consumption dataset of the building.

[0070] In this embodiment of the invention, when the data processing module performs missing value imputation on the heterogeneous multi-source energy consumption data collected within the building, and unifies the imputed heterogeneous multi-source energy consumption data to the same dimension to obtain the standardized energy consumption dataset of the building, it is specifically used for:

[0071] Identify missing timestamp data in heterogeneous multi-source energy consumption data;

[0072] Time series interpolation is performed on the missing timestamp data to obtain multi-source energy consumption heterogeneous data after missing values ​​are filled.

[0073] Remove the abnormal data points that deviate from the overall data distribution pattern in the multi-source energy consumption heterogeneous data after the missing value filling is completed, and obtain the cleaned multi-source energy consumption heterogeneous data;

[0074] Based on the statistical distribution parameters of the feature data in the cleaned multi-source energy consumption heterogeneous data, the feature data is converted into standardized data with the same unit of measurement to obtain the standardized energy consumption dataset of the building.

[0075] Specifically, firstly, all timestamp information in the heterogeneous energy consumption data from multiple sources is collected and arranged in chronological order. Then, a preset data collection period is determined, such as once per hour. After that, the interval between two adjacent timestamps is checked one by one. If the interval between two adjacent timestamps is greater than the preset collection period and there are no other timestamp records between these two timestamps, then the time point corresponding to this interval is the missing timestamp. In this way, all energy consumption data corresponding to missing timestamps are found, and the missing timestamp data is obtained.

[0076] Furthermore, for each missing timestamp data, first find the nearest timestamp with valid energy consumption data before the missing timestamp and its corresponding energy consumption data, then find the nearest timestamp with valid energy consumption data after the missing timestamp and its corresponding energy consumption data. Then, based on the time length between these two valid timestamps and the corresponding changes in energy consumption data, calculate the energy consumption data corresponding to the missing timestamp according to the time ratio. Fill the calculated energy consumption data into the position corresponding to the missing timestamp, and complete the filling operation for all missing timestamp data to obtain multi-source heterogeneous energy consumption data after missing value filling.

[0077] Furthermore, for each type of feature data in the multi-source energy consumption heterogeneous data after missing value imputation, the values ​​of all data in that type of feature data are first counted. Then, these values ​​are arranged in ascending order. The value at the 25th percentile of the arranged data is taken as the first quartile, and the value at the 75th percentile is taken as the third quartile. The interquartile range is obtained by subtracting the first quartile from the third quartile. Then, the lower limit and upper limit of outlier judgment are calculated. The lower limit is the first quartile minus 1.5 times the interquartile range, and the upper limit is the third quartile plus 1.5 times the interquartile range. Then, data points in that type of feature data whose values ​​are less than the lower limit or greater than the upper limit are marked as outlier data points. All outlier data points in all types of feature data are removed from the multi-source energy consumption heterogeneous data after missing value imputation, resulting in cleaned multi-source energy consumption heterogeneous data.

[0078] Furthermore, each type of feature data is extracted from the cleaned heterogeneous multi-source energy consumption data. First, the average value of all data in each type of feature data is calculated. Then, the average value of the squares of the differences between all data in each type of feature data and the average value of that type of data is calculated. Then, the square root of this average value is taken to obtain the standard deviation of that type of feature data. After that, for each data point in each type of feature data, the average value of that type of feature data is subtracted from the value of the data point. Then, the result is divided by the standard deviation of that type of feature data to obtain the standardized value of each data point. After converting all feature data into the corresponding standardized values, the standardized energy consumption dataset of the building is obtained.

[0079] In summary, the standardized multi-source energy consumption heterogeneous data processing workflow provides high-quality data support for building energy consumption prediction and optimization, and has significant beneficial effects.

[0080] In summary, using time series interpolation to fill missing values ​​can fully match the temporal correlation of energy consumption data, avoid data distortion caused by simple filling methods, remove abnormal data points that deviate from the overall distribution, effectively filter out interference information, ensure the consistency between the filled data and the actual energy consumption operation pattern of the building, and significantly improve the accuracy and reliability of the data.

[0081] In summary, by unifying the dimensions of the statistical distribution parameters of the cleaned data, the influence of differences in the collection dimensions and units of multi-source heterogeneous data can be eliminated. This ensures that each feature data participates in subsequent calculations under the same standard, clearing obstacles for accurately measuring the importance of features in the feature weight allocation stage and efficiently integrating multi-source information in the feature fusion stage. It also avoids subsequent analysis errors caused by dimensional bias, laying a solid data foundation for improving the accuracy of subsequent energy consumption prediction and generating targeted energy consumption optimization schemes.

[0082] The feature weight allocation module 102 is used to determine the importance weights of different feature dimensions in the same data source in the standardized energy consumption dataset, and the correlation weights between different data sources in the standardized energy consumption dataset.

[0083] In this embodiment of the invention, when the feature weight allocation module determines the importance weights of different feature dimensions within the same data source in the standardized energy consumption dataset, and the correlation weights between different data sources in the standardized energy consumption dataset, it is specifically used for:

[0084] Based on the correlation between the internal feature dimensions of the data source and the building energy consumption value in the standardized energy consumption dataset, an importance weight value is assigned to the internal feature dimensions of the data source.

[0085] Based on the data complementarity and information overlap between different data sources in the standardized energy consumption dataset, the correlation weight values ​​between the different data sources are determined.

[0086] Establish a table mapping feature dimension importance weights to data source relevance weights;

[0087] The corresponding relationship table is used as a weight identifier and associated with the standardized energy consumption dataset for storage to obtain the energy consumption weight allocation scheme of the building.

[0088] Specifically, a data source is selected from the standardized energy consumption dataset. All feature dimensions and their corresponding building energy consumption values ​​are extracted from this data source. For each feature dimension, the change in building energy consumption value is observed when its value changes. If a change in the value of a certain feature dimension leads to a large change in the building energy consumption value, then the correlation between the feature dimension and the building energy consumption value is high. If the change is small, the correlation is low. All feature dimensions are sorted according to their correlation, and the feature dimension with the highest correlation is assigned the largest importance weight value. The weight values ​​are then assigned in descending order of correlation until the weight assignment of all feature dimensions in the data source is completed. This operation is repeated to process all data sources, and finally the importance weight value of the feature dimensions within each data source is obtained.

[0089] Furthermore, two different data sources are selected from the standardized energy consumption dataset. First, the data complementarity is analyzed to see if the missing energy consumption information in the first data source can be found in the second data source. The more missing information found, the stronger the complementarity. Then, the information overlap is analyzed by counting the proportion of the number of identical energy consumption information in the two data sources to the total number of information in the two data sources. The lower the proportion, the lower the overlap. Based on a comprehensive judgment of complementarity and overlap, the two data sources with strong complementarity and low overlap have a higher correlation weight value, while the two data sources with weak complementarity and high overlap have a lower correlation weight value. The correlation weight values ​​between all different data sources are calculated in this way to obtain the correlation weight values ​​between different data sources.

[0090] Furthermore, first create a table. The rows of the table list all the data sources in the standardized energy consumption dataset. Each data source row is further subdivided into all the feature dimensions within that data source. After each feature dimension column, set a "Feature Dimension Importance Weight" column and fill in the corresponding feature dimension importance weight value that was previously assigned. Then, add a "Data Source Relevance Weight" column on the right side of the table. For each data source, fill in the determined relevance weight value between it and all other data sources in this column to ensure that the importance weight of each feature dimension can be mapped to the relevance weight of its data source and other data sources. After completing the table content filling, a correspondence table between feature dimension importance weight and data source relevance weight is obtained.

[0091] Furthermore, the established correspondence table between the importance weights of feature dimensions and the relevance weights of data sources is used as a weight identifier and associated with the standardized energy consumption dataset. Specifically, under each data source entry in the standardized energy consumption dataset, the importance weight values ​​of all feature dimensions corresponding to that data source are appended. At the same time, in the data source association area of ​​the dataset, the relevance weight values ​​between that data source and all other data sources are appended to ensure that the weight identifiers completely match the corresponding data in the dataset. After completing the association and storage, the building's energy consumption weight allocation scheme is obtained.

[0092] In summary, by assigning importance weights based on the correlation between the internal feature dimensions of the data source and the building energy consumption value in the standardized energy consumption dataset, we can accurately identify the feature dimensions that play a key driving role in energy consumption changes. This ensures that the weight allocation is highly matched with the actual energy consumption impact pattern, avoiding the problem of weakening core feature information and interfering with the analysis due to equal weighting, and strengthening the effective mining of the intrinsic value of the data.

[0093] In summary, determining relevance weights based on the complementarity and information overlap between different data sources scientifically quantifies the inherent connections between multi-source data. This approach fully leverages the synergistic gains of complementary data, improving the overall efficiency of multi-source information utilization, while avoiding redundancy caused by repeated calculations of overlapping information, thus ensuring the rationality of cross-source data integration. Furthermore, establishing and storing a mapping table between feature dimension importance weights and data source relevance weights provides a clear and traceable weighting basis for subsequent feature fusion stages, ensuring an orderly and accurate fusion process and laying a solid foundation for generating high-quality fused feature vectors.

[0094] The feature fusion module 103 is used to perform weighted fusion of the deep features extracted from the multi-source energy consumption heterogeneous data based on the importance weight and the relevance weight, to obtain the multi-source feature vector of the building, and to concatenate and reduce the dimensionality of the multi-source feature vector to obtain the fused feature vector of the building.

[0095] In this embodiment of the invention, when the feature fusion module performs weighted fusion of deep features extracted from the multi-source energy consumption heterogeneous data based on the importance weight and the relevance weight to obtain the multi-source feature vector of the building, it is specifically used for:

[0096] The deep features of the feature dimension within the same data source are weighted according to the importance weight to obtain the weighted feature representation of the building;

[0097] Based on the relevance weights, the weighted feature representations from the different data sources are fused across sources to obtain the preliminary fused features of the building.

[0098] The preliminary fusion features after dimension alignment are merged to form a high-dimensional fusion feature representation of the building.

[0099] Redundant information in the high-dimensional fusion feature representation is removed while key features are retained to obtain the multi-source feature vector of the building.

[0100] In this embodiment of the invention, when the feature fusion module performs dimensionality reduction by concatenating the multi-source feature vectors to obtain the fused feature vector of the building, it is specifically used for:

[0101] The multi-source feature vectors from different data sources are arranged and combined in a preset order to obtain a preliminary concatenated vector of the multi-source feature vectors;

[0102] The features after feature information condensation are mapped to a low-dimensional feature space to obtain the dimensionality-reduced feature representation of the building.

[0103] The reduced-dimensional feature representation is standardized to obtain the fused feature vector of the building.

[0104] Specifically, the same data source is selected, and the deep features corresponding to all feature dimensions within the data source are extracted. At the same time, the importance weight values ​​assigned to each feature dimension within the data source are retrieved. For the deep features of each feature dimension, the feature value is multiplied by the corresponding importance weight value to obtain the weighted deep features of that feature dimension. After the weighted operation of the deep features of all feature dimensions within the data source is completed, all weighted deep features are integrated into a unified feature set to obtain the weighted feature representation of the building.

[0105] Furthermore, the weighted feature representations corresponding to different data sources are obtained, and the determined correlation weight values ​​between each pair of different data sources are retrieved. For each group of different data sources to be fused, all feature values ​​in the weighted feature representation of each data source are multiplied by the correlation weight values ​​between that data source and other data sources. Then, all feature values ​​that have been weighted are added together according to the feature category to complete the fusion operation of the weighted feature representations of all different data sources and obtain the preliminary fusion features of the building.

[0106] Furthermore, all preliminary fusion features are first subjected to dimension alignment processing. The number and meaning of feature dimensions contained in each preliminary fusion feature are checked. Preliminary fusion features with different numbers of dimensions are supplemented with missing feature dimensions (the supplemented feature values ​​are determined according to the average level of the feature dimension). Preliminary fusion features with inconsistent dimension meanings are adjusted to have unified feature dimension meanings. After all preliminary fusion features have completed dimension alignment, all feature values ​​of these dimension-aligned preliminary fusion features are sequentially spliced ​​and integrated according to the feature dimension order to form a feature set containing all aligned feature dimensions, thus obtaining the high-dimensional fusion feature representation of the building.

[0107] Furthermore, all feature dimensions in the high-dimensional fusion feature representation are analyzed to identify feature dimensions whose values ​​are repeated and do not contribute additionally to the building energy consumption analysis (i.e., redundant information). For example, if the values ​​of two feature dimensions are completely consistent across all data points and both reflect the same energy consumption-related attribute, these redundant feature dimensions are deleted from the high-dimensional fusion feature representation. Only those feature dimensions that have a significant impact on changes in building energy consumption and can reflect key energy consumption patterns are retained. The values ​​of the retained key feature dimensions are arranged in order to obtain the building's multi-source feature vector.

[0108] Specifically, first, determine the preset order of the multi-source feature vectors, such as setting it according to the order of data source collection time. Then, collect all multi-source feature vectors from different data sources. Each multi-source feature vector maintains the original order of its internal feature elements. Next, according to the preset order, arrange all feature elements of the multi-source feature vector from the first data source in sequence. Then, arrange all feature elements of the multi-source feature vector from the second data source immediately after the feature elements from the first data source in sequence. Continue in this manner until all multi-source feature vectors from all data sources are processed. The final continuous feature element sequence is the initial spliced ​​vector of the multi-source feature vectors.

[0109] Furthermore, information analysis is performed on all feature elements in the initial spliced ​​vector to identify feature elements that express similar energy consumption information and group them into a group. Information integration is performed on the feature elements within each group. For example, the energy consumption change trend commonly reflected by the feature elements within each group is taken as the condensed information of that group. Then, according to the required low-dimensional feature space dimension, a corresponding low-dimensional position is assigned to each group of condensed information. The condensed information of all groups is mapped to each position in the low-dimensional feature space. The condensed information at each position constitutes a low-dimensional feature element. The set of all low-dimensional feature elements is the dimensionality-reduced feature representation of the building.

[0110] Furthermore, firstly, all low-dimensional feature elements are extracted from the dimensionality-reduced feature representation, and the average value of these low-dimensional feature elements is calculated. Then, the difference between each low-dimensional feature element and the average value is calculated, and the maximum fluctuation range among all differences is counted. Then, for each low-dimensional feature element, the average value is subtracted from the value of the element, and the result is adjusted to a uniform value range according to the maximum fluctuation range, such as between 0 and 1. Each adjusted low-dimensional feature element becomes a standardized feature element. All standardized feature elements are arranged in the order of the original dimensionality-reduced feature representation, and the resulting vector is the fused feature vector of the building.

[0111] In summary, providing high-quality feature inputs for building energy consumption prediction through scientific feature processing logic has significant value.

[0112] In summary, weighting deep features from the same data source based on importance weights can accurately strengthen features that significantly affect energy consumption and weaken the interference of secondary features, making single-source feature representations more consistent with the energy consumption change patterns. Furthermore, cross-source fusion using correlation weights can fully explore complementary information from different data sources while reducing redundancy caused by information overlap. This allows multi-source feature vectors to comprehensively and accurately cover energy consumption-related information, avoiding information imbalance or redundant accumulation during multi-source data integration.

[0113] In summary, concatenating multi-source feature vectors in a preset order ensures the orderliness of the feature structure; mapping to a low-dimensional space can condense key information, reduce data dimensionality and computational complexity, and avoid the curse of dimensionality; finally, standardization processing ensures that the fused feature vectors are on a uniform scale, clearing the scale difference obstacle for the subsequent energy consumption prediction module to extract local temporal pattern features and long-term temporal dependencies, reducing prediction errors caused by feature quality issues, and laying a solid foundation for improving the accuracy of energy consumption prediction.

[0114] The energy consumption prediction module 104 is used to extract the local time-series pattern features and long-term time-series dependencies of the fused feature vector, and to predict the energy consumption of the building based on the local time-series pattern features and the long-term time-series dependencies, so as to obtain the predicted building energy consumption value of the building in a specific time period in the future.

[0115] In this embodiment of the invention, when the energy consumption prediction module extracts the local temporal pattern features and long-term temporal dependencies of the fused feature vector, it is specifically used for:

[0116] The fused feature vector is divided into overlapping local time segments according to continuous temporal characteristics to obtain the local temporal window of the fused feature vector;

[0117] Capture the energy consumption fluctuation patterns and periodic characteristics within the short term in the local time window to obtain the local time pattern features of the fused feature vector;

[0118] Based on the feature evolution patterns between the local time windows, the long-term dependencies of the fused feature vector are captured.

[0119] In this embodiment of the invention, the energy consumption prediction module performs energy consumption prediction on the building based on the local time-series pattern features and the long-term time-series dependency to obtain the predicted energy consumption value of the building in a specific future time period, specifically for:

[0120] The local temporal pattern features are integrated with the long-term temporal dependencies to construct the temporal features of the building;

[0121] By analyzing the correspondence between historical energy consumption data and time-series characteristics, an energy consumption prediction model for the building is constructed.

[0122] The energy consumption of the time-series feature representation is predicted using the prediction model to generate the initial predicted building energy consumption value of the building.

[0123] By eliminating abnormal fluctuations in the initial building energy consumption forecast, the building energy consumption forecast for a specific future time period is obtained.

[0124] In this embodiment of the invention, the mathematical expression of the energy consumption prediction model is as follows:

[0125] ;

[0126] In the formula, For a moment The predicted energy consumption value, The weights of the historical time-series features in the historical energy consumption data are given. The size of the local timing window. The index for the time-series feature. For the first Weights of each time-series feature, for Quantization values ​​of temporal characteristics at any given time. The weighting of the impact of the rate of change in energy consumption in the historical energy consumption data. For time factors, For the local timing window, Let be the influence function of the rate of change of energy consumption. For the target energy consumption value, For time integration variables, The weight of the building in the current time series state. Let be the feature mapping function of the building in the current time series state. for A set of temporal and temporal characteristics.

[0127] Specifically, firstly, according to the chronological order of the fused feature vectors, all fused feature vectors are arranged sequentially to form a continuous time series. Then, the window length and overlap ratio of the local time segments are set. The window length is the number of continuous time series fused feature vectors contained, and the overlap ratio is the proportion of the number of fused feature vectors shared by two adjacent local time segments to the window length. Starting from the first fused feature vector in the continuous time series, fused feature vectors of the corresponding window length are extracted to form the first local time segment. Then, the extraction starting point is moved backward according to the overlap ratio. For example, when the window length is 10 fused feature vectors and the overlap ratio is 50%, the second local time segment is extracted starting from the 6th fused feature vector. This extraction operation is repeated until all fused feature vectors in the continuous time series are covered, thus obtaining the local time series window of the fused feature vectors.

[0128] Furthermore, for each local time series window, the value of each fused feature vector within the window is recorded one by one, and the changes in the value over time are observed. If the value repeatedly changes from low to high and then back to low in a short period of time, or if the value suddenly rises and then drops rapidly, the trend and magnitude of these value changes are recorded as short-term energy consumption fluctuation patterns. At the same time, it is observed whether there are similar values ​​that repeat at fixed time intervals within the window. For example, if the values ​​corresponding to every 3 fused feature vectors are similar and the intervals are consistent, this fixed interval repetition pattern is recorded as a periodic feature. By integrating the energy consumption fluctuation patterns and periodic features of each local time series window, the local time series pattern features of the fused feature vector are obtained.

[0129] Furthermore, all local time series windows are arranged sequentially according to time. The local time series pattern characteristics of two adjacent local time series windows are compared to see if the amplitude of the energy consumption fluctuation pattern changes from large to small or from small to large, and whether the fixed interval of the periodic feature is extended or shortened. Then, the local time series pattern characteristics of multiple local time series windows are compared to see if the energy consumption fluctuation pattern of the earlier window will cause similar fluctuations in the later window, and whether the periodic feature of the earlier window will affect the periodic pattern of the later window. The correlation of feature evolution between these different local time series windows is sorted out to capture the long-term dependency relationship of the fused feature vector.

[0130] Specifically, the local time-series pattern features corresponding to each local time-series window are first extracted, including short-term energy consumption fluctuation patterns and periodic features. Then, the long-term time-series dependencies between these local time-series windows are extracted, that is, the correlation of feature evolution patterns between different windows. According to the chronological order, the local time-series pattern features of each local time-series window are matched with their corresponding long-term time-series dependencies, so that the local features of each time period can be associated with the overall long-term evolution pattern. This is integrated to form a feature set that covers the complete time dimension, contains local details and long-term correlations, and constructs the temporal features of the building.

[0131] Furthermore, historical energy consumption data of the building over a period of time is collected. This data must include specific energy consumption values ​​and corresponding time information. At the same time, following the same method used to construct the time-series features of the current building, the time-series features of the historical energy consumption data corresponding to the time period are extracted. A one-to-one correspondence between the values ​​of the historical energy consumption data and the corresponding time-series features is established. For example, the historical time-series features of a certain week are associated with the actual energy consumption values ​​of that week. Based on all such correspondences, a set of rules is formed that can output corresponding energy consumption values ​​based on the input time-series features, thus constructing an energy consumption prediction model for the building.

[0132] Furthermore, a specific future time period for which energy consumption forecasting is required is determined, and the corresponding time-series feature representation is extracted. The structure and dimensions of this time-series feature representation are consistent with the historical time-series features used when constructing the energy consumption forecasting model. This time-series feature representation is input into the constructed energy consumption forecasting model. Based on the correspondence rules between the historical time-series features and historical energy consumption data stored internally, the model finds the historical correspondence that best matches the input time-series feature representation, outputs the corresponding energy consumption value, and generates the initial building energy consumption forecast value for the building.

[0133] Furthermore, the generated initial building energy consumption forecasts are arranged in chronological order, and the overall numerical trend is observed. Values ​​that suddenly increase or decrease significantly and are completely inconsistent with the trend of forecasts in adjacent time periods are identified as initial forecasts with abnormal fluctuations. By referring to the trend of normal initial forecasts adjacent to the abnormal fluctuation values, reasonable values ​​that conform to the overall trend are calculated. The abnormal fluctuation values ​​are replaced with these reasonable values. After eliminating all abnormal fluctuations, the building energy consumption forecast for a specific time period in the future is obtained.

[0134] Specifically, It originates from the process of analyzing the correspondence between historical energy consumption data and time-series characteristics, and is determined by assessing the overall impact of historical time-series characteristics on energy consumption prediction results; It originates from the local temporal window length set when dividing the fused feature vector into local time segments based on continuous temporal features; It is an index set to distinguish the position of each time series feature when extracting time series features at different times, and the index is taken sequentially according to the order of time series features within the local time series window; This originates from the process of assigning importance weights to time-series features, based on the first... The degree of correlation between each time-series characteristic and energy consumption value is determined; Originating from The quantization of temporal features at a given moment converts the local temporal pattern features and long-term dependencies at that moment into specific numerical values. It is derived from the process of analyzing historical energy consumption data and is determined by assessing the overall impact of the energy consumption change rate on the energy consumption prediction results. It is the target forecast time determined based on energy consumption forecast demand; It originates from the local temporal window range determined when dividing the local time segments of the fused feature vector; It originates from the process of analyzing the correlation between the rate of change of energy consumption and the actual energy consumption value in historical energy consumption data, and summarizes the functional form that can reflect this correlation; Data derived from collected historical energy consumption data of buildings; In calculating the integral of the rate of change of energy consumption, in order to cover from arrive The time range is set by the variable; Derived from the process of constructing temporal features, this involves evaluating the current temporal state for each time step. The extent of the impact of energy consumption prediction results is determined; This is derived from the process of analyzing the correspondence between the current time series characteristics and the energy consumption prediction value, and it summarizes the process of mapping the current time series characteristics into a function form that affects energy consumption prediction. Originating from The integration of temporal and temporal characteristics will The local temporal pattern characteristics and long-term dependencies at each moment are summarized to form the pattern.

[0135] Furthermore, this formula is used to calculate time. Energy consumption forecast The first part contributes predicted values ​​based on historical time-series features. The overall influence of historical time-series features on the predicted value is determined by w_i, while the importance of each historical time-series feature is determined by w_i. The first part consists of the specific quantitative results of each historical time-series feature. The combination of these three elements allows historical time-series features to participate in energy consumption prediction according to their importance. The second part contributes to the predicted value through the historical energy consumption change rate. The integral range determines the overall impact of the energy consumption change rate on the predicted value, and covers the local time series window. Within the time, The energy consumption change rate is converted into a form that affects the predicted value, thereby capturing the impact of the energy consumption change trend within a local time window on the predicted value; the third part contributes to the predicted value based on the current time series state. Determine the degree of influence of the current time series state on the predicted value. It is the set of all temporal features at the current moment. The current time-series feature set is transformed into a form that influences the predicted value, allowing the current time-series state to participate in energy consumption prediction. These three parts work together, integrating the influence of historical time-series features, historical energy consumption change rates, and the current time-series state, to ultimately obtain the predicted energy consumption value at time t. .

[0136] In summary, when extracting local temporal pattern features from the fused feature vector, dividing overlapping local time segments according to continuous time series can accurately capture short-term energy consumption fluctuations and periodic features, avoiding the omission of key short-term energy consumption changes (such as energy consumption fluctuations caused by short-term equipment start-ups and shutdowns and temporary load adjustments). At the same time, capturing long-term temporal dependencies based on the feature evolution patterns between local time windows can effectively grasp the long-term trend of energy consumption changes over time (such as the energy consumption differences between weekdays and weekends, and different seasons), achieving multi-dimensional and full-cycle coverage of energy consumption changes, and providing comprehensive temporal feature support for prediction.

[0137] In summary, when predicting energy consumption based on two types of features, the prediction model constructed by integrating local and long-term time-series features and combining them with historical energy consumption data can establish a more realistic energy consumption mapping relationship, avoiding the one-sidedness of prediction caused by relying solely on short-term or long-term features. Subsequent processing to eliminate abnormal fluctuations in the initial prediction value further filters out accidental interference factors, resulting in more accurate and stable energy consumption prediction values ​​for specific future time periods. This provides high-quality data for the energy consumption optimization module to match abnormal patterns and locate high-energy-consumption areas, ensuring the scientific and effective nature of subsequent energy consumption management decisions.

[0138] The energy consumption optimization module 105 is used to match the predicted building energy consumption value with the energy consumption anomaly patterns in the pre-acquired building operation characteristic library, and generate an energy consumption optimization report for the building based on the high energy consumption area corresponding to the energy consumption anomaly pattern.

[0139] In this embodiment of the invention, when the energy consumption optimization module performs the following steps: matching the predicted building energy consumption value with an abnormal energy consumption pattern in the pre-acquired building operation characteristic database, and generating an energy consumption optimization report for the building based on the high energy consumption area corresponding to the abnormal energy consumption pattern, the module is specifically used for:

[0140] The predicted building energy consumption is compared and analyzed with the standard energy consumption patterns in the pre-acquired building operation characteristic database to obtain the abnormal fluctuation range of the building's energy consumption.

[0141] Based on the abnormal energy consumption fluctuation range, the abnormal energy consumption patterns stored in the pre-acquired building operation characteristic library are matched to obtain the abnormal pattern type of the building and the corresponding high energy consumption area.

[0142] Based on the equipment operating status and energy consumption characteristics of the high-energy-consumption area, locate the specific equipment with low energy efficiency;

[0143] Based on the energy consumption data of the specific equipment and the energy-saving measures knowledge base corresponding to the abnormal mode types in the pre-acquired building operation characteristic library, an energy consumption optimization report for the building is generated.

[0144] Specifically, first, the standard energy consumption pattern corresponding to the current building is extracted from the pre-acquired building operation characteristic library. This pattern includes the normal energy consumption value range of the building under different operating conditions (such as weekdays, rest days, and different seasons) and different time periods. Then, the previously obtained building energy consumption prediction value is retrieved and classified according to the same operating conditions and time periods. The building energy consumption prediction value is compared point by point with the normal energy consumption value range of the corresponding category in the standard energy consumption pattern. When the building energy consumption prediction value exceeds the normal energy consumption value range of the corresponding category, the time interval and energy consumption value interval of the excess point are recorded. All consecutive excess time intervals and corresponding energy consumption value intervals are integrated to obtain the abnormal energy consumption fluctuation interval of the building.

[0145] Furthermore, all stored energy consumption anomaly patterns are retrieved from the pre-acquired building operation characteristic library. Each energy consumption anomaly pattern contains specific abnormal energy consumption fluctuation characteristics (such as fluctuation amplitude, duration, and time period), corresponding anomaly pattern type (such as equipment overload anomaly, insulation failure anomaly), and associated high-energy consumption area information. The abnormal energy consumption fluctuation characteristics (including fluctuation amplitude, duration, and specific time period) presented by the building's abnormal energy consumption fluctuation range are compared one by one with the abnormal energy consumption fluctuation characteristics of each energy consumption anomaly pattern in the library. The energy consumption anomaly pattern with a complete match is found. The anomaly pattern type marked by the matching pattern is the building's anomaly pattern type, and the high-energy consumption area marked by the matching pattern is the building's high-energy consumption area.

[0146] Furthermore, first list the names and numbers of all equipment in the high-energy-consuming areas of the building, then collect real-time operating status data (such as equipment start / stop status, operating load percentage, and key operating parameter values) and real-time energy consumption characteristic data (such as energy consumption per unit time and the trend of energy consumption with load changes). Analyze whether the operating status of each piece of equipment meets its rated operating requirements. If the operating status of an equipment deviates from the rated requirements (such as operating load exceeding the rated range or key parameters deviating from standard values), and its energy consumption characteristic data shows that its energy consumption per unit time is higher than the energy consumption value of similar equipment during normal operation, or its energy consumption shows an abnormal increase with load changes, then that equipment is a specific piece of equipment with low energy efficiency. All equipment that meets this condition is then screened out.

[0147] Furthermore, first, organize the energy consumption data of specific energy-inefficient equipment, including the average daily energy consumption, peak energy consumption, and energy consumption difference compared to similar normal equipment over the past period. Then, from the pre-acquired building operation characteristic database, find the energy-saving measure knowledge base corresponding to the building's abnormal mode type. This knowledge base contains energy-saving solutions for various inefficient equipment under this abnormal mode type (such as equipment parameter adjustment methods, component repair suggestions, and energy-saving equipment replacement models), implementation steps, and expected energy-saving rates. Associate the energy consumption data of specific equipment with the corresponding energy-saving solutions in the knowledge base. For each inefficient piece of equipment, explain its energy consumption problem, suitable energy-saving solutions, and expected energy reduction after implementation. Organize this content by equipment category, supplement the overall building energy-saving goals and implementation priorities, and generate a building energy consumption optimization report.

[0148] In summary, by comparing the predicted building energy consumption with the standard energy consumption patterns in the pre-acquired building operation characteristic library, the range of abnormal energy consumption fluctuations can be accurately identified. Based on the energy consumption anomaly patterns stored in the range matching library, the type of anomaly pattern and the corresponding high-energy consumption area can be quickly identified. This avoids the inefficiency of "blind investigation" in traditional energy management, achieves accurate positioning of energy consumption anomalies, makes energy consumption problem identification more targeted, and reduces the investment of ineffective manpower and time costs.

[0149] In summary, by combining the operating status and energy consumption characteristics of equipment in high-energy-consuming areas to pinpoint specific equipment with low energy efficiency, energy consumption anomalies can be refined from the "regional level" to the "equipment level," avoiding the predicament of only knowing the regional anomaly but not being able to identify the specific problematic equipment. At the same time, relying on a pre-stored knowledge base of energy-saving measures, optimization reports are generated by combining specific equipment energy consumption data. This ensures that the energy-saving recommendations in the reports are highly adapted to the actual situation of the equipment and the type of anomaly mode, rather than generalized universal recommendations. This makes energy consumption optimization measures easier to implement, effectively supporting the precise control and efficient reduction of building energy consumption, and ensuring the scientific and practical nature of energy consumption management decisions.

[0150] Reference Figure 2The diagram shown is a flowchart illustrating a building energy consumption prediction method based on multi-source data fusion according to an embodiment of the present invention. In this embodiment, the building energy consumption prediction method based on multi-source data fusion includes:

[0151] S1. Impute missing values ​​in the heterogeneous multi-source energy consumption data collected within the building, and unify the heterogeneous multi-source energy consumption data after missing value imputation to the same dimension to obtain the standardized energy consumption dataset of the building.

[0152] S2. Determine the importance weights of different feature dimensions in the same data source within the standardized energy consumption dataset, and the correlation weights between different data sources within the standardized energy consumption dataset;

[0153] S3. Based on the importance weight and the relevance weight, the deep features extracted from the multi-source energy consumption heterogeneous data are weighted and fused to obtain the multi-source feature vector of the building, and the multi-source feature vector is spliced ​​and dimensionality reduced to obtain the fused feature vector of the building.

[0154] S4. Extract the local temporal pattern features and long-term temporal dependencies of the fused feature vector, and perform energy consumption prediction on the building based on the local temporal pattern features and the long-term temporal dependencies to obtain the predicted building energy consumption value for a specific time period in the future.

[0155] S5. Based on the predicted building energy consumption value, match the abnormal energy consumption patterns in the pre-acquired building operation characteristic library, and generate an energy consumption optimization report for the building based on the high energy consumption area corresponding to the abnormal energy consumption pattern.

[0156] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0157] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0158] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A building energy consumption prediction system based on multi-source data fusion, characterized in that, The system includes a data processing module, a feature weight allocation module, a feature fusion module, an energy consumption prediction module, and an energy consumption optimization module, wherein: The data processing module is used to fill in missing values ​​in the heterogeneous multi-source energy consumption data collected within the building, and to unify the heterogeneous multi-source energy consumption data after missing value filling to the same dimension, thereby obtaining the standardized energy consumption dataset of the building. The feature weight allocation module is used to determine the importance weights of different feature dimensions in the same data source in the standardized energy consumption dataset, and the correlation weights between different data sources in the standardized energy consumption dataset. The feature fusion module is used to perform weighted fusion of the deep features extracted from the multi-source energy consumption heterogeneous data based on the importance weight and the relevance weight, to obtain the multi-source feature vector of the building, and to concatenate and reduce the dimensionality of the multi-source feature vector to obtain the fused feature vector of the building. The energy consumption prediction module is used to extract the local temporal pattern features and long-term temporal dependencies of the fused feature vector, and to predict the energy consumption of the building based on the local temporal pattern features and the long-term temporal dependencies, so as to obtain the predicted building energy consumption value for a specific time period in the future. The energy consumption optimization module is used to match the predicted building energy consumption value with the energy consumption anomaly patterns in the pre-acquired building operation characteristic library, and generate an energy consumption optimization report for the building based on the high energy consumption area corresponding to the energy consumption anomaly pattern.

2. The building energy consumption prediction system based on multi-source data fusion as described in claim 1, characterized in that, When the data processing module performs missing value imputation on the heterogeneous multi-source energy consumption data collected within the building, and unifies the imputed heterogeneous multi-source energy consumption data to the same dimension to obtain the standardized energy consumption dataset of the building, it is specifically used for: Identify missing timestamp data in heterogeneous multi-source energy consumption data; Time series interpolation is performed on the missing timestamp data to obtain multi-source energy consumption heterogeneous data after missing values ​​are filled. Remove the abnormal data points that deviate from the overall data distribution pattern in the multi-source energy consumption heterogeneous data after the missing value filling is completed, and obtain the cleaned multi-source energy consumption heterogeneous data; Based on the statistical distribution parameters of the feature data in the cleaned multi-source energy consumption heterogeneous data, the feature data is converted into standardized data with the same unit of measurement to obtain the standardized energy consumption dataset of the building.

3. The building energy consumption prediction system based on multi-source data fusion as described in claim 1, characterized in that, When determining the importance weights of different feature dimensions within the same data source in the standardized energy consumption dataset, and the correlation weights between different data sources in the standardized energy consumption dataset, the feature weight allocation module is specifically used for: Based on the correlation between the internal feature dimensions of the data source and the building energy consumption value in the standardized energy consumption dataset, an importance weight value is assigned to the internal feature dimensions of the data source. Based on the data complementarity and information overlap between different data sources in the standardized energy consumption dataset, the correlation weight values ​​between the different data sources are determined. Establish a table mapping feature dimension importance weights to data source relevance weights; The corresponding relationship table is used as a weight identifier and associated with the standardized energy consumption dataset for storage to obtain the energy consumption weight allocation scheme of the building.

4. The building energy consumption prediction system based on multi-source data fusion as described in claim 3, characterized in that, When the feature fusion module performs weighted fusion of deep features extracted from the multi-source energy consumption heterogeneous data based on the importance weight and the relevance weight to obtain the multi-source feature vector of the building, it is specifically used for: The deep features of the feature dimension within the same data source are weighted according to the importance weight to obtain the weighted feature representation of the building; Based on the relevance weights, the weighted feature representations from the different data sources are fused across sources to obtain the preliminary fused features of the building. The preliminary fusion features after dimension alignment are merged to form a high-dimensional fusion feature representation of the building. Redundant information in the high-dimensional fusion feature representation is removed while key features are retained to obtain the multi-source feature vector of the building.

5. The building energy consumption prediction system based on multi-source data fusion as described in claim 4, characterized in that, When the feature fusion module performs dimensionality reduction by concatenating and refining the multi-source feature vectors to obtain the fused feature vector of the building, it is specifically used for: The multi-source feature vectors from different data sources are arranged and combined in a preset order to obtain a preliminary concatenated vector of the multi-source feature vectors; The features after feature information condensation are mapped to a low-dimensional feature space to obtain the dimensionality-reduced feature representation of the building. The reduced-dimensional feature representation is standardized to obtain the fused feature vector of the building.

6. The building energy consumption prediction system based on multi-source data fusion as described in claim 1, characterized in that, When the energy consumption prediction module extracts the local temporal pattern features and long-term temporal dependencies of the fused feature vector, it is specifically used for: The fused feature vector is divided into overlapping local time segments according to continuous temporal characteristics to obtain the local temporal window of the fused feature vector; Capture the energy consumption fluctuation patterns and periodic characteristics within the short term in the local time window to obtain the local time pattern features of the fused feature vector; Based on the feature evolution patterns between the local time windows, the long-term dependencies of the fused feature vector are captured.

7. The building energy consumption prediction system based on multi-source data fusion as described in claim 6, characterized in that, The energy consumption prediction module performs energy consumption prediction on the building based on the local time-series pattern features and the long-term time-series dependency to obtain the predicted energy consumption value of the building for a specific future time period. Specifically, it is used for: The local temporal pattern features are integrated with the long-term temporal dependencies to construct the temporal features of the building; By analyzing the correspondence between historical energy consumption data and time-series characteristics, an energy consumption prediction model for the building is constructed. The energy consumption of the time-series feature representation is predicted using the prediction model to generate the initial predicted building energy consumption value of the building. By eliminating abnormal fluctuations in the initial building energy consumption forecast, the building energy consumption forecast for a specific future time period is obtained.

8. The building energy consumption prediction system based on multi-source data fusion as described in claim 7, characterized in that, The mathematical expression of the energy consumption prediction model is as follows: ； In the formula, For a moment The predicted energy consumption value, The weights of the historical time-series features in the historical energy consumption data are given. The size of the local timing window. The index of the time series feature, For the first Weights of each time-series feature, for Quantized values ​​of temporal characteristics at any given time. The weighting of the impact of the rate of change in energy consumption in the historical energy consumption data. For time factors, For the local timing window, Let be the influence function of the rate of change of energy consumption. For the target energy consumption value, For time integration variables, The weight of the building in the current time series state. Let be the feature mapping function of the building in the current time series state. for A set of temporal and temporal characteristics.

9. The building energy consumption prediction system based on multi-source data fusion as described in claim 1, characterized in that, When the energy consumption optimization module performs the following steps: matching the predicted building energy consumption value with an anomaly pattern in the pre-acquired building operation characteristic database, and generating an energy consumption optimization report for the building based on the high-energy-consumption area corresponding to the anomaly pattern, the module is specifically used for: The predicted building energy consumption is compared and analyzed with the standard energy consumption patterns in the pre-acquired building operation characteristic database to obtain the abnormal fluctuation range of the building's energy consumption. Based on the abnormal energy consumption fluctuation range, the abnormal energy consumption patterns stored in the pre-acquired building operation characteristic library are matched to obtain the abnormal pattern type of the building and the corresponding high energy consumption area. Based on the equipment operating status and energy consumption characteristics of the high-energy-consumption area, locate the specific equipment with low energy efficiency; Based on the energy consumption data of the specific equipment and the energy-saving measures knowledge base corresponding to the abnormal mode types in the pre-acquired building operation characteristic library, an energy consumption optimization report for the building is generated.

10. A building energy consumption prediction method based on multi-source data fusion, characterized in that, The method includes: S1. Impute missing values ​​in the heterogeneous multi-source energy consumption data collected within the building, and unify the heterogeneous multi-source energy consumption data after missing value imputation to the same dimension to obtain the standardized energy consumption dataset of the building. S2. Determine the importance weights of different feature dimensions in the same data source within the standardized energy consumption dataset, and the correlation weights between different data sources within the standardized energy consumption dataset; S3. Based on the importance weight and the relevance weight, the deep features extracted from the multi-source energy consumption heterogeneous data are weighted and fused to obtain the multi-source feature vector of the building, and the multi-source feature vector is spliced ​​and dimensionality reduced to obtain the fused feature vector of the building. S4. Extract the local temporal pattern features and long-term temporal dependencies of the fused feature vector, and perform energy consumption prediction on the building based on the local temporal pattern features and the long-term temporal dependencies to obtain the predicted building energy consumption value for a specific time period in the future. S5. Based on the predicted building energy consumption value, match the abnormal energy consumption patterns in the pre-acquired building operation characteristic library, and generate an energy consumption optimization report for the building based on the high energy consumption area corresponding to the abnormal energy consumption pattern.

Images