Data analysis method and system applied to building energy management

Abstract

The embodiment of the application provides a data analysis method and system applied to building energy management, effectively captures the space-time correlation and cross-domain dependent change among building energy monitoring data through graph self-attention processing and mutual correlation mining, generates a graph representation vector sequence of electric energy, water consumption, gas consumption or heat consumption, mines the dependent change relationship of adjacent data, forms dependent change relationship vectors of the first and second energy use cross-domains, and realizes extended analysis of energy flow intervals. Based on the dependent change relationship vectors, mode analysis is performed on a target energy entity, a mode analysis result reflecting the energy state is generated, and then building energy allocation and intelligent control are optimized, and the energy management efficiency and intelligent level are improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to a data analysis method and system for building energy management. Background Art

[0002] With the rapid development of society and the acceleration of urbanization, building energy consumption continues to grow and has become a significant component of global energy consumption. To effectively manage and reduce building energy consumption, building energy management systems (BEMS) are widely used in various buildings. These systems collect and analyze various energy data within the building, such as electricity, water, gas, and heat consumption, to monitor and optimize energy usage.

[0003] However, traditional building energy management methods often ignore the spatiotemporal correlation and cross-domain dependency between data when processing and analyzing these data. This limitation leads to a lack of comprehensiveness and accuracy in the formulation of energy management strategies, and is unable to fully realize the potential value of the data. Summary of the Invention

[0004] In order to at least overcome the above-mentioned deficiencies in the prior art, the purpose of this application is to provide a data analysis method and system for building energy management.

[0005] In a first aspect, the present application provides a data analysis method for building energy management, which is applied to a building energy management system. The method comprises:

[0006] Performing graph self-attention processing on each building energy monitoring data in the building energy monitoring data set to generate a graph representation vector sequence for each building energy monitoring data; each building energy monitoring data is one of electric energy monitoring data, water consumption monitoring data, gas consumption monitoring data, or heat consumption monitoring data, and each graph representation vector sequence for the building energy monitoring data is one of an electric energy graph representation vector, a water consumption graph representation vector, a gas consumption graph representation vector, or a heat consumption graph representation vector;

[0007] Performing mutual correlation mining on the graph representation vector sequences of the building energy monitoring data adjacent to each spatiotemporal domain, respectively, to generate a first energy usage cross-domain dependency relationship vector between the building energy monitoring data adjacent to each spatiotemporal domain; the first energy usage cross-domain dependency relationship vector represents the dependency relationship between the building energy monitoring data at an instantaneous energy flow interval, and the mutual correlation mining includes feature embedding analysis;

[0008] generating a second energy usage cross-domain dependency relationship vector based on each dependency relationship vector of the first energy usage cross-domain and a graph representation vector sequence of each building energy monitoring data; the second energy usage cross-domain dependency relationship vector represents a dependency relationship between the building energy monitoring data of an extended energy flow interval, and the second energy usage cross-domain is greater than the first energy usage cross-domain;

[0009] performing pattern analysis on a target energy entity in the building energy monitoring data according to the dependency change relationship vector of the second energy usage across domains, and generating a pattern analysis result reflecting an energy status pattern;

[0010] Building energy allocation is performed based on the pattern analysis results and the associated building energy allocation information, or the regulation system in the building is intelligently regulated based on the pattern analysis results.

[0011] In a second aspect, an embodiment of the present application further provides a building energy management system, which includes a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores a computer program, and the computer program is loaded and executed according to the processor to implement the data analysis method applied to building energy management according to the first aspect above.

[0012] Based on the technical solutions of any of the above aspects, this embodiment first performs graph self-attention processing on various energy monitoring data within a building to generate corresponding graph representation vector sequences, allowing the system to effectively capture the intrinsic connections and importance between various energy data. Secondly, the graph representation vector sequences of energy monitoring data adjacent to each other in each spatiotemporal domain are mined for mutual correlation, generating a first cross-domain dependency relationship vector for energy use. This not only characterizes the dependency relationship between building energy monitoring data at instantaneous energy flow intervals, but also further deepens the understanding of the association between data through feature embedding analysis, providing a more comprehensive and accurate analysis of energy use. Furthermore, based on the first cross-domain dependency relationship vectors for energy use and the graph representation vector sequences for each energy monitoring data, this method generates a second cross-domain dependency relationship vector for energy use, which can reflect the dependency relationship of energy use over a longer time span or a wider spatial range, providing a more macro perspective for building energy management. Ultimately, by analyzing the cross-domain dependency relationship vector of the second energy usage, the state pattern of the target energy entity in the building energy monitoring data can be accurately identified, and the corresponding pattern analysis results can be generated. This can not only be used directly to guide the allocation of building energy, but also serve as an important basis for intelligent control of the regulation system in the building. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required to be used in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0014] Figure 1 A flow chart of a data analysis method for building energy management provided in an embodiment of the present application;

[0015] Figure 2 A schematic block diagram of the functional structure of a building energy management system for implementing the above-mentioned data analysis method for building energy management provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] The following description is intended to enable one of ordinary skill in the art to implement and rely on this application, and is provided in the context of a specific application scenario and its requirements. It will be apparent to one of ordinary skill in the art that various modifications may be made to the disclosed embodiments, and that the general principles defined herein may be applied to other embodiments and application scenarios without departing from the principles and scope of this application. Therefore, this application is not limited to the described embodiments, but should be accorded the broadest scope consistent with the claims.

[0017] See also Figure 1 As shown, the present application provides a data analysis method applied to building energy management, comprising the following steps.

[0018] Step S110 : Performing graph self-attention processing on each building energy monitoring data in the building energy monitoring data set to generate a graph representation vector sequence for each building energy monitoring data. Each building energy monitoring data is one of electric energy monitoring data, water consumption monitoring data, gas consumption monitoring data, or heat consumption monitoring data, and each graph representation vector sequence for the building energy monitoring data is one of an electric energy graph representation vector, a water consumption graph representation vector, a gas consumption graph representation vector, or a heat consumption graph representation vector.

[0019] The building energy monitoring data set can be a collection of various energy usage data from multiple buildings, typically collected through various sensors and metering devices installed within the buildings. For example, real-time monitoring data on electricity, water, gas, and heat consumption for all buildings in a large residential complex or commercial complex can be aggregated to form a building energy monitoring data set. Building energy monitoring data refers to data on the usage or consumption of a specific energy source (such as electricity, water, gas, or heat) within a specific timeframe for a single building. For example, the electricity monitoring data for an office building may include the building's electricity consumption at different time periods throughout the day.

[0020] Graph self-attention processing is a method that uses the self-attention mechanism in graph neural networks to process graph-structured data. In graph data, nodes can represent entities (such as buildings) and edges can represent relationships between entities (such as energy flow relationships). The self-attention mechanism allows the model to focus on other nodes that are most relevant to a node when processing it. For example, when processing building energy monitoring data, the building energy management system will treat each building as a node and establish edges based on the energy flow relationships between buildings. Then, through graph self-attention processing, the building energy management system can analyze the degree of correlation between the energy use of each building and its adjacent buildings, and generate a graph representation vector sequence based on this.

[0021] The term "graph representation vector sequence" refers to the process of converting graph-structured data into a series of vectors (i.e., numerical vectors). This captures the graph's nodes, edges, and the relationships between them, facilitating subsequent mathematical calculations and pattern recognition. For example, a graph representation vector for each building's electricity, water, gas, or heat consumption could be a numerical vector that captures the statistical characteristics of that building's energy usage over a period of time (e.g., average, peak, and volatility).

[0022] Electricity, water, gas, and heat monitoring data refer to the usage or consumption of different energy types (electricity, water, natural gas, and heat) in a building. These data are typically collected and recorded using specialized metering devices (such as electricity, water, gas, and heat meters). For example, electricity monitoring data might include real-time current, voltage, and power readings for each area within a building; water monitoring data might include real-time flow readings for each water point within the building; gas monitoring data typically refers to natural gas consumption; and heat monitoring data might include energy consumption readings for heating or cooling systems.

[0023] The graph representation vectors for electricity, water, gas, and heat consumption refer to the graph representation vectors obtained by applying graph self-attention to the various types of energy monitoring data described above. Each type of energy monitoring data generates a corresponding graph representation vector, which captures the correlations and patterns in specific energy usage between buildings. For example, in the case of electricity, a graph representation vector for electricity could be a numerical vector that contains the correlations between a building's electricity usage and its neighboring buildings. This vector can reflect the changing trends in a building's electricity usage over different time periods and the degree of mutual influence between the building and its neighboring buildings. The same logic applies to the graph representation vectors for water, gas, and heat consumption.

[0024] Based on the above terminology, in this embodiment, the building energy management system receives a set of building energy monitoring data from multiple buildings. This set includes monitoring data on electricity, water, gas, and heat consumption for each building. The building energy management system first performs graph self-attention processing on this set of building energy monitoring data.

[0025] Specifically, the building energy management system treats each building as a node in a graph, and the energy flows between buildings as edges. The system then uses a graph self-attention mechanism to analyze the relationships between energy monitoring data from each node (i.e., building) and its adjacent nodes. This allows the system to generate a sequence of graph representation vectors for each building's electricity, water, gas, and heat consumption.

[0026] Step S120 performs correlation mining on the graph-represented vector sequences of the building energy monitoring data adjacent in each spatiotemporal domain to generate a first cross-domain dependency relationship vector for energy usage between the adjacent building energy monitoring data in each spatiotemporal domain. The first cross-domain dependency relationship vector for energy usage represents the dependency relationship between the building energy monitoring data at an instantaneous energy flow interval. The correlation mining includes feature embedding analysis.

[0027] In spatiotemporal data analysis, spatiotemporal adjacency refers to data points or entities that are adjacent or close in time and space. For example, in this embodiment, this typically refers to energy monitoring data from buildings that are geographically close and synchronized or close in time. For example, the energy usage data of two adjacent office buildings during the same time period can be considered spatiotemporally adjacent. When analyzing this data, the building energy management system will consider its temporal and spatial correlation.

[0028] Correlation mining is a data analysis technique that aims to discover statistical relationships or patterns between different variables or data points in a dataset. In the context of building energy monitoring, this involves identifying correlations or influences between energy usage across different buildings. For example, a building energy management system might analyze the energy usage data of two adjacent buildings and discover that when energy usage increases in one building, energy usage in the other building also tends to increase. This relationship is discovered through correlation mining.

[0029] The first cross-domain dependency relationship vector for energy usage is a numerical vector that describes an instantaneous or short-term dependency relationship between different energy types and different buildings. This relationship is derived by analyzing a graph representation vector sequence of energy monitoring data from adjacent buildings in the spatiotemporal domain. For example, suppose there are two adjacent buildings, A and B. The building energy management system analyzes their electricity and water consumption monitoring data and discovers that when building A's electricity usage suddenly increases, building B's water consumption also increases within a short period of time. This instantaneous dependency relationship can be encoded as a numerical vector (i.e., the first cross-domain dependency relationship vector for energy usage) for subsequent energy management and regulation decisions.

[0030] Feature embedding analysis is an in-depth data mining technique that aims to discover deep correlations or embedding relationships between different features (or characteristics) in a dataset. For example, in this embodiment, this can involve identifying the mutual influence and dependencies between different building energy usage characteristics. Taking electricity and water consumption as an example, the building energy management system can use feature embedding analysis to discover that in some cases, increased electricity use can lead to reduced water consumption (for example, because a more efficient cooling system reduces water use), or vice versa. This deep feature embedding relationship helps to more accurately understand the complexity and dynamics of building energy use.

[0031] Based on the above description, in this embodiment, after obtaining the graph representation vector sequence of each building's energy monitoring data, the building energy management system begins to mine the mutual correlation of these graph representation vector sequences. The purpose of this step is to discover the dependent change relationship of energy usage between buildings adjacent in time and space domains.

[0032] For example, a building energy management system can detect that when a building's electricity usage increases, the water consumption of adjacent buildings also increases accordingly. This relationship is represented by the dependency relationship between building energy monitoring data at instantaneous energy flow intervals. Using methods such as feature embedding analysis, the building energy management system quantifies this relationship as a cross-domain dependency relationship vector for primary energy usage.

[0033] Step S130: Based on the dependency relationship vectors of the first energy usage span and the graph representation vector sequence of the building energy monitoring data, a second energy usage span dependency relationship vector is generated. The second energy usage span dependency relationship vector represents the dependency relationship between the building energy monitoring data over an extended energy flow interval, and the second energy usage span is greater than the first energy usage span.

[0034] The second energy usage cross-domain dependency relationship vector is a numerical vector that describes the interdependence of energy usage between buildings over a larger range or longer time interval. Compared to the first energy usage cross-domain, the second energy usage cross-domain considers a wider range of temporal and spatial scales. For example, if the first energy usage cross-domain focuses on energy usage changes within minutes or hours between adjacent buildings, the second energy usage cross-domain can be extended to energy usage patterns within adjacent regions or cities over days, weeks, or even months. In this way, the second energy usage cross-domain dependency relationship vector can reveal more macroscopic and long-term energy usage trends and mutual influences.

[0035] Extending the energy flow interval refers to expanding the temporal or spatial scope of analysis to a larger interval when considering energy usage dependencies. Compared to instantaneous or short-term energy flow intervals, extended energy flow intervals allow for observation of energy usage dynamics over longer periods or wider areas. For example, in an instantaneous energy flow interval, one might focus on energy usage changes between two adjacent buildings over a period of a few minutes. However, in an extended energy flow interval, energy usage patterns across multiple buildings in the same area can be observed over days or weeks, and their interdependencies analyzed.

[0036] In one possible example, consider an energy monitoring dataset encompassing multiple buildings, each with detailed data on electricity, water, gas, and heat consumption. The goal of this embodiment is to analyze this data to understand the interdependencies in energy use across buildings and provide decision support for energy management and regulation.

[0037] First, we use graph self-attention to transform this energy monitoring data into a graph representation vector sequence. In this process, each building is considered a node, and the energy flows between buildings are considered edges. This self-attention mechanism captures the correlation between each building's energy usage and its neighbors.

[0038] Next, we perform correlation mining on the graph-represented vector sequences of energy monitoring data from adjacent buildings in the spatiotemporal domain. This generates first-order energy usage cross-domain dependency relationship vectors. These vectors describe the interdependence of energy usage between adjacent buildings over a short period of time. For example, we can find that when electricity usage in one building increases, gas consumption in its neighboring buildings also increases accordingly.

[0039] The first cross-domain dependency vector for energy use and the original graph representation vector sequence are then used to generate a second cross-domain dependency vector for energy use. This process considers a wider range of temporal and spatial scales to reveal broader, longer-term energy use trends and interactions. For example, it can be discovered that electricity usage in a building within a region shows an upward trend over several weeks, and this trend is associated with increased water consumption in other buildings within the region.

[0040] Ultimately, by analyzing the inter-domain dependency vectors of secondary energy use, we can more accurately understand the complexity and dynamics of energy use across buildings and provide effective decision support for energy management and regulation. For example, these analysis results can be used to optimize energy allocation plans, adjust operating parameters of building equipment, or implement energy-saving measures.

[0041] That is, in this embodiment, based on the first energy usage cross-domain dependency relationship vectors generated in step S120 and the graph representation vector sequence of each building energy monitoring data, the building energy management system further generates a second energy usage cross-domain dependency relationship vector. The purpose of this step is to discover the building energy usage dependency relationship over a longer time span or a wider spatial range.

[0042] For example, a building energy management system can detect that, during a specific time period, if electricity usage increases across buildings within a certain area, gas consumption in certain buildings outside that area also increases accordingly. This relationship represents a dependent variation relationship between building energy monitoring data over an extended energy flow interval. The building energy management system quantifies this relationship as a second energy usage cross-domain dependent variation relationship vector.

[0043] Step S140 , performing pattern analysis on the target energy entity in the building energy monitoring data according to the second energy usage cross-domain dependency relationship vector, and generating a pattern analysis result reflecting an energy status pattern.

[0044] A target energy entity refers to a specific energy-using object or system that requires special attention or management. This could be a piece of equipment within a building, a specific area, or the energy usage of the entire building. For example, the air conditioning system in an office building could be considered a target energy entity. When monitoring and analyzing the building's energy usage, the building energy management system could pay special attention to the energy consumption of the air conditioning system and optimize and regulate it as needed.

[0045] Pattern analysis aims to identify recurring patterns or regularities with specific structure and meaning within complex data sets. In building energy monitoring, pattern analysis is used to discover specific patterns in energy usage data. These patterns can reveal energy efficiency, waste, or potential optimization opportunities. For example, a building energy management system might analyze a building's energy usage data and discover a significant peak in energy usage between 3:00 PM and 5:00 PM each day. This recurring peak is a pattern of energy usage, reflecting the building's workload or equipment usage during this time period.

[0046] Energy state patterns refer to specific states or behavioral patterns in a building's energy use. These patterns can be cyclical, trend-based, or sudden, reflecting a building's energy demand and usage at different times and under different conditions. For example, a commercial building's energy usage can be significantly lower on weekends than on weekdays because business activity decreases during the weekend. This cyclical variation is an energy state pattern.

[0047] Therefore, in this embodiment, the building energy management system performs pattern analysis on the target energy entity in the building energy monitoring data based on the cross-domain dependency relationship vector of the second energy usage. The purpose of this step is to identify and understand the patterns and laws of building energy usage.

[0048] For example, a building energy management system might discover that a building's electricity usage is consistently above average during a specific time period, potentially indicating issues such as energy waste or equipment failure. The building energy management system identifies this energy status pattern as an abnormality and generates corresponding pattern analysis results.

[0049] Step S150 : performing building energy allocation based on the pattern analysis result and the associated building energy allocation information, or performing intelligent regulation on the regulation system in the building according to the pattern analysis result.

[0050] In this embodiment, finally, the building energy management system performs building energy allocation based on the pattern analysis results and the associated building energy allocation information, or performs intelligent regulation on the regulation system in the building according to the pattern analysis results.

[0051] For example, if a building's energy management system identifies energy waste in a particular building, it can optimize the building's energy usage plan based on the building's energy allocation information to reduce waste. Alternatively, if the building's energy management system identifies equipment failure in a particular building, resulting in abnormal energy usage, it can automatically adjust the operating parameters of the building's temperature control system, lighting system, and other equipment to reduce the impact of the failure on energy usage, and notify maintenance personnel to address the issue promptly.

[0052] Building energy allocation information refers to information about how energy is allocated or allocated among different building energy users. This information can include information about energy sources, allocation ratios, and allocation times. It is used to ensure that energy supply within the building meets demand while optimizing energy efficiency. For example, if a building has multiple areas requiring power, the building energy management system can dynamically adjust the proportion of energy allocated to each area based on each area's real-time power demand and efficiency. These adjustments are made based on building energy allocation information.

[0053] Smart control is an automated control system approach that uses advanced algorithms and sensor technology to monitor and adjust energy-using devices and systems within a building in real time. The goal of smart control is to optimize energy efficiency, improve comfort, and reduce energy waste.

[0054] In an example of this step, in a smart office building, an intelligent control system can automatically adjust the air conditioning system's operating mode and set temperature based on readings from indoor temperature and humidity sensors to ensure a comfortable office environment with minimal energy consumption. This automatic adjustment is a typical application of intelligent control. Consider a smart office building equipped with various sensors and metering devices to monitor energy usage, such as electricity, water, gas, and heat, in real time. The building energy management system collects this energy monitoring data and converts it into a graph representation vector sequence using methods such as graph self-attention processing.

[0055] Next, the building energy management system (BEM) mines the interdependencies of the graph-representing vector sequence to generate a first energy usage cross-domain dependency relationship vector and a second energy usage cross-domain dependency relationship vector. These dependency relationship vectors reveal the interdependencies and long-term trends between different energy usage entities within the building. The BEM then performs pattern analysis on the target energy entity (such as the air conditioning system) based on the second energy usage cross-domain dependency relationship vector. By analyzing historical and real-time data, the BEM discovers that the air conditioning system's energy consumption is high during a certain period in the summer and that there is room for optimization. Based on this pattern analysis result and the associated building energy allocation information, the BEM can develop an optimization strategy to reduce the air conditioning system's energy consumption. For example, the BEM can increase the set temperature while ensuring comfort, optimize the operating hours of the air conditioning equipment, or introduce more efficient cooling technology.

[0056] Finally, the building energy management system (BEM) intelligently regulates the building's conditioning systems, such as air conditioning, based on optimization strategies. By monitoring and automatically adjusting equipment operating parameters in real time, the BEM ensures that the air conditioning system meets comfort requirements while minimizing energy consumption. This intelligent regulation not only improves energy efficiency but also helps reduce operating costs and minimize environmental impact.

[0057] Based on the above steps, this embodiment first performs graph self-attention processing on various energy monitoring data within the building, generating corresponding graph representation vector sequences, allowing the system to effectively capture the intrinsic connections and importance between various energy data. Secondly, the graph representation vector sequences of energy monitoring data adjacent to each other in each spatiotemporal domain are mined for mutual correlation, generating a first cross-domain dependency relationship vector for energy use. This not only characterizes the dependency relationship between building energy monitoring data within instantaneous energy flow intervals, but also further deepens the understanding of the association between data through feature embedding analysis, providing a more comprehensive and accurate analysis of energy use. Furthermore, based on the first cross-domain dependency relationship vectors for energy use and the graph representation vector sequences for each energy monitoring data, this method generates a second cross-domain dependency relationship vector for energy use, which can reflect the dependency relationship of energy use over a longer time span or a wider spatial range, providing a more macro perspective for building energy management. Ultimately, by analyzing the cross-domain dependency relationship vector of the second energy usage, the state pattern of the target energy entity in the building energy monitoring data can be accurately identified, and the corresponding pattern analysis results can be generated. This can not only be used directly to guide the allocation of building energy, but also serve as an important basis for intelligent control of the regulation system in the building.

[0058] In a possible implementation, before step S110, the method further includes:

[0059] Step S101: Obtain an energy data stream obtained by energy monitoring a target energy entity within a preset monitoring period, parse the energy data stream to obtain a sequence of energy data points, sample data from the sequence of energy data points, and generate the building energy monitoring data set. Alternatively, obtain the building energy monitoring data set obtained by sampling energy data from the target energy entity within the preset monitoring period.

[0060] For example, during a specific monitoring period of a day, such as from 9 a.m. to 5 p.m., the building energy management system obtains energy usage data of multiple target energy entities in real time through the data interface connected to each building energy monitoring system. These energy usage data streams include real-time readings of various energy types such as electricity, water consumption, and gas consumption. Each data stream consists of a timestamp and the corresponding energy usage.

[0061] After receiving these energy data streams, the building energy management system first cleans and formats the data to ensure consistency and accuracy. It then parses these data streams into a series of energy data points in chronological order, each containing specific energy usage and time information. For example, a building energy management system might parse an electrical energy data stream into a sequence of multiple data points, each representing electrical energy usage at a specific point in time.

[0062] Because energy data point sequences can be very large and contain a large number of data points, building energy management systems need to sample these energy data point sequences to reduce the data volume and retain key information. Building energy management systems can adopt different sampling strategies, such as equal-interval sampling and threshold-based sampling. Taking equal-interval sampling as an example, the building energy management system can select a data point from the energy data point sequence at a certain time interval (such as every minute or every hour) as the sampling result.

[0063] After data sampling, the building energy management system obtains a simplified building energy monitoring data set containing key energy usage information. This building energy monitoring data set is a structured data set, in which each data entry contains key fields such as sampling time, energy type, and energy usage. This data set provides the basis for subsequent graph self-attention processing and correlation mining.

[0064] In addition to acquiring and processing energy data streams in real time through the aforementioned steps, building energy management systems can also directly obtain pre-processed building energy monitoring data sets from the building energy monitoring system. This pre-processing process can include data cleaning, formatting, sampling, and other steps to ensure data quality and usability. Directly acquiring this pre-processed data set can save the building energy management system processing time and accelerate subsequent data analysis.

[0065] In a possible implementation, step S120 may include:

[0066] A graph representation vector sequence of the first building energy monitoring data in the building energy monitoring data adjacent to each spatiotemporal domain is used as a reference vector, and a graph representation vector sequence of the second building energy monitoring data in the building energy monitoring data adjacent to each spatiotemporal domain is used as a tracking vector. The reference vector and the tracking vector are subjected to focused tracking based on vector dependency, and a first energy usage cross-domain dependency relationship vector is generated between the building energy monitoring data adjacent to each spatiotemporal domain.

[0067] In this embodiment, the building energy management system begins mining the correlations among graph-represented vector sequences of energy monitoring data from adjacent buildings across time and space. Specifically, the building energy management system selects the graph-represented vector sequence of energy monitoring data from a single building (referred to as the first building) as a baseline vector. This baseline vector represents the dynamic pattern of energy usage in the first building.

[0068] Then, the building energy management system selects a graph representation vector sequence of energy monitoring data of another building (referred to as the second building) adjacent to the first building in the temporal and spatial domain as a tracking vector, which represents the dynamic pattern of energy use of the second building.

[0069] The building energy management system then performs focused tracking of the reference vector and the tracking vector based on vector dependency. This means that the building energy management system analyzes the dynamic relationship between the two vector sequences, especially the dependency relationship between them, which can manifest as a time delay, amplitude scaling, or other forms of complex association.

[0070] Through focused tracking analysis, the building energy management system can identify the interdependence of energy use between the first building and the second building. This interdependence may indicate that when the energy use of the first building increases, the energy use of the second building will also increase accordingly, or there may be other similar correlation patterns.

[0071] Finally, the building energy management system encodes this interdependence into a vector, called the first energy usage cross-domain dependency relationship vector. This dependency relationship vector captures the dynamic correlation of energy usage between the first building and the second building, providing important information for subsequent pattern analysis and intelligent regulation.

[0072] Through this process, the building energy management system can systematically analyze the interrelationships in energy usage among various buildings in the urban energy monitoring system, thereby providing strong support for energy management and optimization.

[0073] In a possible implementation, the reference vector includes a vector index and vector content.

[0074] In the processing of building energy management systems, the reference vector is not just a string of numbers or values, but also contains two important pieces of information: vector index and vector content.

[0075] Vector index: This can be thought of as a timestamp or location marker, indicating the specific location or time point of the vector content in the spatiotemporal domain. For example, if the baseline vector represents the energy usage of a building over a week, the vector index might be the time stamp for each day.

[0076] Vector content: This is the core part of the baseline vector and contains the actual energy usage data or related information. In the example above, the vector content may be the energy usage of the building on each day.

[0077] When processing this data, the building energy management system considers both the vector index and the vector content to ensure that the dynamic changes and interrelationships in energy use can be accurately captured in subsequent analysis.

[0078] The performing focused tracking on the reference vector and the tracking vector based on vector dependency to generate a first energy usage cross-domain dependency relationship vector between the building energy monitoring data adjacent in each spatiotemporal domain includes:

[0079] Step S121 , performing focus tracking based on vector-dependent changes on the vector index and the tracking vector to generate a vector focus tracking result.

[0080] In this embodiment, the building energy management system starts to perform focused tracking analysis on the vector index of the reference vector and the tracking vector, which means that the building energy management system will closely monitor the temporal or spatial variation trends of the two vectors and look for the dependency relationship between them.

[0081] Specifically, the building energy management system can compare the vector index of the baseline vector and the values ​​of the tracking vector at the same or similar time points to analyze whether their changing trends are consistent or regular. For example, if the baseline vector's energy usage suddenly increases at a certain point in time, the building energy management system will check whether the tracking vector shows a similar increase at the same or later time points.

[0082] Through this focus tracking analysis, the building energy management system can generate a vector focus tracking result that captures the dynamic dependency between the reference vector and the tracking vector. This dynamic dependency may manifest as a time delay, amplitude scaling, or other forms of complex correlation.

[0083] Step S122: Performing a logical relationship quantitative analysis on the vector focus tracking result to generate a logical relationship focus quantization coefficient.

[0084] After obtaining the vector focus tracking results, the building energy management system will further perform a logical relationship quantitative analysis on the results. This means that the building energy management system will try to convert this dynamic dependency relationship into a specific value or coefficient in order to more accurately describe this relationship.

[0085] For example, a building energy management system can calculate the correlation coefficient, covariance, or other statistics between the baseline vector and the tracking vector to quantify the degree of association between them. This quantitative coefficient (called the logical relationship focus quantification coefficient) can provide an objective metric for evaluating the correlation between the two vectors.

[0086] Step S123 , focusing on the quantization coefficient and the vector content according to the logical relationship, and generating a first cross-domain dependency change relationship vector of the energy usage between the building energy monitoring data adjacent in each spatiotemporal domain.

[0087] Finally, the building energy management system will combine the logical relationship focus quantification coefficient and the vector content of the benchmark vector to generate a first energy usage cross-domain dependency relationship vector. This dependency relationship vector not only contains the dynamic dependency relationship information between the benchmark vector and the tracking vector (reflected by the logical relationship focus quantification coefficient), but also contains the actual energy usage data (reflected by the vector content).

[0088] This first cross-domain dependency relationship vector of energy usage can be regarded as a high-level feature representation that integrates spatiotemporal information and the intercorrelations of energy usage, and can provide strong data support for subsequent energy management model analysis, energy efficiency evaluation, and intelligent energy regulation.

[0089] In a possible implementation, step S121 may include:

[0090] Step S1211 defines the initial window size of a sliding window used for focus tracking and sets a sensitivity threshold for focus tracking. The sliding window is used to slide on the vector index of the reference vector and compare it with the tracking vector. The sensitivity threshold is used to determine the similarity or difference between the tracking vector and the reference vector within any sliding window.

[0091] In this embodiment, when processing the focus tracking of the reference vector and the tracking vector, the building energy management system first defines an initial window size of a sliding window, for example, setting it to include data of 5 consecutive time points in the reference vector. This sliding window will slide on the vector index of the reference vector and compare it with the tracking vector.

[0092] The building energy management system also sets a sensitivity threshold, such as 0.8, to determine the similarity or difference between the tracking vector and the reference vector within any sliding window. If the similarity metric between two vectors within a sliding window exceeds this threshold, they are considered to have a significant interdependence relationship within that window.

[0093] In step S1212, a sliding window with an adjustable window size is constructed based on the initial window size, taking the starting point of the reference vector as the starting point. The sliding window is dynamically resized as the focus tracking proceeds to adapt to the dependent changes between the reference vector and the tracking vector.

[0094] In this embodiment, the building energy management system uses the starting point of the reference vector as the starting point and constructs a sliding window based on the initial window size. The size of this sliding window is not fixed, but can be dynamically adjusted as the focus tracking progresses.

[0095] For example, if the building energy management system finds that the tracking vector has a significant correlation with the baseline vector within a sliding window, the sliding window size can be reduced to more accurately capture this change. Conversely, if the building energy management system does not find a significant correlation within multiple consecutive sliding windows, the sliding window size can be increased to cover more data points for analysis.

[0096] Step S1213 : calculating, within each sliding window, a similarity measurement parameter between the subsequence of the reference vector and the corresponding portion of the tracking vector, wherein the similarity measurement parameter is used to quantify the degree of matching between the reference vector and the tracking vector within the local window.

[0097] In this embodiment, within each sliding window, the building energy management system calculates a similarity metric between a subsequence of the reference vector and a corresponding portion of the tracking vector. This similarity metric can be a correlation coefficient, cosine similarity, or other metrics that measure the similarity between two sequences.

[0098] For example, a building energy management system can use cosine similarity to calculate the similarity between two vectors within a sliding window. Cosine similarity ranges from -1 to 1, with values ​​closer to 1 indicating more consistent directions between the two vectors, meaning their patterns of change within the window are more similar.

[0099] In step S1214, when the similarity metric exceeds a preset sensitivity threshold, the sliding window is marked as a dependent change point, and the size and position of the sliding window are dynamically adjusted based on the identified dependent change point. A dependent change point indicates that the tracking vector has a significant change correlation with the reference vector within the sliding window.

[0100] Step S1215: If no dependent change greater than the preset change floating value is detected in multiple consecutive sliding windows, the window size of the sliding window is increased based on the first preset window control parameter; if a dependent change greater than the preset change floating value is detected in multiple consecutive sliding windows, the window size of the sliding window is reduced based on the second preset window control parameter.

[0101] When the similarity metric calculated by the building energy management system exceeds a preset sensitivity threshold, the sliding window is marked as a dependent change point, which means that the tracking vector has a significant change correlation with the reference vector within the sliding window.

[0102] Based on the identified dependent change points, the building energy management system dynamically adjusts the size and position of the sliding window. For example, if significant change correlations are detected within multiple consecutive sliding windows, the building energy management system can gradually reduce the size of the sliding window to more accurately locate these change points.

[0103] If no significant change correlation is detected within multiple consecutive sliding windows, the building energy management system will increase the sliding window size based on the first preset window control parameter to cover more data points for analysis. Conversely, if a significant change correlation is detected within multiple consecutive sliding windows, the building energy management system will reduce the sliding window size based on the second preset window control parameter.

[0104] Step S1216 , gradually moving the sliding window along the vector index of the reference vector, and repeating the above steps until the entire reference vector is traversed, and recording all identified dependent change points and corresponding window information.

[0105] The building energy management system gradually moves the sliding window along the vector index of the reference vector and repeats the above steps to perform focus tracking analysis. During this process, the building energy management system records all identified dependent change points and the corresponding window information (including window size, position, and corresponding similarity measurement parameters).

[0106] When the building energy management system traverses the entire reference vector, it will obtain a list of key change points and change patterns of the tracking vector relative to the reference vector in the time and space domain. This list is sorted according to the position and time order of the change point on the reference vector.

[0107] Step S1217, sorting all identified dependent change points according to the position and time sequence of the dependent change points on the baseline vector, and generating a vector focus tracking result based on the sorted dependent change point list and corresponding window information, wherein the vector focus tracking result includes the key change points and change patterns of the tracking vector relative to the baseline vector in the spatiotemporal domain.

[0108] Finally, the building energy management system generates a vector focus tracking result based on the sorted list of dependent change points and the corresponding window information. This result is a high-level feature representation that integrates spatiotemporal information and the interrelationships between energy use, providing strong data support for subsequent energy management model analysis, energy efficiency assessment, and intelligent energy regulation.

[0109] For example, a building energy management system can visualize vector focus tracking results as a chart or report, showing the relationship between the tracking vector and the baseline vector over different time periods and spatial locations. Such visualizations can help energy managers more intuitively understand the dynamic changes and interrelationships in building energy use, thereby developing more effective energy management strategies.

[0110] In a possible implementation manner, each dependency relationship vector of the first energy usage across domains includes a dependency relationship vector of each energy usage node in the first energy usage across domains.

[0111] Step S130 may include:

[0112] Step S131 : determining a potential situation vector based on a graph representation vector sequence corresponding to a first energy usage node in the building energy monitoring data set.

[0113] In this embodiment, the building energy management system first processes the graph representation vector sequence corresponding to the first energy usage node in the building energy monitoring data set. This graph representation vector sequence contains the energy usage data, device status information, environmental factors, and other information for this energy usage node at different time points. By analyzing this graph representation vector sequence, the building energy management system can identify potential situational information such as the node's energy usage pattern and changing trends.

[0114] For example, building energy management systems use deep learning, pattern recognition, and other algorithms to process these graph-represented vector sequences, extract key features, and form a latent state vector. This latent state vector abstractly represents the energy usage of the primary energy-using node, capturing important information and features about that node in both time and space.

[0115] Step S132: Integrate the dependency change relationship vector of the second energy usage node across the first energy usage domain with the potential situation vector to generate an integrated vector. The first energy usage node is an energy usage node preceding the second energy usage node.

[0116] Next, the building energy management system will integrate the dependency relationship vector of the second energy usage node across the first energy usage domain with the potential situation vector. This dependency relationship vector describes the dynamic dependency relationship between the second energy usage node and the benchmark vector (or other related vectors).

[0117] The integration process may involve weighted summation, feature concatenation, and fusion of the two vectors, depending on the specific application scenario and data characteristics. Through integration, the building energy management system obtains an integrated vector that not only contains the potential status information of the first energy-using node but also reflects the dependent change relationship of the second energy-using node.

[0118] Step S133: determining flow migration weight information according to the integrated vector and the first conversion array.

[0119] The building energy management system determines flow migration weight information based on the integrated vector and a first conversion array. The first conversion array may be a predefined matrix or function that maps the integrated vector to a flow migration weight space. The flow migration weight information reflects the weight of the impact of the potential situation vector of the previous energy-using node (i.e., the first energy-using node) on the potential situation vector of the current energy-using node (i.e., the second energy-using node).

[0120] This transfer weight information can help the building energy management system understand the transmission and transfer patterns of energy use between different nodes, as well as the interdependencies between nodes. For example, if the energy use of the previous node has a significant impact on the energy use of the current node, the corresponding transfer weight will be larger.

[0121] Step S134: Determine flow transition information based on the integrated vector and the second conversion array, and determine a first reference potential situation vector based on the potential situation vector, the flow transition information, and the third conversion array. The flow transition weight information reflects the weight of influence of the potential situation vector of the previous energy-using node on the potential situation vector of the current energy-using node, and the flow transition information reflects the transition information between the potential situation vector of the previous energy-using node and the dependency relationship vector of the current energy-using node.

[0122] The building energy management system then determines flow and integration information based on the integrated vector and the second transformation array. This flow and integration information reflects the integration of the potential state vector of the previous energy usage node and the dependent change relationship vector of the current energy usage node. This integration may manifest as the fusion, complementarity, or mutual enhancement of the two vectors in the feature space.

[0123] At the same time, the building energy management system also determines a first reference potential state vector based on the potential state vector, the flow and integration information, and the third conversion array. This first reference potential state vector is a comprehensive prediction or estimate of the current energy use node after considering the potential state and dependent change relationship of the previous node.

[0124] Step S135 : generating a second energy usage cross-domain dependency change relationship vector based on the first reference potential situation vector and the flow migration weight information.

[0125] Finally, the building energy management system generates a second cross-domain dependency relationship vector for energy usage based on the first reference potential situation vector and the flow migration weight information. This dependency relationship vector not only contains the current energy usage node's own information and characteristics (reflected by the first reference potential situation vector), but also incorporates the impact of the previous node's potential situation on the current node (reflected by the flow migration weight information).

[0126] This second cross-domain dependency relationship vector of energy usage can be regarded as a high-level feature representation that integrates spatiotemporal information, inter-node dependencies, and dynamic changes in energy usage. It can provide strong data support and decision-making basis for subsequent energy usage prediction, optimization, and control.

[0127] In a possible implementation, step S140 may include:

[0128] Step S141 : determining a characteristic distance between the cross-domain dependency relationship vector of the second energy usage and each preset pattern feature vector in the preset pattern feature vector sequence.

[0129] In this embodiment, the building energy management system first processes the second energy usage cross-domain dependency vector, a high-level feature representation that captures the dynamics and interdependencies of energy usage. To perform pattern analysis on this dependency vector, the building energy management system compares it with each of the preset pattern feature vectors in the sequence of preset pattern feature vectors.

[0130] A preset pattern feature vector sequence is a predefined set of vectors, each of which represents a known energy state or usage pattern. These patterns may be derived from historical data, expert knowledge, or other sources and have been verified to be effective and representative in real-world applications.

[0131] The building energy management system calculates the characteristic distance between the second energy usage cross-domain dependency relationship vector and each preset pattern feature vector in the preset pattern feature vector sequence. This characteristic distance can be a Euclidean distance, cosine similarity, Mahalanobis distance, or other distance, depending on the characteristics of the vector and the requirements of the application scenario. The size of the characteristic distance reflects the similarity or difference between the two vectors in the feature space.

[0132] Step S142 : selecting a target characteristic distance that meets a preset distance requirement from the obtained characteristic distances.

[0133] After calculating all characteristic distances, the building energy management system selects target characteristic distances that meet the requirements based on preset distance requirements. This preset distance requirement can be a threshold, an interval, or a sorting rule, depending on the needs of the application scenario and the characteristics of the analysis target.

[0134] For example, a building energy management system may select the first few preset pattern feature vectors with the smallest feature distance as the vectors corresponding to the target feature distance, because these vectors are closest to the second energy usage cross-domain dependency relationship vector in the feature space and are most likely to represent the same or similar energy state patterns.

[0135] Step S143 , taking the state pattern portrait of the preset pattern feature vector corresponding to the target feature distance as the pattern analysis result.

[0136] Finally, the building energy management system uses the state pattern profile of the preset pattern feature vector corresponding to the target feature distance as the pattern analysis result. The state pattern profile is a visual or symbolic representation of the energy state pattern represented by the preset pattern feature vector, helping users or managers more intuitively understand and identify the current energy usage status.

[0137] For example, if the preset pattern feature vector corresponding to the target feature distance represents a "peak power consumption pattern," the pattern analysis result generated by the building energy management system may be a report or chart containing information such as the peak power consumption period, peak power consumption, and duration. This result can provide a basis and support for subsequent energy management decisions.

[0138] In a possible implementation, the graph representation vector sequence is extracted from a graph interrelationship mining network.

[0139] Before step S110, the method further includes:

[0140] In step A110, graph self-attention processing is performed on the first sample building energy monitoring data based on the basic network model, and the energy usage status pattern in the first sample building energy monitoring data is predicted based on the parsed graph representation vector sequence to generate a status pattern prediction result.

[0141] In this embodiment, the building energy management system first uses a basic network model to process a first sample of building energy monitoring data. This first sample of building energy monitoring data may include various monitoring indicators such as power consumption, temperature, and humidity in different areas of the building. This basic network model applies graph self-attention, a technique that enables the model to automatically focus on important connections between different nodes in the graph when processing graph-structured data.

[0142] During processing, the basic network model uses the graph self-attention mechanism to assign weights to each data point (i.e., node in the graph) in the first example of building energy monitoring data. These weights reflect the relevance and importance of the data points. In this way, the model can capture key information about the building's energy usage patterns.

[0143] After completing the graph self-attention process, the basic network model uses the processing results to predict the energy usage status pattern in the first example building energy monitoring data. The prediction result may be one or more status pattern labels, such as "peak power consumption mode" or "energy saving mode." These labels describe and categorize the current energy usage status of the building.

[0144] The building energy management system will save the predicted results for subsequent comparison and evaluation with the actual marked results.

[0145] Step A120 : determining a prediction error parameter according to the state mode prediction result and the state mode labeling result of the first sample building energy monitoring data.

[0146] To evaluate the prediction performance of the basic network model, the building energy management system needs to compare the prediction results with the state pattern annotation results of the first sample building energy monitoring data. The state pattern annotation results are previously obtained by experts or other reliable methods and represent the actual state of building energy usage.

[0147] The building energy management system calculates the difference, or error, between the predicted and labeled results. This error can be measured using metrics such as precision, recall, and F1 score. These metrics constitute the prediction error parameter, reflecting the accuracy and reliability of the model in predicting energy usage patterns.

[0148] Step A130: Training the basic network model based on the prediction error parameters to generate a target network model.

[0149] Based on the prediction error parameters, the building energy management system trains and optimizes the underlying network model. During training, the building energy management system adjusts the model's parameters and structure based on the error parameters, enabling the model to more accurately capture key information about the building's energy usage patterns in subsequent prediction tasks.

[0150] After multiple rounds of training and optimization, the building energy management system will obtain a target network model with better performance. This target network model can more accurately predict energy usage patterns when processing similar building energy monitoring data.

[0151] Step A140 prunes the classification units used for state mode prediction in the target network model, and uses the target network model with the pruned classification units as the graph self-correlation mining network. The graph self-correlation mining network is used to construct a machine learning model with the correlation mining network and the temporal cycle control network.

[0152] After obtaining the target network model, the building energy management system will further optimize it. Specifically, the building energy management system will prune the classification unit used for state pattern prediction in the target network model. This classification unit may be a neural network layer or a specific group of neurons in the model.

[0153] The purpose of pruning is to remove redundant or unnecessary parts of the model to reduce the complexity of the model and improve the operation efficiency. Through pruning, the building energy management system can obtain a more streamlined and high-performance graph self-interrelationship mining network.

[0154] Finally, the building energy management system combines this graph self-correlation mining network with other components, such as the correlation mining network and the temporal cycle control network, to construct a complete machine learning model. This machine learning model leverages the strengths of various network structures to perform comprehensive and in-depth analysis and processing of building energy monitoring data.

[0155] In one possible implementation, the method further includes:

[0156] Step B110, using the graph self-correlation mining network in the machine learning model to perform graph self-attention processing on each second sample building energy monitoring data in the sample building energy monitoring data sequence, to generate a prediction graph representation vector sequence for each second sample building energy monitoring data.

[0157] In this embodiment, the sample building energy monitoring data sequence received by the building energy management system may include the building energy usage at multiple time points. For each second sample building energy monitoring data in the sample building energy monitoring data sequence, the building energy management system uses the graph self-correlation mining network in the machine learning model to perform graph self-attention processing.

[0158] During this process, the graph self-correlation mining network automatically focuses on key information and inter-node correlations within the sample building energy monitoring data. For example, it may discover unusually high energy usage at certain time points, which are associated with other important events (such as weather changes or equipment failures). By weighting and aggregating this information, the network can generate a sequence of predictive graph representation vectors for each second sample of building energy monitoring data.

[0159] Step B120, using the correlation mining network in the machine learning model to perform focused tracking based on vector dependency on the prediction graph representation vector sequence of the second sample building energy monitoring data adjacent to each time and space domain, to generate a first energy usage cross-domain prediction dependency relationship vector between the second sample building energy monitoring data adjacent to each time and space domain.

[0160] In this embodiment, the building energy management system then utilizes the interrelationship mining network within the machine learning model to focus on and track the sequence of predicted graph representation vectors of the second example building energy monitoring data that are adjacent in time and space. Here, spatiotemporal adjacency refers to data points that are adjacent or related in time and space.

[0161] The interdependence mining network tracks and analyzes each data point adjacent to each other in spatiotemporal domain based on the dependency relationships in the prediction graph representation vector sequence. For example, the interdependence mining network might discover that energy usage in a certain area has been rising continuously over a certain period of time, and this upward trend is closely correlated with changes in energy usage in adjacent areas. By capturing and quantifying these dependencies, the interdependence mining network can generate a cross-domain predicted dependency relationship vector for first energy usage between the energy monitoring data of second-example buildings adjacent to each other in spatiotemporal domain.

[0162] Step B130, using the timing loop control network in the machine learning model to perform timing processing on each predicted dependency relationship vector of the first energy usage across domains and each predicted graph representation vector sequence of the second sample building energy monitoring data, to generate the predicted dependency relationship vector of the second energy usage across domains.

[0163] The building energy management system then uses the time-series recurrent control network in the machine learning model to perform time-series processing on the first energy usage cross-domain prediction dependency relationship vectors and the sequence of prediction graph representation vectors of the second example building energy monitoring data. The time-series recurrent control network has the ability to handle time dependencies in sequential data.

[0164] During this process, the temporal recurrent control network dynamically models and predicts the sequence based on the temporal information and data features in the input sequence. For example, the temporal recurrent control network might capture the cyclical variation pattern of energy usage throughout the day and how this cyclical variation pattern adjusts to external factors such as seasons and weather. By modeling and analyzing these temporal dependencies and variation patterns, the temporal recurrent control network can generate a second vector of predicted cross-domain energy usage dependencies.

[0165] Step B140 , predicting the energy usage state pattern in the second sample building energy monitoring data according to the second energy usage cross-domain prediction dependency change relationship vector, and generating a state pattern training result.

[0166] Based on the second energy usage cross-domain prediction dependency relationship vector, the building energy management system predicts the energy usage status pattern in the second example building energy monitoring data. The prediction results may include information such as energy usage status, energy consumption level, and energy saving potential at different time points.

[0167] For example, a building energy management system might predict that during a certain period of time, the lighting system in a certain area will be in a high energy consumption state, while the air conditioning system will be in a low energy consumption state. These prediction results can provide building managers with targeted energy-saving suggestions and optimization plans.

[0168] Step B150: Training the machine learning model based on a training error parameter between the state pattern training result and the state pattern labeling result of the energy usage state pattern.

[0169] Finally, the building energy management system trains and optimizes the machine learning model based on the training error parameter between the state pattern training results and the state pattern labeling results of the energy usage state pattern. The state pattern labeling results are the true state labels previously obtained by experts or other reliable methods.

[0170] By comparing the differences and errors between predicted and labeled results, the building energy management system can evaluate the model's predictive performance and adjust the model's parameters and structure based on the error parameters. For example, if the model's predicted results at a certain point in time deviate significantly from the actual labeled results, the building energy management system can improve the model's predictive performance by adjusting the model's weight distribution or adding new training samples. After multiple rounds of training and optimization, the machine learning model will be able to better adapt to actual building energy monitoring scenarios and provide more accurate predictions.

[0171] In a possible implementation, the predicted dependency relationship vector of the first energy usage across domains includes a predicted dependency relationship vector of each energy usage node in the first energy usage across domains.

[0172] Step B130 may include:

[0173] Step B131 : determining a potential situation vector based on a graph representation vector sequence corresponding to a first energy usage node in the building energy monitoring data set.

[0174] In this embodiment, the building energy management system first processes cross-domain energy usage data, including predicted dependency relationship vectors for multiple energy usage nodes (e.g., lighting systems, air conditioning systems, etc.). The predicted dependency relationship vector for each energy usage node reflects the correlation between that node's energy usage and other nodes or external factors.

[0175] For example, the energy use of lighting systems may vary with the duration of daylight, while the energy use of air conditioning systems may be closely related to outdoor temperature. The building energy management system captures the interrelationships between these nodes and forms a first energy use cross-domain prediction dependency relationship vector.

[0176] When processing the second example building energy monitoring data, the building energy management system will first focus on the graph representation vector sequences corresponding to the first energy usage node. These sequences contain historical information and patterns of energy usage.

[0177] The building energy management system analyzes these graph-representing vector sequences to extract potential energy usage patterns—that is, overall trends and patterns in energy usage. For example, by analyzing historical data, the building energy management system might discover that energy usage at a particular node has been increasing over a specific time period. This trend information is encoded as a potential pattern vector, which serves as the basis for subsequent predictions.

[0178] Step B132: Integrate the predicted dependency relationship vector of the second energy usage node across the first energy usage domain with the potential situation vector to generate a predicted integrated vector.

[0179] The building energy management system then integrates the predicted dependency relationship vectors of the first energy usage and the second energy usage node across the domain with the potential situation vector. This step combines the interrelationships between nodes with the overall trend information.

[0180] For example, if the energy use of the air conditioning system is closely related to the outdoor temperature, and the overall trend shows that the outdoor temperature is rising, the building energy management system can predict that the energy use of the air conditioning system will also increase. This integrated information is encoded as a prediction integration vector, which combines the interrelationships between nodes and the overall trend information.

[0181] Step B133: Process the predicted integrated vector using the flow migration impact prediction unit in the timing loop control network to generate predicted flow migration weight information.

[0182] The building energy management system uses the flow migration impact prediction unit in the time-series cyclic control network to process the predicted integrated vector. This flow migration impact prediction unit analyzes the time dependency and change pattern of the integrated vector and generates predicted flow migration weight information.

[0183] These predicted flow and migration weights reflect the importance of energy use flows and migrations at different points in time and between different nodes. For example, lighting energy use may be more important during certain time periods, while air conditioning energy use may dominate during other time periods. This weight information provides a critical reference for subsequent forecasts.

[0184] Step B134: Process the predicted integrated vector using the flow transfer and blending unit in the timing cycle control network to generate predicted flow transfer and blending information.

[0185] The building energy management system further processes the predicted integrated vector using a flow blending unit within the sequential cyclic control network. This unit captures the multi-source information and interactions within the integrated vector to generate predicted flow blending information. This predicted flow blending information reflects the mutual influence and blending of energy usage between different nodes. For example, energy usage between the lighting and air conditioning systems may be somewhat complementary. When lighting energy usage increases, the air conditioning system may correspondingly reduce energy usage to maintain indoor comfort. This blending information provides a more comprehensive perspective for subsequent predictions.

[0186] Step B135 : Processing the predicted flow blending information and the predicted integrated vector using the potential situation prediction unit in the sequential cyclic control network to generate a second reference potential situation vector.

[0187] The building energy management system utilizes a potential situation prediction unit within the sequential cyclic control network to process the predicted flow and integration information and the predicted integrated vector. This potential situation prediction unit comprehensively considers the integration information and the trend information in the integrated vector to generate a second reference potential situation vector. This second reference potential situation vector provides a prediction and reference for future energy usage trends, integrating various information such as historical trends, inter-node relationships, and integration. For example, based on historical data and current environmental factors, the building energy management system may predict that the building's overall energy use will increase within a certain period of time in the future.

[0188] Step B136: Generate a second energy usage cross-domain predicted dependency relationship vector based on the second reference potential situation vector and the predicted flow migration weight information.

[0189] Finally, the building energy management system generates a second cross-domain predicted dependency relationship vector for energy use based on the second reference potential state vector and the predicted flow and migration weight information. This predicted dependency relationship vector predicts and describes the future relationships between energy use nodes. By integrating multiple factors, including potential state, flow and migration weights, and integration information, it provides building managers with a comprehensive forecast and reference for future energy use. For example, the building energy management system may predict that energy use between the lighting system and the air conditioning system will exhibit a specific relationship or change pattern within a certain time period. This predictive information is crucial for developing energy-saving strategies and optimizing energy use.

[0190] Figure 2In an embodiment of the present application, a building energy management system 100 is provided, including a processor 1001, a memory 1003, and program code stored in the memory 1003. The processor 1001 executes the above program code to implement the steps of the data analysis method applied to building energy management.

[0191] Figure 2 The illustrated building energy management system 100 includes a processor 1001 and a memory 1003. The processor 1001 and the memory 1003 are connected, for example, via a bus 1002. Optionally, the building energy management system 100 may also include a transceiver 1004, which can be used for data exchange between the building energy management system 100 and other building energy management systems 100, such as data transmission and / or data reception. It should be noted that in actual scheduling, the number of transceivers 1004 is not limited to one, and the structure of the building energy management system 100 does not constitute a limitation on the embodiments of this application.

[0192] The processor 1001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor 1001 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like.

[0193] The bus 1002 may include a path for transmitting information between the above components. The bus 1002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus 1002 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 2 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0194] The memory 1003 can be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium that can be used to carry or store program code and can be read by a computer, without limitation here.

[0195] The memory 1003 is used to store program codes for executing the embodiments of the present application, and the execution is controlled by the processor 1001. The processor 1001 is used to execute the program codes stored in the memory 1003 to implement the steps shown in the above method embodiments.

[0196] An embodiment of the present application provides a computer-readable storage medium having program code stored thereon. When the program code is executed by a processor, the steps and corresponding contents of the aforementioned method embodiment can be implemented.

[0197] The above is only an optional implementation method for some implementation scenarios of this application. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the solution of this application, other similar implementation methods based on the technical ideas of this application also fall within the protection scope of the embodiments of this application.

Claims

1. A data analysis method for building energy management, characterized in that: Applied to a building energy management system, the method includes: Performing graph self-attention processing on each building energy monitoring data in the building energy monitoring data set to generate a graph representation vector sequence for each building energy monitoring data; each building energy monitoring data is one of electric energy monitoring data, water consumption monitoring data, gas consumption monitoring data, or heat consumption monitoring data, and each graph representation vector sequence for the building energy monitoring data is one of an electric energy graph representation vector, a water consumption graph representation vector, a gas consumption graph representation vector, or a heat consumption graph representation vector; Performing mutual correlation mining on the graph representation vector sequences of the building energy monitoring data adjacent to each spatiotemporal domain, respectively, to generate a first energy usage cross-domain dependency relationship vector between the building energy monitoring data adjacent to each spatiotemporal domain; the first energy usage cross-domain dependency relationship vector represents the dependency relationship between the building energy monitoring data at an instantaneous energy flow interval; the mutual correlation mining includes characteristic embedding analysis, which is used to identify mutual influence and dependency relationships between energy usage characteristics of different buildings; generating a second energy usage cross-domain dependency relationship vector based on each dependency relationship vector of the first energy usage cross-domain and a graph representation vector sequence of each building energy monitoring data; the second energy usage cross-domain dependency relationship vector represents a dependency relationship between the building energy monitoring data of an extended energy flow interval, and the second energy usage cross-domain is greater than the first energy usage cross-domain; performing pattern analysis on a target energy entity in the building energy monitoring data according to the dependency change relationship vector of the second energy usage across domains, and generating a pattern analysis result reflecting an energy status pattern; Building energy allocation is performed based on the pattern analysis results and the associated building energy allocation information, or the regulation system in the building is intelligently regulated based on the pattern analysis results.

2. The data analysis method for building energy management according to claim 1, characterized in that: Before performing graph self-attention processing on each building energy monitoring data in the building energy monitoring data set to generate a graph representation vector sequence for each building energy monitoring data, the method further includes: Acquiring an energy data stream obtained by energy monitoring the target energy entity within a preset monitoring period, parsing the energy data stream to obtain an energy data point sequence, sampling data in the energy data point sequence, and generating the building energy monitoring data set; Alternatively, a building energy monitoring data set obtained by sampling energy data of the target energy entity within the preset monitoring period is obtained.

3. The data analysis method for building energy management according to claim 1, characterized in that: The mutual correlation mining of the graph representation vector sequences of the building energy monitoring data adjacent to each spatiotemporal domain to generate the first energy usage cross-domain dependency change relationship vector between the building energy monitoring data adjacent to each spatiotemporal domain includes: A graph representation vector sequence of the first building energy monitoring data in the building energy monitoring data adjacent to each spatiotemporal domain is used as a reference vector, and a graph representation vector sequence of the second building energy monitoring data in the building energy monitoring data adjacent to each spatiotemporal domain is used as a tracking vector. The reference vector and the tracking vector are subjected to focused tracking based on vector dependency, and a first energy usage cross-domain dependency relationship vector is generated between the building energy monitoring data adjacent to each spatiotemporal domain.

4. The data analysis method for building energy management according to claim 3, characterized in that: The reference vector includes a vector index and a vector content; The performing focused tracking on the reference vector and the tracking vector based on vector dependency to generate a first energy usage cross-domain dependency relationship vector between the building energy monitoring data adjacent in each spatiotemporal domain includes: performing focus tracking based on vector-dependent changes on the vector index and the tracking vector to generate a vector focus tracking result; Performing a logical relationship quantitative analysis on the vector focus tracking result to generate a logical relationship focus quantization coefficient; A first cross-domain dependent change relationship vector of energy usage between the building energy monitoring data adjacent in each spatiotemporal domain is generated based on the logic relationship focused quantization coefficient and the vector content.

5. The data analysis method for building energy management according to claim 4, characterized in that: The performing focus tracking on the vector index and the tracking vector based on vector-dependent changes to generate a vector focus tracking result includes: Defining an initial window size for a sliding window used for focus tracking, and setting a sensitivity threshold for focus tracking. The sliding window is used to slide over the vector index of the reference vector and compare it with the tracking vector. The sensitivity threshold is used to determine the similarity or difference between the tracking vector and the reference vector within any sliding window. Taking the starting point of the reference vector as the starting point, constructing a corresponding sliding window with an adjustable window size based on the initial window size, wherein the sliding window is dynamically resized as focus tracking proceeds to adapt to the dependent changes between the reference vector and the tracking vector; In each sliding window, calculating a similarity metric parameter between a subsequence of the reference vector and a corresponding portion of the tracking vector, wherein the similarity metric parameter is used to quantify a degree of matching between the reference vector and the tracking vector in the local window; When the similarity measure exceeds a preset sensitivity threshold, the sliding window is marked as a dependent change point, and the size and position of the sliding window are dynamically adjusted based on the identified dependent change point; the dependent change point indicates that the tracking vector has a significant change correlation with the reference vector within the sliding window; If no dependent change greater than the preset change floating value is detected in a plurality of consecutive sliding windows, the window size of the sliding window is increased based on the first preset window control parameter; if a dependent change greater than the preset change floating value is detected in a plurality of consecutive sliding windows, the window size of the sliding window is reduced based on the second preset window control parameter; Stepwise moving the sliding window along the vector index of the reference vector, and repeating the above steps until the entire reference vector is traversed, recording all identified dependent change points and corresponding window information; All identified dependent change points are sorted according to the position and time sequence of the dependent change points on the baseline vector, and a vector focus tracking result is generated based on the sorted dependent change point list and the corresponding window information. The vector focus tracking result includes the key change points and change patterns of the tracking vector relative to the baseline vector in the spatiotemporal domain.

6. The data analysis method for building energy management according to claim 1, characterized in that: The first energy usage cross-domain dependency change relationship vectors include the first energy usage cross-domain dependency change relationship vectors of energy usage nodes; The generating of the second energy usage cross-domain dependency relationship vector based on the first energy usage cross-domain dependency relationship vectors and the graph representation vector sequence of the building energy monitoring data includes: determining a potential situation vector based on a graph representation vector sequence corresponding to a first energy usage node in the building energy monitoring data set; Integrating the dependency change relationship vector of the second energy usage node across the first energy usage domain with the potential situation vector to generate an integrated vector; the first energy usage node is an energy usage node preceding the second energy usage node; determining flow migration weight information based on the integrated vector and the first conversion array; Determining flow blending information based on the integrated vector and the second conversion array, and determining a first reference potential situation vector based on the potential situation vector, the flow blending information, and the third conversion array; wherein the flow migration weight information reflects the influence weight of the potential situation vector of the previous energy usage node on the potential situation vector of the current energy usage node, and the flow blending information reflects the blending information of the potential situation vector of the previous energy usage node and the dependent change relationship vector of the current energy usage node; Based on the first reference potential situation vector and the flow migration weight information, a second energy usage cross-domain dependency change relationship vector is generated.

7. The data analysis method for building energy management according to any one of claims 1 to 6, characterized in that: The step of performing pattern analysis on the target energy entity in the building energy monitoring data based on the cross-domain dependency relationship vector of the second energy usage to generate a pattern analysis result reflecting the energy status pattern includes: Determining a characteristic distance between the cross-domain dependency change relationship vector of the second energy use and each preset mode feature vector in the preset mode feature vector sequence; Selecting a target characteristic distance that meets the preset distance requirement from the obtained characteristic distances; The state pattern portrait of the preset pattern feature vector corresponding to the target feature distance is used as the pattern analysis result.

8. The data analysis method for building energy management according to any one of claims 1 to 6, characterized in that: The graph representation vector sequence is extracted from a graph interrelationship mining network; Before performing graph self-attention processing on each building energy monitoring data in the building energy monitoring data set, the method further includes: Performing graph self-attention processing on the first sample building energy monitoring data based on the basic network model, and predicting the energy usage state pattern in the first sample building energy monitoring data based on the parsed graph representation vector sequence to generate a state pattern prediction result; determining a prediction error parameter according to the state mode prediction result and the state mode labeling result of the first sample building energy monitoring data; Training the basic network model based on the prediction error parameter to generate a target network model; The classification units in the target network model used for state mode prediction are pruned, and the target network model after the classification unit pruning is used as the graph self-correlation mining network; wherein, the graph self-correlation mining network is used to construct a machine learning model with the correlation mining network and the timing cycle control network.

9. The data analysis method for building energy management according to claim 8, characterized in that: The method further comprises: Performing graph self-attention processing on each second sample building energy monitoring data in the sample building energy monitoring data sequence using a graph self-correlation mining network in the machine learning model to generate a sequence of predicted graph representation vectors for each second sample building energy monitoring data; Using the correlation mining network in the machine learning model, focusing and tracking the prediction graph representation vector sequences of the second sample building energy monitoring data adjacent in each spatiotemporal domain based on vector dependency, to generate a first energy usage cross-domain prediction dependency relationship vector between the second sample building energy monitoring data adjacent in each spatiotemporal domain; Performing time series processing on each predicted dependency relationship vector of the first energy usage across domains and each predicted graph representation vector sequence of the second sample building energy monitoring data using a time series loop control network in the machine learning model to generate a predicted dependency relationship vector of the second energy usage across domains; Predicting an energy usage state pattern in the second sample building energy monitoring data according to the second energy usage cross-domain prediction dependency change relationship vector to generate a state pattern training result; Training the machine learning model based on a training error parameter between the state pattern training result and the state pattern labeling result of the energy usage state pattern; The predicted dependency relationship vector of the first energy usage across domains includes the predicted dependency relationship vector of each energy usage node across the first energy usage across domains; Performing time series processing on each predicted dependency relationship vector of the first energy usage across domains and each predicted graph representation vector sequence of the second sample building energy monitoring data using a time series loop control network in the machine learning model to generate the predicted dependency relationship vector of the second energy usage across domains includes: determining a potential situation vector based on a graph representation vector sequence corresponding to a first energy usage node in the building energy monitoring data set; Integrating the predicted dependency relationship vector of the second energy usage node across the first energy usage domain with the potential situation vector to generate a predicted integrated vector; Processing the predicted integrated vector using a flow migration impact prediction unit in the timing cycle control network to generate predicted flow migration weight information; Processing the predicted integrated vector using a flow transfer and blending unit in the timing cycle control network to generate predicted flow transfer and blending information; Processing the predicted flow blending information and the predicted integrated vector using a potential situation prediction unit in the sequential cyclic control network to generate a second reference potential situation vector; Based on the second reference potential situation vector and the predicted flow migration weight information, a second energy usage cross-domain predicted dependency change relationship vector is generated.

10. A building energy management system, characterized in that: The building energy management system includes a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores machine-executable instructions, and the machine-executable instructions are loaded and executed by the processor to implement the data analysis method for building energy management described in any one of claims 1 to 9.

Images