Artificial intelligence-based building energy consumption analysis method and system

Abstract

The application relates to the technical field of building energy consumption analysis, and discloses a building energy consumption analysis method and system based on artificial intelligence. The method comprises the following steps: collecting building environment data to form an energy consumption basic data set; processing the data set to generate an energy consumption feature vector; constructing a prediction model to obtain an energy consumption predictor; comparing actual data by using the predictor to identify abnormalities and generate a report; formulating an optimization scheme based on the report to generate a control instruction; executing the instruction to record change data, update the feature library, and complete a closed loop. The application realizes closed loop management of accurate prediction, abnormality detection, optimization control and effect evaluation of building energy consumption, so that building energy use efficiency is improved, and energy waste is reduced.

Description

Technical Field

[0001] This application relates to the field of building energy consumption analysis technology, and in particular to a building energy consumption analysis method and system based on artificial intelligence. Background Technology

[0002] With increasing awareness of energy consumption and environmental protection, building energy conservation has become a crucial global issue. Building energy consumption accounts for a large proportion of total energy consumption; statistics show that it accounts for approximately 40% of global energy consumption. Traditional building energy consumption analysis methods mainly rely on manual experience and simple statistical analysis, such as periodic meter readings and manual comparison of energy consumption data. With the development of Internet of Things (IoT) technology, Building Energy Management Systems (BEMS) have begun to be applied to building energy consumption monitoring, achieving automated monitoring of building energy consumption through the installation of various sensors and metering devices. In recent years, with the rapid development of artificial intelligence technology, some studies have begun to explore the application of machine learning algorithms to building energy consumption prediction and anomaly detection, such as using support vector machines and random forests to build energy consumption prediction models, or using clustering algorithms to identify abnormal energy consumption patterns.

[0003] However, existing building energy consumption analysis methods still have many shortcomings. First, most methods fail to fully consider the spatiotemporal correlation of building energy consumption, ignoring the mutual influence between energy consumption in different regions and time periods, resulting in limited prediction accuracy. Second, existing methods typically use static thresholds for anomaly detection, which cannot adapt to the dynamic changes in building energy consumption with seasons, weather, and usage patterns, easily leading to false alarms and missed alarms. Third, they lack adaptive learning capabilities, failing to automatically adjust model parameters based on newly acquired data and feedback results, making it difficult to adapt to the long-term evolution of building energy consumption characteristics. Fourth, existing methods often treat energy consumption prediction, anomaly detection, and energy-saving optimization as independent processes, lacking an integrated closed-loop management mechanism, making it difficult to accurately evaluate and continuously improve the effectiveness of energy-saving measures. Finally, existing methods have limited ability to fuse and process multi-source heterogeneous data, making it difficult to fully utilize the rich information provided by multi-dimensional data such as building temperature, humidity, and occupancy density. Summary of the Invention

[0004] This application provides an artificial intelligence-based building energy consumption analysis method and system, which enables closed-loop management of accurate prediction, anomaly detection, optimized control, and effect evaluation of building energy consumption, thereby improving building energy efficiency and reducing energy waste.

[0005] Firstly, this application provides an artificial intelligence-based building energy consumption analysis method, which includes: collecting building temperature, humidity, energy consumption, and occupancy density data through multi-dimensional sensors to obtain a basic building energy consumption dataset; performing data missing filling and outlier processing on the basic building energy consumption dataset, extracting building usage periods, environmental parameters, and energy load characteristics to generate a building energy consumption feature vector; establishing a building energy consumption prediction model considering time factors and spatial relationships based on the building energy consumption feature vector, and training it with historical data to obtain a building energy consumption predictor; calculating the expected energy consumption value using the building energy consumption predictor, comparing the actual energy consumption data with the expected energy consumption value, identifying abnormal energy consumption intervals, and generating a building energy consumption anomaly report; based on the building energy consumption anomaly report, combined with the building usage plan and environmental conditions, formulating an energy optimization parameter adjustment scheme, and generating a building energy-saving control command; executing the building energy-saving control command, recording energy consumption changes before and after the adjustment, calculating energy efficiency improvement indicators, updating the building energy consumption feature library, and completing the building energy consumption analysis closed loop.

[0006] Secondly, this application provides an artificial intelligence-based building energy consumption analysis system, the artificial intelligence-based building energy consumption analysis system comprising:

[0007] The data acquisition module is used to collect data on building temperature, humidity, energy consumption and personnel density through multi-dimensional sensors to obtain a basic dataset of building energy consumption.

[0008] The extraction module is used to perform data missing imputation and outlier processing based on the building energy consumption basic dataset, extract building usage time period, environmental parameters and energy load characteristics, and generate building energy consumption feature vector;

[0009] A module is established to build a building energy consumption prediction model that considers time factors and spatial relationships based on the building energy consumption feature vector, and to obtain a building energy consumption predictor through training with historical data.

[0010] The comparison module is used to calculate the expected energy consumption value using the building energy consumption predictor, compare the actual energy consumption data with the expected energy consumption value, identify abnormal energy consumption intervals, and generate a building energy consumption anomaly report.

[0011] The generation module is used to formulate an energy optimization parameter adjustment scheme based on the building energy consumption anomaly report, combined with the building usage plan and environmental conditions, and generate building energy-saving control instructions;

[0012] The update module is used to execute the building energy-saving control command, record the energy consumption change data before and after the adjustment, calculate the energy efficiency improvement index, update the building energy consumption feature library, and complete the building energy consumption analysis closed loop.

[0013] A third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the computer device to execute the above-described artificial intelligence-based building energy consumption analysis method.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described artificial intelligence-based building energy consumption analysis method.

[0015] The technical solution provided in this application acquires comprehensive building energy consumption data by collecting building temperature, humidity, energy consumption, and occupancy density data through multi-dimensional sensors. This provides a rich data foundation for subsequent analysis and solves the problem of single data sources in traditional methods. By imputing missing data and handling outliers in the building energy consumption dataset, and extracting building usage periods, environmental parameters, and energy load characteristics, a building energy consumption feature vector is generated. This significantly improves data quality and feature representation capabilities, providing high-quality input for model training. Based on the building energy consumption feature vector, a building energy consumption prediction model considering time factors and spatial relationships is established. This fully captures the spatiotemporal correlation of building energy consumption, overcomes the limitation of traditional models that ignore the mutual influence of energy consumption between regions, and improves prediction accuracy. The building energy consumption predictor calculates the expected energy consumption value, compares the actual energy consumption data with the expected energy consumption value, identifies abnormal energy consumption intervals, and generates a building energy consumption anomaly report. This realizes a shift from passive discovery to proactive early warning, greatly shortening the anomaly detection time. Based on building energy consumption anomaly reports, combined with building usage plans and environmental conditions, a targeted energy optimization parameter adjustment scheme was developed, generating building energy-saving control instructions. This achieved precise energy saving, avoiding the comfort degradation caused by the "one-size-fits-all" approach of traditional methods. By executing building energy-saving control instructions, recording energy consumption changes before and after adjustments, calculating energy efficiency improvement indicators, and updating the building energy consumption feature database, a closed-loop building energy consumption analysis was completed, achieving self-learning and continuous optimization. This scheme fully leverages the advantages of artificial intelligence algorithms in specific building energy consumption management fields, particularly in spatiotemporal data processing. It captures the temporal dependencies of energy consumption through a bidirectional long short-term memory network, models the spatial relationships between building areas through a graph convolutional network, and dynamically adjusts the importance weights of different features through an attention mechanism. These algorithmic features contribute to the scheme by adaptively handling the dynamic changes in building energy consumption and identifying complex energy consumption patterns, thereby achieving more accurate predictions and more refined control. Furthermore, a genetic algorithm is used to solve a multi-objective optimization problem, balancing energy saving, cost, and comfort dimensions to achieve a globally optimal energy management strategy. Incremental learning mechanisms enable models to continuously learn from new data, improve prediction accuracy, and adapt to the long-term evolution of building energy consumption characteristics. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of an embodiment of the building energy consumption analysis method based on artificial intelligence in this application.

[0018] Figure 2 This is a schematic diagram of one embodiment of the building energy consumption analysis system based on artificial intelligence in this application.

[0019] Figure 3 This is a schematic block diagram of the structure of the computer device in an embodiment of the present invention. Detailed Implementation

[0020] This application provides an artificial intelligence-based method and system for building energy consumption analysis. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0021] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the building energy consumption analysis method based on artificial intelligence in this application includes:

[0022] Step S101: Collect building temperature, humidity, energy consumption and personnel density data through multi-dimensional sensors to obtain basic data set of building energy consumption;

[0023] Step S102: Based on the building energy consumption basic dataset, perform data missing imputation and outlier processing, extract building usage time period, environmental parameters and energy load characteristics, and generate building energy consumption feature vector;

[0024] Step S103: Based on the building energy consumption feature vector, establish a building energy consumption prediction model that considers time factors and spatial relationships, and obtain a building energy consumption predictor through training with historical data.

[0025] Step S104: Calculate the expected energy consumption value using the building energy consumption predictor, compare the actual energy consumption data with the expected energy consumption value, identify abnormal energy consumption intervals, and generate a building energy consumption anomaly report.

[0026] Step S105: Based on the building energy consumption anomaly report, and in conjunction with the building usage plan and environmental conditions, formulate an energy optimization parameter adjustment plan and generate building energy-saving control instructions;

[0027] Step S106: By executing building energy-saving control commands, record the energy consumption change data before and after the adjustment, calculate the energy efficiency improvement index, update the building energy consumption feature library, and complete the building energy consumption analysis closed loop.

[0028] It is understood that the executing entity of this application can be an AI-based building energy consumption analysis system, or it can be a terminal or a server; no specific limitation is made here. This application's embodiment uses a server as an example for illustration.

[0029] Specifically, a building energy consumption baseline dataset is obtained by collecting data on building temperature, humidity, energy consumption, and occupant density using multi-dimensional sensors. Specifically, a temperature sensor network is deployed in each functional area of ​​the building to collect temperature change data, forming a spatiotemporal temperature data matrix; simultaneously, humidity distribution data within the building is collected using multi-point humidity sensors, and combined with ventilation status parameters to generate a humidity environmental feature set; energy metering devices collect energy consumption data for major energy-consuming equipment in the building, recording equipment operating power curves and constructing an equipment energy consumption database; and occupant flow and density data are collected using personnel detection devices to establish an occupant activity pattern feature table. These data undergo time synchronization processing, invalid data points are removed, and missing time periods are supplemented, ultimately forming a building energy consumption baseline dataset containing multi-dimensional information. Based on the building energy consumption baseline dataset, data missingity is filled and outlier processing is performed, and building usage time periods, environmental parameters, and energy load characteristics are extracted to generate a building energy consumption feature vector. The specific processing steps include filling missing data using a sliding time window average filling method, which calculates the average of valid data within a certain time window before and after the missing point; then identifying outliers using an improved Z-score method, calculating the deviation of each data point from the surrounding data, marking points with excessive deviations as outliers and replacing them with the median of the preceding and following data points; subsequently, segmenting the smoothed data into time series segments based on energy consumption load variation characteristics into working periods, transition periods, and idle periods; performing noise reduction using wavelet transform to decompose the energy consumption data into trend terms, seasonal terms, and residual terms; extracting environmental parameter features such as temperature difference, thermal inertia index, solar radiation impact coefficient, humidity change rate, and population density distribution from the processed data; and finally integrating the time period segmentation table, energy consumption decomposition features, and environmental impact factor set through a feature fusion matrix to form a complete building energy consumption feature vector.

[0030] Based on building energy consumption feature vectors, a building energy consumption prediction model considering time factors and spatial relationships is established. This model is trained using historical data to obtain a building energy consumption predictor. The process first divides the building energy consumption feature vectors into temporal and spatial feature sub-vectors. The temporal feature sub-vectors are processed using a bidirectional long short-term memory network (LSTM), which calculates hidden states through forward and backward time steps to capture long-term dependencies. The spatial feature sub-vectors are then subjected to graph convolution, representing each building region as a node in the graph and the physical connections between regions as edges. A message passing mechanism is used to capture the mutual influence of energy consumption between different regions. The processed temporal relationship feature matrix and spatial correlation feature graph are fused using an attention mechanism, and the weight coefficients of each feature channel are calculated for weighted combination. The fused features are then paired with historical building energy consumption data to form training samples, and the model parameters are optimized using a mini-batch gradient descent algorithm. Finally, cross-validation is used to evaluate the model's predictive stability and accuracy under different data partitions, and the model with the best performance is selected as the building energy consumption predictor.

[0031] The system utilizes a building energy consumption predictor to calculate expected energy consumption values. Actual energy consumption data is then compared with these expected values ​​to identify abnormal energy consumption intervals and generate a building energy consumption anomaly report. Specifically, the process involves inputting current building parameter data into the predictor for forward calculation to obtain expected energy consumption values ​​at each time point; collecting actual building energy consumption data for the current period and aligning it with the expected energy consumption sequence; calculating the difference between the actual and expected energy consumption sequences to generate an energy consumption residual sequence; analyzing the residual sequence using a dynamic threshold detection algorithm to calculate a threshold for each time point, identifying anomalies when the absolute value of the residual exceeds the corresponding threshold; performing continuity analysis on the anomalies, marking intervals with three or more consecutive anomalies as energy consumption anomaly intervals; and classifying the anomalies based on their characteristic parameters using a decision tree algorithm to identify the anomaly type and generate a building energy consumption anomaly report that includes the anomaly location, type, severity, and suggested handling methods.

[0032] Based on building energy consumption anomaly reports, combined with building usage plans and environmental conditions, an energy optimization parameter adjustment scheme is formulated to generate building energy-saving control instructions. This step first extracts anomaly types and energy consumption deviation data from the anomaly reports to construct an optimization target list; collects building usage plan data, including work schedules, occupancy plans, and special event times; obtains future environmental condition forecast data; constructs a multi-objective optimization function, including energy consumption indicators, economic cost indicators, and comfort indicators, sets weight coefficients for each indicator, and solves for the optimal parameter combination using a genetic algorithm; decomposes the obtained equipment operating parameter adjustment scheme into equipment-level control parameters; finally, converts the control parameter table into an instruction format conforming to the building control system communication protocol to form building energy-saving control instructions.

[0033] By executing building energy-saving control commands, recording energy consumption changes before and after adjustments, calculating energy efficiency improvement indicators, and updating the building energy consumption feature library, a closed-loop building energy consumption analysis is completed. Specifically, the process involves sending energy-saving control commands to the building control system and recording the execution timestamp; real-time monitoring of the adjusted operating status using multi-dimensional sensors to collect various data after adjustments; time-aligned comparison of the adjusted data with historical data from the same period before adjustments to calculate energy consumption differences; statistical analysis of energy consumption changes to calculate indicators such as average energy consumption reduction rate, peak energy consumption reduction, and load balance; fusion of key parameters with building energy consumption feature vectors to update the building energy consumption feature library; and incremental training of the predictor using the updated feature library to form a closed-loop optimization analysis system.

[0034] For example, after collecting a month's worth of basic energy consumption data through multi-dimensional sensors, it was found that 10% of the air conditioning energy consumption data during weekday working hours was missing. This missing data was filled using a sliding time window averaging method, with the window size set to 3 hours before and after the data point. The processed dataset showed that peak energy consumption periods were 9-11 am and 2-4 pm on weekdays. Wavelet transform was used to separate the daily and weekly cyclical patterns. The building energy consumption feature vector contained key information such as time features (weekday / non-weekday, daytime / nighttime), environmental features (indoor / outdoor temperature difference, solar radiation intensity), and usage features (personnel density distribution). The predictor trained based on these features demonstrated a prediction accuracy of over 90% on the test set. In a real-world application, the predictor found that actual energy consumption on a Tuesday afternoon was 30% higher than expected, lasting for 4 hours. The anomaly report identified this as excessive cooling caused by improper air conditioning parameter settings. Based on this report, the system generated control commands to adjust the air conditioning set temperature and airflow direction. After executing the commands, the energy consumption in that area decreased to normal levels, and the optimal operating parameters for that area were updated in the feature library.

[0035] In this embodiment, multi-dimensional sensors are used to collect data on building temperature, humidity, energy consumption, and occupancy density, obtaining a comprehensive building energy consumption dataset. This provides a rich data foundation for subsequent analysis and solves the problem of single data sources in traditional methods. By imputing missing data and handling outliers in the building energy consumption dataset, building usage periods, environmental parameters, and energy load characteristics are extracted to generate building energy consumption feature vectors. This significantly improves data quality and feature representation capabilities, providing high-quality input for model training. Based on the building energy consumption feature vectors, a building energy consumption prediction model considering time factors and spatial relationships is established. This fully captures the spatiotemporal correlation of building energy consumption, overcomes the limitation of traditional models that ignore the mutual influence of energy consumption between regions, and improves prediction accuracy. The building energy consumption predictor is used to calculate the expected energy consumption value. The actual energy consumption data is compared with the expected energy consumption value to identify abnormal energy consumption intervals and generate building energy consumption anomaly reports. This realizes the transformation from passive discovery to proactive early warning, greatly shortening the anomaly detection time. Based on building energy consumption anomaly reports, combined with building usage plans and environmental conditions, a targeted energy optimization parameter adjustment scheme was developed, generating building energy-saving control instructions. This achieved precise energy saving, avoiding the comfort degradation caused by the "one-size-fits-all" approach of traditional methods. By executing building energy-saving control instructions, recording energy consumption changes before and after adjustments, calculating energy efficiency improvement indicators, and updating the building energy consumption feature database, a closed-loop building energy consumption analysis was completed, achieving self-learning and continuous optimization. This scheme fully leverages the advantages of artificial intelligence algorithms in specific building energy consumption management fields, particularly in spatiotemporal data processing. It captures the temporal dependencies of energy consumption through a bidirectional long short-term memory network, models the spatial relationships between building areas through a graph convolutional network, and dynamically adjusts the importance weights of different features through an attention mechanism. These algorithmic features contribute to the scheme by adaptively handling the dynamic changes in building energy consumption and identifying complex energy consumption patterns, thereby achieving more accurate predictions and more refined control. Furthermore, a genetic algorithm is used to solve a multi-objective optimization problem, balancing energy saving, cost, and comfort dimensions to achieve a globally optimal energy management strategy. Incremental learning mechanisms enable models to continuously learn from new data, improve prediction accuracy, and adapt to the long-term evolution of building energy consumption characteristics.

[0036] In one specific embodiment, the process of performing step S101 may specifically include the following steps:

[0037] Temperature change data for each area is collected by a network of temperature sensors arranged according to the functional zones of the building, forming a spatiotemporal temperature data matrix.

[0038] Humidity distribution data within the building is collected using multi-point humidity sensors, and a humidity environment feature set is generated by combining ventilation status parameters.

[0039] Energy consumption data of major energy-consuming equipment in buildings are collected by energy metering devices, the operating power curves of the equipment are recorded, and an equipment energy consumption database is constructed.

[0040] Data on the flow and density of people in the building are collected using personnel detection devices to establish a table of personnel activity patterns.

[0041] Time synchronization processing is performed on the temperature spatiotemporal data matrix, humidity environmental feature set, equipment energy consumption database and personnel activity pattern feature table to remove invalid data points, supplement missing time period data, and generate a basic building energy consumption dataset.

[0042] Normalize the various types of data in the building energy consumption basic dataset to eliminate dimensional differences and obtain the building energy consumption basic dataset.

[0043] Specifically, when collecting temperature change data in different areas of the building through a network of temperature sensors arranged according to the building's functional zones, temperature sensors are installed in different functional areas of the building, such as office areas, meeting rooms, corridors, and equipment rooms. These sensors are connected to the data acquisition gateway via RS-485 bus or wireless communication. Multiple temperature measurement points are arranged in each area, typically with sensors installed in the four corners and center of each room, and the acquisition frequency is set to once every 5 minutes. The collected temperature data is organized according to both time and space dimensions to form a temperature spatiotemporal data matrix. The rows of this matrix represent different time points, the columns represent different spatial locations, and the matrix element values ​​are the temperature values ​​at the corresponding time points and locations. When collecting humidity distribution data within the building through multi-point humidity sensors, the humidity sensors are usually installed in the same location as the temperature sensors, with the same acquisition frequency and an accuracy of ±1%RH. In addition to humidity data, the operating status of the building's ventilation system also needs to be collected, including parameters such as fresh air volume, return air ratio, and fan speed. The humidity data is combined with ventilation status parameters for processing to calculate the rate of change of air humidity, absolute humidity value, and correlation coefficient between humidity and ventilation status, thereby generating a humidity environmental feature set. This feature set contains not only raw humidity data, but also information on humidity change trends and humidity regulation capabilities.

[0044] When collecting energy consumption data for major energy-consuming equipment in a building using energy metering devices, power monitoring devices, such as smart meters and sub-metering instruments, need to be installed on the building's main circuits and on each major electrical device. These devices monitor the power consumption of each device in real time with an accuracy of ±0.5%. The monitoring data includes parameters such as real-time power, cumulative power consumption, and power factor. Based on the power change curves recorded by the equipment's operating status, the start-stop characteristics, steady-state operating power, peak power, and other characteristics of the equipment are analyzed. The data is then categorized and organized according to multiple dimensions such as equipment type, functional area, and usage period to construct an equipment energy consumption database. Each record in this database includes fields such as equipment ID, timestamp, power value, cumulative power consumption, and operating status identifier. When collecting data on personnel flow and density within a building using personnel detection devices, various technologies are mainly employed, including infrared sensors, camera-based people counting systems, and WIFI probes. These devices are installed at various entrances and main passages of the building to record personnel entry and exit and the density of people in the area in real time. The collected data undergoes preliminary processing to form raw data records containing information such as time, location, number of people, and direction of movement. By analyzing these records, the characteristics of human activity in each area were extracted, such as peak hours, activity routes, and duration of stay, and a human activity pattern characteristic table was established. This characteristic table describes the patterns of human activity in each area over different time periods.

[0045] When performing time synchronization processing on the temperature spatiotemporal data matrix, humidity environmental feature set, equipment energy consumption database, and personnel activity pattern feature table, the first step is to ensure that all data has a timestamp in a standard format. The first step in time synchronization processing is to align data from different sources according to their timestamps, unifying them to the same time granularity, such as using 15-minute intervals. During this process, for data with a collection frequency higher than the unified granularity, the average value within the time window is used as a representative; for data with a collection frequency lower than the unified granularity, interpolation methods are used to supplement data at intermediate time points. The second step is to identify and remove invalid data points, including outliers caused by sensor malfunctions and missing values ​​due to communication interruptions. Criteria for determining invalid data include values ​​exceeding a reasonable range, excessive fluctuations within a short period, and data remaining completely unchanged for a long period. The third step is to supplement missing data for specific time periods. For short-term data gaps, linear interpolation or the average of preceding and following data is used for filling; for long-term data gaps, data from similar dates and time periods are used as replacements. After these processes, a building energy consumption basic dataset with continuous time dimensions and corresponding data for various time periods is obtained.

[0046] Normalization of various data types in the building energy consumption dataset aims to eliminate dimensional differences between them, facilitating subsequent comprehensive analysis. The normalization process employs a min-max normalization method, mapping all data to the [0,1] interval. For each data type, its historical maximum and minimum values ​​are first determined, and then converted using the formula: Normalized value = (Original value - Minimum value) / (Maximum value - Minimum value). This process maps different data types to the same numerical range, ensuring data comparability. Simultaneously, to address the possibility of new data exceeding historical ranges, data exceeding the original range is truncated: data exceeding the maximum value is normalized to 1, and data below the minimum value is normalized to 0. After normalization, a unified building energy consumption dataset is obtained, laying the foundation for subsequent feature extraction and model training.

[0047] Taking the data acquisition process of an office building as an example, the building is divided into four functional zones: administrative office area, technology research and development area, meeting area, and public area. A temperature sensor network with a total of 86 temperature measurement points is installed in each zone, recording temperature data every 5 minutes to form an 86-column spatiotemporal temperature data matrix. Simultaneously, humidity sensors are connected to the ventilation system controller to record humidity values ​​and ventilation parameters, generating a humidity environment feature set including humidity change rate and absolute humidity. Major energy-consuming equipment includes central air conditioning, lighting systems, office equipment, and elevators. Each system is equipped with a power monitoring device, recording energy consumption data every 15 minutes, constructing a database of energy consumption for each item. Personnel activity is monitored using infrared counters installed at entrances and main passageways, recording changes in personnel density in different areas, and establishing a personnel activity pattern feature table. The four types of data are synchronized using timestamps and unified to a 15-minute granularity. The missing data points of 8 temperature points and 3 humidity points are filled using linear interpolation. Abnormal energy consumption data points are filtered by setting thresholds. Finally, all data are mapped to the [0,1] interval using the min-max normalization method, forming a basic dataset of building energy consumption.

[0048] In one specific embodiment, the process of performing step S102 may specifically include the following steps:

[0049] The missing values ​​in the basic building energy consumption dataset were filled using the sliding time window average filling method to obtain a preliminary filled dataset;

[0050] An improved Z-score method is used to identify outliers in the initially filled dataset. The identified outliers are then replaced with the median of the preceding and following data points to generate a smoothed dataset.

[0051] The smoothed dataset is segmented into time series segments, and divided into working periods, transition periods, and idle periods according to the building energy load variation characteristics, forming a time period segmentation table;

[0052] The smooth dataset is denoised using wavelet transform to separate the trend, seasonal and residual terms of building energy consumption, and to construct energy consumption decomposition features.

[0053] Five environmental parameters—temperature difference, thermal inertia index, solar radiation impact coefficient, humidity change rate, and population density distribution—are extracted from the smoothed dataset to generate a set of environmental impact factors.

[0054] The time period division table, energy consumption decomposition characteristics, and environmental impact factor set are integrated through a feature fusion matrix to generate a building energy consumption feature vector.

[0055] Specifically, a sliding time window average imputation method is used to fill missing values ​​in the building energy consumption dataset. This method selects valid data within a certain time range before and after the missing point and calculates the average value to fill the missing point. In practice, a window size W is set, usually 24 hours. For a missing point t, valid data points within the time range of tW / 2 to t+W / 2 are searched, and the average value of these data points is calculated as the imputation value. When consecutive missing values ​​exceed a set threshold, such as more than 6 hours, a similar day pattern substitution method is used, that is, historical data of the same time period on the same type of day (weekday or rest day) are searched for and used to fill the missing value. This sliding window average imputation method considers the temporal continuity and periodicity of the data, and the resulting preliminary imputation dataset retains the basic trend of the original data. When identifying outliers in the preliminary imputation dataset using an improved Z-score method, the deviation of each data point from its surrounding data is calculated. The ordinary Z-score method subtracts the mean from the data point and divides by the standard deviation, while the improved Z-score method considers the temporal characteristics of the data and uses the mean and standard deviation within a local time window for judgment. Specifically, for each data point x(t), the mean μ and standard deviation σ of the data within its time window [tW / 2, t+W / 2] are calculated, and then the Z-score Z(t) = (x(t) - μ) / σ is calculated. When |Z(t)| exceeds a preset threshold (usually set to 3.0), it is identified as an outlier and replaced with the median of the data within the same time window. This method effectively identifies sudden data anomalies while avoiding misjudging normal seasonal fluctuations as anomalies. The replaced dataset is called a smoothed dataset, which eliminates sudden anomalies in the original data and is more suitable for subsequent pattern analysis.

[0056] Time series segmentation of a smoothed dataset is performed to identify energy consumption patterns across different time periods. Based on the characteristics of building energy load variations, a day is typically divided into working hours, transitional periods, and idle periods. Working hours refer to the time when the building operates normally and personnel activity is frequent, such as 9:00-18:00 for an office building; transitional periods refer to the time periods when the building transitions from idle to working or vice versa, such as 7:00-9:00 and 18:00-20:00; idle periods refer to the time periods when the building is largely unused, such as 20:00-7:00. The segmentation process determines the start and end times of each period by analyzing the combined changes in energy consumption data, personnel density data, and equipment operating status. For each working day and non-working day, a time period segmentation table is established, recording the start and end times of different periods and the statistical values ​​of energy consumption characteristics for each period, such as average, standard deviation, and maximum value.

[0057] Denoising smoothed datasets using wavelet transform is a crucial step in time series data analysis. Wavelet transform is a multi-resolution analysis method capable of representing signals simultaneously in the time and frequency domains. In building energy consumption analysis, the Daubechies wavelet basis is employed, with a decomposition level of 5. Energy consumption time series data is decomposed into coefficients across multiple frequency bands using wavelet transform. High-frequency coefficients correspond to short-term fluctuations (residual terms), mid-frequency coefficients to periodic variations (seasonal terms), and low-frequency coefficients to long-term trends (trend terms). Noise is removed by thresholding the coefficients in each frequency band (e.g., soft or hard thresholding), and the signal is then reconstructed to obtain the denoised data. Based on the wavelet decomposition results, energy consumption decomposition features can be constructed, including trend features (reflecting long-term energy consumption trends), seasonal features (reflecting periodic energy consumption patterns), and residual features (reflecting random disturbances). These features reveal the multi-scale variation patterns of building energy consumption.

[0058] The mathematical expression of wavelet transform is as follows:

[0059]

[0060] Among them, W ψ f(s,∈) are the wavelet coefficients of the function f(t) at scale s and translation ∈, ψ * It is the complex conjugate of the wavelet function, and f(t) represents the energy consumption time series. Through decomposition at different scales, a multi-scale representation of the energy consumption data can be obtained:

[0061]

[0062] Among them, A J (t) is the approximate part (trend term) of the J-layer wavelet decomposition, D j (t) represents the detailed part of the j-th layer, where J is the total decomposition layer number. Seasonal terms typically correspond to D in the middle layers.j (t), while the residual term is the most frequent detail.

[0063] Extracting environmental parameter features from smoothed datasets is crucial for establishing the correlation between energy consumption and environmental factors. Temperature difference features calculate the difference between indoor and outdoor temperatures, reflecting the intensity of heat conduction; thermal inertia indices represent the building's response speed to temperature changes, calculated through the time delay between indoor and outdoor temperature changes; the solar radiation impact coefficient is obtained by analyzing the correlation between energy consumption and solar radiation intensity; humidity change rate calculates the amount of humidity change per unit time; and personnel density distribution is calculated based on personnel monitoring data, determining the personnel density and its time-varying characteristics in each area. These five environmental parameter features together constitute the environmental impact factor set, comprehensively characterizing the impact of environmental factors on building energy consumption.

[0064] The calculation of the environmental impact factor set involves multiple formulas. Taking the thermal inertia index H_I as an example, its calculation formula is as follows:

[0065]

[0066] Where, ΔT in It is the change in indoor temperature, ΔT out It is the change in external temperature, Δt delay This refers to the time lag between indoor temperature changes and outdoor temperature changes. The formula for calculating the solar radiation impact coefficient S_C is:

[0067]

[0068] Wherein, cov(E,I) s () is the energy consumption E and solar radiation intensity I s covariance, σ E and These are the standard deviations of energy consumption and solar radiation intensity, respectively. The time period segmentation table, energy consumption decomposition features, and environmental impact factor set are integrated using a feature fusion matrix to generate a building energy consumption feature vector. The feature fusion matrix is ​​a weight matrix used to adjust the importance of different types of features in the final feature vector. The fusion process first standardizes all types of features, and then generates the final feature vector through a weighted combination. The formula for feature fusion is:

[0069] F = [W] T ·T,W D ·D,W E ·E]

[0070] Where F is the final feature vector, T is the time-period feature, D is the decomposed feature, E is the environmental feature, and W is the environmental feature. T W D W EThese are the corresponding weight coefficient matrices. The final generated building energy consumption feature vector is a high-dimensional vector that includes the temporal characteristics, decomposition characteristics, and environmental correlation characteristics of building energy consumption, providing comprehensive feature input for subsequent prediction models.

[0071] Taking a commercial building as an example, a weekly energy consumption dataset acquired through multi-dimensional sensors revealed missing data for the air conditioning system, specifically the four-hour data period from 10:00 AM to 2:00 PM on Tuesday. A sliding time window averaging method was used to fill the missing data, with a window size of 24 hours. The average values ​​for the same time periods on Monday and Wednesday (10:00 AM to 2:00 PM) were calculated and used to fill the missing data. An abnormal energy consumption peak was identified on Thursday afternoon, with a Z-score of 4.2, exceeding the set threshold of 3.0. This value was replaced with the median within the same time window. Time series analysis of the smoothed dataset determined the building's working hours to be 8:00 AM to 7:00 PM, transitional periods to 7:00 AM to 8:00 AM and 7:00 PM to 9:00 PM, and idle periods to 9:00 PM to 7:00 AM. Energy consumption data was decomposed using wavelet transform to extract a trend term reflecting long-term energy consumption growth, a seasonal term reflecting weekday / restday differences, and a residual term reflecting random disturbances. Simultaneously, environmental data showed an average indoor-outdoor temperature difference of 6.2℃, a thermal inertia index of 1.8 hours·℃ / ℃, a solar radiation impact coefficient of 0.72, an average humidity change rate of 1.4% / hour, and peak population density between 10:00 and 15:00. These features were integrated using a feature fusion matrix to generate a complete building energy consumption feature vector, characterizing the temporal variation patterns of building energy consumption and environmental influencing factors.

[0072] In one specific embodiment, the process of executing step S103 may specifically include the following steps:

[0073] The building energy consumption feature vector is segmented into time feature subvectors representing time characteristics and spatial feature vectors representing spatial distribution.

[0074] The time feature subvectors are processed by a bidirectional long short-term memory network. The hidden state is calculated through forward and backward time steps to capture long-term dependencies, extract the temporal variation pattern of building energy consumption, and form a time relationship feature matrix.

[0075] Graph convolution is performed on spatial feature subvectors to represent each area of ​​the building as a node in the graph and the physical connection between areas as an edge. The node feature update is calculated through a message passing mechanism to capture the mutual influence of energy consumption between different areas of the building and generate a spatial association feature graph.

[0076] The temporal relationship feature matrix and the spatial correlation feature map are fused through an attention mechanism. The weight coefficients of each feature channel are calculated and weighted together to construct a spatiotemporal fusion feature representation.

[0077] The spatiotemporal fusion feature representation is paired with historical building energy consumption data to form a training sample set. The parameters are iteratively optimized by the mini-batch gradient descent algorithm. The gradient is calculated using a loss function that combines the weighted average absolute percentage error and the root mean square error. The connection weights are updated, and L2 regularization is introduced to prevent overfitting, thus forming an energy consumption prediction calculation framework.

[0078] The dataset is divided into K subsets using the K-fold cross-validation method. K-1 subsets are used for training and 1 subset for validation in turn. The prediction stability and accuracy of the energy consumption prediction calculation framework under different data partitions are evaluated, and the model parameter configuration with the best validation performance is selected as the building energy consumption predictor.

[0079] Specifically, the building energy consumption feature vector is segmented into temporal feature sub-vectors representing temporal characteristics and spatial feature sub-vectors representing spatial distribution. This segmentation is based on the properties of the features. The temporal feature sub-vectors contain the temporal variation characteristics of energy consumption, such as the time series of historical energy consumption values, working / non-working day markers, time period markers within a day, and seasonal cycle characteristics. The spatial feature sub-vectors contain the spatial layout of the building and the characteristics of each area, such as the area, functional type, relative positional relationship, and equipment distribution of each area. Vector segmentation uses a feature attribute classification method, assigning each feature to the corresponding sub-vector according to its physical meaning, ensuring the integrity and independence of the information. The temporal feature sub-vectors are processed using a bidirectional long short-term memory (BiLSTM) network to capture the temporal variation patterns of building energy consumption. A BiLSTM network is a special type of recurrent neural network capable of learning sequence information from both past and future directions simultaneously. During processing, the temporal feature sub-vectors are used as the input sequence, and the network consists of forward LSTM layers and backward LSTM layers. The forward LSTM layer processes the sequence from beginning to end, capturing the information flow from the past to the present; the backward LSTM layer processes the sequence from end to beginning, capturing the information flow from the future to the present. Each LSTM layer in both directions computes its hidden state, and these hidden states are then concatenated to form a complete representation. Specifically, for each time point, the forward LSTM computes the forward hidden state, and the backward LSTM computes the backward hidden state; the final output at that time point is the concatenation of the two. Through this bidirectional processing, the network can comprehensively capture long-term dependencies, extract the periodic, trend, and sudden changes in building energy consumption, and form a temporal relationship feature matrix containing rich temporal information. Graph convolution operations on spatial feature vectors are used to capture the energy consumption interactions between different areas of the building. Graph Convolutional Networks (GCNs) are neural networks specifically designed for processing graph-structured data. In building energy consumption analysis, each functional area of ​​the building is represented as a node in a graph, with node features being the spatial feature vectors of that area; the physical connections between areas (such as adjacency, shared equipment, etc.) are represented as edges in the graph, constructing an adjacency matrix. Graph convolution operations update node features through a message-passing mechanism. Each node considers not only its own features but also information from its neighbors. In the specific computation process, for each node, feature updates are achieved by aggregating its own features and those of its neighbors. The stacking of multiple layers of graph convolutions allows each node to perceive a broader range of information, extending from directly adjacent nodes to the global relationships of the entire building. Through graph convolution operations, a spatial relational feature map is generated. This feature map not only contains the characteristics of each region itself but also encodes the propagation relationships of energy consumption effects between regions.

[0080] The fusion of temporal relationship feature matrix and spatial correlation feature map through an attention mechanism aims to comprehensively consider information from both temporal and spatial dimensions. The core idea of ​​the attention mechanism is to calculate the importance weights of different features and then perform a weighted combination based on these weights. In practice, the temporal relationship feature matrix and spatial correlation feature map are first converted into a unified form, and then their attention scores are calculated. The calculation process includes: calculating the query vector, key vector, and value vector for each feature channel; calculating the similarity between the query vector and the key vector to obtain the attention score; normalizing the attention scores to obtain weight coefficients; and finally, using the weight coefficients to perform a weighted summation of the value vectors. This approach adaptively adjusts the importance of temporal and spatial features in different contexts, resulting in a spatiotemporal fusion feature representation that retains both the dynamic changes of the time series and the structured knowledge of spatial layout.

[0081] Pairing spatiotemporal fusion feature representations with historical building energy consumption data to form a training sample set is the preparatory work for model training. Each training sample consists of input features (spatiotemporal fusion feature representations) and a target value (the actual energy consumption value at the corresponding time point). The training process uses a mini-batch gradient descent algorithm to iteratively optimize parameters. Each time, a small batch of samples (usually 64) is randomly selected from the training sample set, the error between the model's predicted value and the true value is calculated, and then the gradient is calculated and the model parameters are updated through backpropagation. The loss function adopts a combination of weighted average absolute percentage error and root mean square error. The former focuses on relative error and is suitable for energy consumption prediction of different magnitudes; the latter focuses on absolute error and is more sensitive to outliers. At the same time, an L2 regularization term is introduced to prevent the model from overfitting by adding a penalty term for the sum of squared parameters to the loss function. During the training process, the parameters are gradually adjusted according to the set learning rate until the loss function converges or the maximum number of training epochs is reached, forming the energy consumption prediction calculation framework.

[0082] K-fold cross-validation is used to evaluate model performance to ensure its generalization ability. K-fold cross-validation randomly divides the complete dataset into K subsets of similar size, typically K is set to 5 or 10. During validation, each time a subset is selected as the validation set, and the remaining K-1 subsets are used as the training set. The model is trained and its performance is evaluated on the validation set. This process is repeated K times, ensuring that each subset is used as the validation set once. Finally, the average performance metric across the K validations is calculated to obtain the model's overall performance under different data partitions. By comparing the cross-validation results under different parameter configurations, the model parameter configuration with the best validation performance is selected as the final building energy consumption predictor. This validation method can comprehensively evaluate the model's stability and adaptability, avoiding performance bias caused by specific data partitions.

[0083] Taking a commercial complex as an example, the building is divided into four functional areas: retail, office, dining, and public areas. Temporal feature vectors (containing 24-hour energy consumption variation curves, weekday markers, seasonal indicators, etc.) and spatial feature vectors (containing area, orientation, equipment configuration, etc.) are segmented from the building's energy consumption feature vector. The temporal feature vector is input into a three-layer BiLSTM network, with each layer containing 128, 256, and 128 units respectively, capturing hourly, daily, and weekly energy consumption variation patterns to form a feature matrix reflecting time dependencies. Simultaneously, the four functional areas are constructed as a graph structure, with the weights of edges between nodes determined based on the physical distance between areas and the energy system connectivity. A three-layer graph convolutional network is used to capture the propagation mechanism of energy consumption impact between areas, such as the temperature impact of air conditioning use in the office area on adjacent public areas. Temporal and spatial features are fused through an attention layer, which automatically learns patterns where spatial factors have higher weights during weekdays, while temporal factors have higher weights at night and on weekends. The fusion features were paired with actual energy consumption data, and trained using a mini-batch gradient descent algorithm with a batch size of 64. The weights of the root mean square error and mean absolute percentage error in the loss function were set to 0.3 and 0.7, respectively. Finally, the performance of different network configurations was evaluated using 10-fold cross-validation, and the parameter combination with the smallest validation error was selected as the final model, which can accurately predict energy consumption changes in different times and regions.

[0084] In one specific embodiment, the process of executing step S104 may specifically include the following steps:

[0085] Input the current building parameter data into the building energy consumption predictor, and obtain the expected energy consumption value at each time point through forward calculation to form the expected energy consumption sequence.

[0086] Collect actual building energy consumption data for the current time period, align it with the same time granularity as the expected energy consumption sequence, and obtain the actual energy consumption sequence;

[0087] Calculate the difference between the actual energy consumption sequence and the expected energy consumption sequence to generate an energy consumption residual sequence;

[0088] The energy consumption residual sequence is analyzed by a dynamic threshold detection algorithm. The threshold τ value at each time point is calculated, and data points whose absolute residual value exceeds the corresponding threshold are identified as outliers.

[0089] Perform continuity analysis on anomalies, mark intervals in which three or more anomalies occur consecutively in time as energy consumption anomaly intervals, and record the start time, duration, maximum deviation value and cumulative deviation amount of the anomaly;

[0090] Based on the characteristic parameters of the energy consumption anomaly intervals, the anomaly intervals are classified using a decision tree algorithm to identify the anomaly type, and a building energy consumption anomaly report is generated according to the anomaly location, anomaly type, severity, and suggested handling methods.

[0091] Specifically, current building parameter data is input into the building energy consumption predictor. Forward computation is used to obtain the expected energy consumption values ​​at each time point, forming an expected energy consumption sequence. Forward computation refers to passing the input data layer by layer from the input layer to the output layer through a trained neural network model to obtain the final prediction result. Specifically, the current building parameter data includes predicted values ​​for the current and future periods, such as temperature, humidity, occupancy density, and building usage plans. This data is processed in the same format as the training data and then input into the predictor. The predictor calculates through each layer of the network and finally outputs the expected energy consumption values ​​at each time point within a future period (e.g., 24 hours). These values ​​are arranged in chronological order to form the expected energy consumption sequence. Actual building energy consumption data for the current period is collected and aligned with the same time granularity as the expected energy consumption sequence to obtain the actual energy consumption sequence. The data collection process involves real-time recording of actual electricity, water, and gas consumption data through the metering equipment of the building energy management system. Since different energy metering devices may have different sampling frequencies, time alignment processing is required to unify all data to the same time granularity (e.g., 15 minutes, 30 minutes, or 1 hour). Alignment processing includes downsampling high-frequency data (e.g., averaging) or interpolating low-frequency data to ensure that the actual energy consumption sequence corresponds one-to-one with the expected energy consumption sequence at time points.

[0092] The difference between the actual energy consumption sequence and the expected energy consumption sequence is calculated to generate an energy consumption residual sequence. The residual calculation formula is: Residual = Actual Value - Expected Value. The residual value at a corresponding time point reflects the degree of deviation between the actual and expected energy consumption. A positive residual indicates that the actual energy consumption is higher than expected, which may indicate energy waste; a negative residual indicates that the actual energy consumption is lower than expected, which may indicate that the equipment is not operating normally or that its utilization rate is lower than expected. The residual sequence retains the time attribute of the original sequence, which is convenient for subsequent time series analysis.

[0093] The energy consumption residual sequence is analyzed using a dynamic threshold detection algorithm to calculate the threshold τ value at each time point. The threshold calculation formula is as follows:

[0094] τ(p)=μ p +k p ·σ p

[0095] Where τ(p) represents the dynamic threshold at time point p, μ p σ represents the mean of the residual sequence within a time window around time point p. p k represents the standard deviation within the window. pThis is an adaptive adjustment coefficient that automatically adjusts based on different time periods and building functional areas. Compared to traditional fixed thresholds, dynamic thresholds can adapt to the time-varying characteristics of energy consumption data, setting a more lenient threshold during periods of high fluctuation and a more stringent threshold during stable periods. When the absolute value of the residual |R(p)| exceeds the corresponding threshold k... p When an outlier occurs, the data point is identified as an anomaly. Continuity analysis is performed on the identified outliers, and intervals with three or more consecutive outliers in time are marked as energy consumption anomaly intervals. Continuity judgment is used to filter out occasional random fluctuations, focusing only on persistent anomalies. For each anomaly interval, several characteristic parameters are recorded: anomaly start time (timestamp of the first outlier), duration (time from the first to the last outlier), maximum deviation (maximum absolute value of residuals within the interval), and cumulative deviation (sum of absolute values ​​of all residuals within the interval). These characteristic parameters comprehensively describe the characteristics of the anomaly interval, providing a basis for subsequent anomaly classification.

[0096] Based on the characteristic parameters of energy consumption anomaly intervals, a decision tree algorithm is used to classify these intervals and identify the anomaly types. The decision tree algorithm is a typical classification method that divides anomaly intervals into different categories by constructing a tree structure. The construction of the decision tree is based on expert knowledge and historical anomaly cases. Each internal node of the tree represents a test for a certain feature (e.g., "whether the duration exceeds 2 hours"), each branch represents the possible outcome of the test (yes / no), and each leaf node represents the classification result (anomaly type). By continuously segmenting the features, the decision tree categorizes anomaly intervals into different types such as equipment failure, control parameter deviation, abnormal energy consumption behavior, or external environmental influences. Each type can be further subdivided into multiple subcategories. The classification results, along with information such as the location, severity, and suggested handling methods of the anomaly intervals, form a complete building energy consumption anomaly report to guide subsequent energy consumption optimization.

[0097] Taking the air conditioning system of an office building as an example, on a summer workday, the day's temperature forecast data, expected staff attendance rate, and meeting schedule are input into the building energy consumption predictor to obtain the hourly expected air conditioning energy consumption value, forming an expected energy consumption sequence. Simultaneously, the actual hourly air conditioning power consumption is recorded in real time through the building energy management system, forming the actual energy consumption sequence. Calculating the difference between the two sequences reveals that the residual value is consistently positive and large between 10:00 AM and 3:00 PM, indicating that the actual energy consumption is significantly higher than the expected value. A dynamic threshold detection algorithm is applied to the residual sequence. First, the mean and standard deviation for different time periods are calculated. For example, the mean residual for the 10:00-11:00 AM period is 2.1 kWh, and the standard deviation is 0.8 kWh. Based on empirical values, an adjustment coefficient of 2.5 is set, resulting in a dynamic threshold of 4.1 kWh for this period. Time points where the residual value exceeds the threshold are marked as outliers. Continuous analysis reveals 18 consecutive outliers from 10:15 AM to 2:45 PM, forming an energy consumption anomaly interval. The system records characteristic parameters such as the start time (10:15), duration (4.5 hours), maximum deviation (6.2 kWh), and cumulative deviation (82.3 kWh) for this interval. These characteristics are analyzed using a decision tree algorithm. The algorithm first determines the duration of the anomaly (>4 hours, proceeds to the left branch), then determines if it coincides with the outdoor temperature peak (yes, proceeds to the right branch), then determines the building area (central area, proceeds to the middle branch), and finally classifies the anomaly as "excessive cooling of air conditioning due to an excessively low temperature setpoint." An anomaly report is generated, including the anomaly location, type, severity, and suggested handling method (adjusting the temperature setpoint upwards by 2°C), providing building managers with accurate energy consumption anomaly diagnosis and handling suggestions.

[0098] In one specific embodiment, the process of executing step S105 may specifically include the following steps:

[0099] Extract anomaly types and energy consumption deviation data from building energy consumption anomaly reports to construct an energy consumption optimization target list;

[0100] Collect building usage plan data, including work schedules, occupancy plans, and special event times, to create a building usage timeline;

[0101] Obtain forecasts of future environmental conditions, including temperature change trends, humidity distribution, and light intensity, and generate prediction sequences of environmental impact factors;

[0102] A multi-objective optimization function is constructed, which includes three dimensions: energy consumption index, economic cost index, and comfort index. Weight coefficients are set for each index, and the optimal parameter combination is solved by genetic algorithm to obtain the equipment operation parameter adjustment scheme.

[0103] The equipment operating parameter adjustment plan is decomposed into equipment-level control parameters, including air conditioning temperature setpoint, fresh air ratio adjustment parameters, lighting intensity parameters, and equipment start-up and shutdown time, and an equipment control parameter table is generated.

[0104] The equipment control parameter table is converted into an instruction format that conforms to the building control system communication protocol to form building energy-saving control instructions.

[0105] Specifically, anomaly types and energy consumption deviation data are extracted from building energy consumption anomaly reports to construct an energy consumption optimization target list. Anomaly types typically include four main categories: equipment failure, control parameter deviation, abnormal energy consumption behavior, and external environmental influences. Each type is further subdivided into multiple subcategories, such as excessively low air conditioning temperature setpoints and excessively high lighting intensity under the control parameter deviation category. Energy consumption deviation data represents the difference between actual and expected energy consumption within the anomaly range, including indicators such as deviation amount, deviation rate, and duration. The energy consumption optimization target list is prioritized based on anomaly type, energy consumption deviation amount, duration, and frequency of occurrence, prioritizing anomalies with large deviations, long durations, and frequent occurrences. Collecting building usage plan data to create a building usage timeline ensures that energy-saving optimization does not affect the building's normal functionality. Building usage plan data originates from the building management system's scheduling information, including work hours (e.g., office hours 9:00-18:00), personnel occupancy plans (e.g., expected number and distribution in each area), and special event times (e.g., meetings, exhibitions, and other temporary events). This data is organized chronologically to form a building usage timeline that covers a future period (usually 7 days). This timeline marks the building's usage status at different times, providing constraints for energy optimization and ensuring energy-saving adjustments are made while meeting usage needs.

[0106] Obtaining future environmental condition forecasts and generating a predicted sequence of environmental impact factors is a crucial step in considering the impact of the external environment on building energy consumption. Future environmental condition forecasts primarily come from meteorological forecast data, including temperature change trends (such as hourly temperature forecasts for the next few days), humidity distribution (such as relative humidity change curves), and light intensity (such as sunshine duration and intensity). This raw meteorological data is processed and converted into environmental impact factors directly related to building energy consumption, such as temperature difference factors (indoor-outdoor temperature difference), humidity factors (indoor-outdoor humidity difference), and light factors (natural daylight availability). The predicted sequence of environmental impact factors is arranged at the same time granularity as the building's usage timeline, providing environmental background information for energy optimization algorithms. Constructing a multi-objective optimization function is the core of energy optimization. The multi-objective optimization function includes three dimensions: energy consumption indicators, economic cost indicators, and comfort indicators, which are comprehensively considered through a weighted summation. The mathematical expression of this optimization function is:

[0107] F=γ1·Econs +γ2·C econ +γ3·D comf

[0108] Where F is the comprehensive optimization objective value, which needs to be minimized; E cons It is an energy consumption indicator, representing the amount of energy used per unit of time; C econ It is an economic cost indicator, taking into account electricity price differences and equipment operating costs at different times; D comf These are comfort indicators, quantifying the deviation of indoor environmental parameters (temperature, humidity, light, etc.) from comfort standards. γ1, γ2, and γ3 are weighting coefficients of the three indicators, satisfying γ1 + γ2 + γ3 = 1, dynamically set according to building type and management needs. A genetic algorithm is used to solve a multi-objective optimization problem, resulting in an equipment operating parameter adjustment scheme. The genetic algorithm is an optimization algorithm inspired by biological evolution, searching for the optimal solution by simulating natural selection and genetic mechanisms. The algorithm flow includes: initializing the population (randomly generating multiple sets of equipment parameter combinations); evaluating fitness (calculating the score of each set of parameters under the optimization function); selection operation (retaining individuals with high fitness); crossover operation (exchanging some parameters between two individuals to generate new individuals); mutation operation (randomly changing some parameters of individuals to increase diversity); and termination condition judgment (stopping when the maximum number of generations is reached or the convergence condition is met, otherwise returning to the fitness evaluation step). After multiple generations of evolution, the algorithm eventually converges to a solution close to the global optimum, forming the equipment operating parameter adjustment scheme. The equipment operating parameter adjustment scheme is decomposed into equipment-level control parameters, generating an equipment control parameter table. The equipment operating parameter adjustment plan is a holistic optimization result that needs to be converted into specific, directly controllable equipment parameters. These control parameters include air conditioning temperature setpoints (e.g., setting the return air temperature of a central air conditioning system to 24℃), fresh air ratio adjustment parameters (e.g., setting the fresh air valve opening to 30%), lighting intensity parameters (e.g., setting the brightness of a dimming system to 80%), and equipment start-up and shutdown times (e.g., advancing the pre-cooling start time of the air conditioning system to 8:00 AM). Each control parameter has a corresponding value range and adjustment step size, and the overall optimization goal must be achieved while meeting equipment operating constraints. The equipment control parameters are organized according to equipment type, control point ID, parameter name, parameter value, and execution time to form an equipment control parameter table.

[0109] The equipment control parameter table is converted into an instruction format conforming to the building control system's communication protocol, forming building energy-saving control instructions. Different building control systems may use different communication protocols, such as BACnet, Modbus, or LonWorks. The conversion process requires mapping each control parameter to a corresponding protocol command according to the target system's protocol specifications, including fields such as device address, function code, data area, and checksum. The building energy-saving control instructions are the final executable set of commands, which are distributed to various controllers and execution devices through the building automation system's communication network to achieve optimized control of building energy consumption.

[0110] Taking a commercial office building as an example, an anomaly of excessively low air conditioning system temperature setpoint was extracted from energy consumption anomaly reports. During the weekday period of 9:00-18:00, actual energy consumption was 15% higher than expected. An optimization list was constructed with adjusting the air conditioning temperature setpoint as the primary objective. Data collection of the building's usage plan revealed that two important meetings were scheduled for the following morning from 10:00-12:00, resulting in a 30% higher personnel density than usual. The afternoon session, from 14:00-17:00, was designated as normal office work. Simultaneously, weather forecasts indicated that the temperature would gradually rise the following day, from 22℃ in the morning to 30℃ in the afternoon, with relative humidity remaining stable at around 65%. Based on this information, a multi-objective optimization function was constructed, setting the weights for energy consumption (0.5), economic cost (0.3), and comfort (0.2). Through iterative calculations using a genetic algorithm (population size 100, iterations 50), a differentiated air conditioning temperature setting scheme was obtained: 24℃ during meeting hours, 26℃ during regular office hours, and 28℃ during non-office hours. Simultaneously, the fresh air ratio was increased to 40% during meeting hours and maintained at 25% during other times. These parameters were decomposed into specific air conditioning system control point parameters, including temperature setpoints for each zone, air supply volume regulating valve openings, and fresh air valve positions. Finally, these parameters were converted into control instructions according to the BACnet protocol format used in the building. These instructions included device ID, attribute ID, value type, adjustment value, and execution time, forming a building energy-saving control instruction set, ready to be executed sequentially according to the planned time, achieving refined control that meets usage needs while saving energy.

[0111] In one specific embodiment, the process of performing step S106 may specifically include the following steps:

[0112] Send building energy conservation control commands to the building control system, initiate the equipment parameter adjustment process, and record the command execution timestamp;

[0113] The building's operational status after adjustments is monitored in real time using multi-dimensional sensors. Data on building temperature, humidity, energy consumption, and personnel density after adjustments are collected to form an energy consumption monitoring dataset after adjustments.

[0114] The adjusted energy consumption monitoring dataset is compared with the historical data of the same period before the adjustment in terms of time alignment. The energy consumption difference under the same conditions is calculated and an energy consumption change data table is generated.

[0115] Statistical analysis was performed on the data in the energy consumption change data table to calculate three energy efficiency improvement indicators: average energy consumption reduction rate, peak energy consumption reduction amount, and load balance degree, and an energy-saving effect assessment report was constructed.

[0116] The key parameters in the energy-saving effect assessment report are integrated with the building energy consumption feature vector to supplement new time series data points and update the building energy consumption feature library.

[0117] The building energy consumption predictor is incrementally trained by updating the building energy consumption feature library to improve prediction accuracy and form a closed-loop optimized building energy consumption analysis system.

[0118] Specifically, building energy-saving control commands are issued to the building control system, initiating the equipment parameter adjustment process and recording the command execution timestamp. This process transmits the previously generated control commands to various levels of controllers, such as air conditioning main unit controllers, lighting control modules, and fan inverters, through the building automation network. Each command includes information such as the controlled object identifier, parameter type, parameter value, and execution time. When the control command is received and executed by the equipment, the system records the execution timestamp, forming a command execution log, recording the command issuance time, reception confirmation time, and actual effective time of the parameters. This timestamp information provides accurate time reference points for subsequent effect evaluation. The building's operational status after the adjustment is monitored in real time using multi-dimensional sensors, collecting data on the adjusted building temperature, humidity, energy consumption, and personnel density to form a post-adjustment energy consumption monitoring dataset. These multi-dimensional sensors are the same as those used in the previous data acquisition phase, including temperature sensors, humidity sensors, power monitoring devices, and personnel density sensors distributed in various functional areas of the building. The monitoring process continuously records the real-time changes of various parameters. The data acquisition frequency is set according to the parameter characteristics, generally 5 minutes / time for temperature and humidity, 15 minutes / time for energy consumption, and 10 minutes / time for personnel density. All collected data is accurately timestamped and stored in a standardized data format, forming an adjusted energy consumption monitoring dataset. This dataset includes not only energy consumption data but also environmental and usage parameters that affect energy consumption, providing a data foundation for a comprehensive assessment of energy-saving effects.

[0119] The adjusted energy consumption monitoring dataset is compared with historical data from the same period before the adjustment using time alignment. The energy consumption difference under identical conditions is calculated, and an energy consumption change data table is generated. Time alignment comparison refers to selecting historical data from the same period before the adjustment (e.g., the same workday last week) as a benchmark and comparing it with the adjusted data. To ensure the accuracy of the comparison, external factors affecting energy consumption, such as weather conditions and building usage, need to be considered. A similar day screening method is used in data processing. Based on indicators such as temperature similarity, humidity similarity, and population density similarity, data closest to the current conditions is selected from historical data as the comparison benchmark. The energy consumption difference is calculated using the formula: Difference value = Adjusted energy consumption - Unadjusted energy consumption, Difference rate = (Adjusted energy consumption - Unadjusted energy consumption) / Unadjusted energy consumption. The calculation results are organized according to dimensions such as equipment type, functional area, and time period to generate an energy consumption change data table, intuitively displaying the actual effect of energy-saving measures. Statistical analysis was performed on the data in the energy consumption change data table to calculate three energy efficiency improvement indicators: average energy consumption reduction rate, peak energy consumption reduction, and load balance. An energy-saving effect assessment report was then constructed. The average energy consumption reduction rate refers to the average percentage reduction in energy consumption over the entire assessment period. It is calculated by weighting the energy consumption reduction rates for each time period (the weight is the time period length). Peak energy consumption reduction refers to the reduction in energy consumption during peak energy consumption periods, focusing on the energy-saving effect during peak energy consumption periods (such as peak air conditioning electricity consumption in summer). Load balance is an indicator that measures the uniformity of energy consumption distribution. It is calculated as the ratio of the standard deviation of energy consumption to the average value. The smaller this value, the more uniform the energy consumption distribution, which is beneficial for reducing equipment capacity and improving energy utilization efficiency. These three indicators evaluate the energy-saving effect from different perspectives, comprehensively forming an energy-saving effect assessment report, including the overall energy saving rate, the energy-saving effect of each item, the energy-saving situation in each region, and economic benefit analysis.

[0120] The key parameters in the energy-saving performance evaluation report are integrated with the building energy consumption feature vector to supplement new time-series data points and update the building energy consumption feature library. The building energy consumption feature library is a structured storage of building energy consumption data, containing historical energy consumption data, environmental parameter data, usage parameter data, and the relationships between them. The update process first extracts key parameters from the energy-saving performance evaluation report, such as the actual energy consumption values ​​of each region under different control parameters and energy response characteristics (e.g., the relationship between temperature adjustment range and energy consumption changes). Then, these parameters are integrated with the existing building energy consumption feature vector to generate new feature vectors, which are then added to the feature library as new data points. The feature library update adopts an incremental update method, retaining the original data while adding new data points, enriching the diversity of the data samples, and improving the representativeness and timeliness of the feature library.

[0121] Incremental training of the building energy consumption predictor, using an updated building energy consumption feature library, improves prediction accuracy and forms a closed-loop optimized building energy consumption analysis system. Incremental training refers to fine-tuning parameters using new data on the basis of the original model, without retraining the entire model. During incremental training, new data is first extracted from the updated feature library to construct a training set; then, using the current predictor model as the initial state, a small learning rate (usually 1 / 10 of the original learning rate) is set, and the parameters are updated using the new training set; finally, the prediction accuracy of the updated model is evaluated. If it is better than the original model, it is replaced; otherwise, the original model is retained. Incremental training allows the model to gradually adapt to dynamic changes in building energy consumption characteristics, such as seasonal changes, equipment aging, and adjustments in usage patterns, continuously improving prediction accuracy. In this way, building energy consumption analysis forms a complete closed-loop optimization system: from data collection to anomaly detection, from optimization strategy to execution evaluation, and then to model updates, each link is interconnected and continuously optimized, achieving intelligent and refined building energy consumption management.

[0122] Taking the optimization of the air conditioning system in an educational building as an example, the generated energy-saving control commands were first sent to the air conditioning automatic control system via the BACnet protocol. The temperature setpoints for different classrooms were adjusted (from a uniform 24℃ to differentiated settings based on actual usage, such as 25℃ for large classrooms and 26℃ for small classrooms). Simultaneously, the start-stop time strategy was modified (starting 30 minutes earlier based on the class schedule instead of a fixed time). All command execution was recorded with precise timestamps. Temperature sensors recorded the actual temperature change curves for each classroom after the adjustment, the power monitoring system collected the power consumption of the air conditioning system, and the personnel density sensor recorded the actual usage of each classroom. These data collectively formed a week's energy consumption monitoring dataset after the adjustment. This data was compared with historical data from the previous week with the same class schedule. Considering that the average temperature this week was 2℃ higher than last week, the baseline energy consumption was adjusted using a temperature correction formula. Finally, the energy consumption differences of the air conditioning system in each classroom at different times were calculated. Statistical analysis shows that the differentiated temperature setting strategy reduced average energy consumption by 8.2 kWh / day, peak load during peak hours (12:00-14:00) by 12 kW, and improved system load balance from 0.42 to 0.31, indicating a more uniform energy distribution. These key parameters, fused with the existing building energy consumption feature vector, were added to the building energy consumption feature library, specifically recording the actual energy consumption performance under different temperature setting strategies. Finally, incremental training of the building energy consumption predictor was performed using this newly added sample data, enabling the predictor to more accurately predict the energy consumption performance of the differentiated temperature strategy.

[0123] The above describes the building energy consumption analysis method based on artificial intelligence in the embodiments of this application. The following describes the building energy consumption analysis system based on artificial intelligence in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the building energy consumption analysis system based on artificial intelligence in this application includes:

[0124] The data acquisition module is used to collect data on building temperature, humidity, energy consumption and personnel density through multi-dimensional sensors to obtain a basic dataset of building energy consumption.

[0125] The extraction module is used to perform data missing imputation and outlier processing based on the building energy consumption basic dataset, extract building usage time period, environmental parameters and energy load characteristics, and generate building energy consumption feature vector;

[0126] A module is established to build a building energy consumption prediction model that considers time factors and spatial relationships based on the building energy consumption feature vector, and to obtain a building energy consumption predictor through training with historical data.

[0127] The comparison module is used to calculate the expected energy consumption value using the building energy consumption predictor, compare the actual energy consumption data with the expected energy consumption value, identify abnormal energy consumption intervals, and generate a building energy consumption anomaly report.

[0128] The generation module is used to formulate an energy optimization parameter adjustment scheme based on the building energy consumption anomaly report, combined with the building usage plan and environmental conditions, and generate building energy-saving control instructions;

[0129] The update module is used to execute the building energy-saving control command, record the energy consumption change data before and after the adjustment, calculate the energy efficiency improvement index, update the building energy consumption feature library, and complete the building energy consumption analysis closed loop.

[0130] Through the collaborative efforts of the aforementioned components, a comprehensive building energy consumption dataset was obtained by collecting data on building temperature, humidity, energy consumption, and occupancy density using multi-dimensional sensors. This provides a rich data foundation for subsequent analysis and solves the problem of single data sources in traditional methods. By imputing missing data and handling outliers in the building energy consumption dataset, and extracting building usage periods, environmental parameters, and energy load characteristics, a building energy consumption feature vector was generated. This significantly improved data quality and feature representation capabilities, providing high-quality input for model training. Based on the building energy consumption feature vector, a building energy consumption prediction model considering time factors and spatial relationships was established. This model fully captures the spatiotemporal correlation of building energy consumption, overcomes the limitation of traditional models that ignore the mutual influence of energy consumption between regions, and improves prediction accuracy. The building energy consumption predictor calculates the expected energy consumption value, compares the actual energy consumption data with the expected energy consumption value, identifies abnormal energy consumption intervals, and generates a building energy consumption anomaly report. This realizes a shift from passive discovery to proactive early warning, greatly shortening the anomaly detection time. Based on building energy consumption anomaly reports, combined with building usage plans and environmental conditions, a targeted energy optimization parameter adjustment scheme was developed, generating building energy-saving control instructions. This achieved precise energy saving, avoiding the comfort degradation caused by the "one-size-fits-all" approach of traditional methods. By executing building energy-saving control instructions, recording energy consumption changes before and after adjustments, calculating energy efficiency improvement indicators, and updating the building energy consumption feature database, a closed-loop building energy consumption analysis was completed, achieving self-learning and continuous optimization. This scheme fully leverages the advantages of artificial intelligence algorithms in specific building energy consumption management fields, particularly in spatiotemporal data processing. It captures the temporal dependencies of energy consumption through a bidirectional long short-term memory network, models the spatial relationships between building areas through a graph convolutional network, and dynamically adjusts the importance weights of different features through an attention mechanism. These algorithmic features contribute to the scheme by adaptively handling the dynamic changes in building energy consumption and identifying complex energy consumption patterns, thereby achieving more accurate predictions and more refined control. Furthermore, a genetic algorithm is used to solve a multi-objective optimization problem, balancing energy saving, cost, and comfort dimensions to achieve a globally optimal energy management strategy. Incremental learning mechanisms enable models to continuously learn from new data, improve prediction accuracy, and adapt to the long-term evolution of building energy consumption characteristics.

[0131] Reference Figure 3 This invention also provides a computer device, which can be a server, and its internal structure can be as follows: Figure 3As shown, the computer device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores the data corresponding to this embodiment. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0132] Those skilled in the art will understand that Figure 3 The structures shown are merely block diagrams of some structures related to the present invention and do not constitute a limitation on the computer devices on which the present invention is applied.

[0133] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0134] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the present invention and embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0135] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0136] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0137] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A building energy consumption analysis method based on artificial intelligence, characterized in that, include: By collecting data on building temperature, humidity, energy consumption, and occupancy density using multi-dimensional sensors, a basic dataset of building energy consumption can be obtained. Missing values ​​in the basic building energy consumption dataset are imputed using a sliding time window averaging method to obtain a preliminary imputed dataset. Outliers in the preliminary imputed dataset are identified using an improved Z-score method, and these outliers are replaced with the median of the preceding and following data points to generate a smoothed dataset. The smoothed dataset is then segmented into time series segments based on the building energy load variation characteristics, dividing it into working periods, transition periods, and idle periods to form a time period division table. Wavelet transform is used to denoise the smoothed dataset, separating the trend, seasonal, and residual terms of building energy consumption to construct energy consumption decomposition features. Five environmental parameters—temperature difference, thermal inertia index, solar radiation impact coefficient, humidity change rate, and population density distribution—are extracted from the smoothed dataset to generate an environmental impact factor set. The time period division table, energy consumption decomposition features, and environmental impact factor set are integrated using a feature fusion matrix to generate a building energy consumption feature vector, improving data quality and feature representation capabilities, and providing high-quality input for model training. Based on the building energy consumption feature vector, a building energy consumption prediction model considering time factors and spatial relationships is established, and a building energy consumption predictor is obtained by training with historical data. The building energy consumption predictor is used to calculate the expected energy consumption value. The actual energy consumption data is compared with the expected energy consumption value to identify abnormal energy consumption intervals and generate a building energy consumption anomaly report. Based on the building energy consumption anomaly report, combined with the building usage plan and environmental conditions, formulate an energy optimization parameter adjustment plan and generate building energy-saving control instructions; The building energy conservation control command is issued to the building control system to initiate the equipment parameter adjustment process and record the command execution timestamp. Real-time monitoring of the building's operational status after adjustment is conducted using multi-dimensional sensors, collecting data on building temperature, humidity, energy consumption, and occupancy density to form a post-adjustment energy consumption monitoring dataset. This dataset is then compared with historical data from the same period before adjustment to calculate energy consumption differences under identical conditions, generating an energy consumption change data table. Statistical analysis is performed on the data in the energy consumption change data table to calculate three energy efficiency improvement indicators: average energy consumption reduction rate, peak energy consumption reduction, and load balance, constructing an energy conservation effect assessment report. Key parameters from the energy conservation effect assessment report are integrated with the building energy consumption feature vector, supplementing new time-series data points and updating the building energy consumption feature library. The updated building energy consumption feature library is used to incrementally train the building energy consumption predictor, improving prediction accuracy and forming a closed-loop optimized building energy consumption analysis system.

2. The building energy consumption analysis method based on artificial intelligence according to claim 1, characterized in that, A building energy consumption basic dataset is obtained by collecting building temperature, humidity, energy consumption, and personnel density data through multi-dimensional sensors. This includes: collecting temperature change data for each area through a temperature sensor network arranged according to building functional zones to form a temperature spatiotemporal data matrix; collecting humidity distribution data within the building through multi-point humidity sensors and combining it with ventilation status parameters to generate a humidity environment feature set; collecting sub-item energy consumption data for major energy-consuming equipment in the building through energy metering devices, recording equipment operating power curves, and constructing an equipment energy consumption database; collecting personnel flow and density data within the building through personnel detection devices to establish a personnel activity pattern feature table; performing time synchronization processing on the temperature spatiotemporal data matrix, humidity environment feature set, equipment energy consumption database, and personnel activity pattern feature table, removing invalid data points, supplementing missing time periods, and generating a building energy consumption basic dataset; and normalizing various types of data in the building energy consumption basic dataset to eliminate dimensional differences, thus obtaining the building energy consumption basic dataset.

3. The building energy consumption analysis method based on artificial intelligence according to claim 1, characterized in that, Based on building energy consumption feature vectors, a building energy consumption prediction model considering time factors and spatial relationships is established. Trained using historical data, a building energy consumption predictor is obtained, comprising: splitting the building energy consumption feature vectors into time feature sub-vectors representing time characteristics and spatial feature sub-vectors representing spatial distribution; processing the time feature sub-vectors using a bidirectional long short-term memory network, calculating hidden states through forward and backward time steps to capture long-term dependencies, extracting the temporal variation patterns of building energy consumption, and forming a time relationship feature matrix; and performing graph convolution operations on the spatial feature sub-vectors, representing each area of ​​the building as nodes in the graph, and the physical connections between areas as edges, calculating node feature updates through a message passing mechanism to capture the mutual influence relationships of energy consumption between different areas of the building, and generating a spatial correlation feature graph. The temporal relationship feature matrix and spatial correlation feature map are fused using an attention mechanism. The weight coefficients of each feature channel are calculated and weighted to construct a spatiotemporal fusion feature representation. The spatiotemporal fusion feature representation is paired with historical building energy consumption data to form a training sample set. The parameters are iteratively optimized using a mini-batch gradient descent algorithm. The gradient is calculated using a loss function combining weighted average absolute percentage error and root mean square error to update the connection weights. At the same time, L2 regularization is introduced to prevent overfitting, forming an energy consumption prediction calculation framework. The dataset is divided into K subsets using the K-fold cross-validation method. K-1 subsets are used for training and 1 subset is used for validation in turn to evaluate the prediction stability and accuracy of the energy consumption prediction calculation framework under different data partitions. The model parameter configuration with the best validation performance is selected as the building energy consumption predictor.

4. The building energy consumption analysis method based on artificial intelligence according to claim 1, characterized in that, The process involves: calculating expected energy consumption values ​​using a building energy consumption predictor; comparing actual energy consumption data with expected energy consumption values ​​to identify abnormal energy consumption intervals and generate a building energy consumption anomaly report. This includes: inputting current building parameter data into the building energy consumption predictor; obtaining expected energy consumption values ​​at each time point through forward computation to form an expected energy consumption sequence; collecting actual building energy consumption data for the current period and aligning it with the expected energy consumption sequence at the same time granularity to obtain the actual energy consumption sequence; calculating the difference between the actual energy consumption sequence and the expected energy consumption sequence to generate an energy consumption residual sequence; analyzing the energy consumption residual sequence using a dynamic threshold detection algorithm, calculating the threshold τ value for each time point, and identifying data points whose absolute residual value exceeds the corresponding threshold as anomalies; performing continuity analysis on the anomalies, marking intervals with three or more consecutive anomalies as energy consumption anomaly intervals, and recording the anomaly start time, duration, maximum deviation value, and cumulative deviation; classifying the anomaly intervals based on their characteristic parameters and using a decision tree algorithm, identifying the anomaly type, and generating a building energy consumption anomaly report according to the anomaly location, anomaly type, severity, and suggested handling methods.

5. The building energy consumption analysis method based on artificial intelligence according to claim 1, characterized in that, Based on building energy consumption anomaly reports, combined with building usage plans and environmental conditions, an energy optimization parameter adjustment scheme is formulated, and building energy-saving control instructions are generated. This includes: extracting anomaly types and energy consumption deviation data from the building energy consumption anomaly reports to construct an energy consumption optimization target list; collecting building usage plan data, including work schedules, occupancy plans, and special event times, to form a building usage time series table; obtaining future environmental condition forecasts, including temperature change trends, humidity distribution, and light intensity, to generate an environmental impact factor prediction sequence; constructing a multi-objective optimization function, including three dimensions: energy consumption indicators, economic cost indicators, and comfort indicators, setting weight coefficients for each indicator, and solving for the optimal parameter combination using a genetic algorithm to obtain an equipment operation parameter adjustment scheme; decomposing the equipment operation parameter adjustment scheme into equipment-level control parameters, including air conditioning temperature setpoints, fresh air ratio adjustment parameters, lighting intensity parameters, and equipment start-up and shutdown times, to generate an equipment control parameter table; and converting the equipment control parameter table into an instruction format conforming to the building control system communication protocol to form building energy-saving control instructions.

6. An artificial intelligence-based building energy consumption analysis system, used to implement the artificial intelligence-based building energy consumption analysis method as described in any one of claims 1-5, characterized in that, include: The data acquisition module is used to collect data on building temperature, humidity, energy consumption and personnel density through multi-dimensional sensors to obtain a basic dataset of building energy consumption. The extraction module is used to perform data missing imputation and outlier handling based on the building energy consumption basic dataset, extract building usage time period, environmental parameters and energy load characteristics, and generate building energy consumption feature vectors. A module is established to build a building energy consumption prediction model that considers time factors and spatial relationships based on the building energy consumption feature vector. The model is trained using historical data to obtain a building energy consumption predictor. The comparison module is used to calculate the expected energy consumption value using the building energy consumption predictor, compare the actual energy consumption data with the expected energy consumption value, identify abnormal energy consumption intervals, and generate a building energy consumption anomaly report. The generation module is used to formulate energy optimization parameter adjustment schemes based on building energy consumption anomaly reports, combined with building usage plans and environmental conditions, and generate building energy-saving control instructions; The update module is used to execute building energy-saving control commands, record energy consumption changes before and after adjustments, calculate energy efficiency improvement indicators, update the building energy consumption feature library, and complete the building energy consumption analysis closed loop.

7. A computer device, characterized in that, It includes a memory and a processor, the memory storing a computer program that can run on the processor, and the processor executing the computer program to implement the artificial intelligence-based building energy consumption analysis method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores a computer program, which, when run by a processor, causes the processor to execute the artificial intelligence-based building energy consumption analysis method as described in any one of claims 1 to 5.

Images