Intelligent building energy-saving control system based on AI energy consumption prediction

Abstract

The invention provides an intelligent building energy-saving control system based on AI energy consumption prediction, and the system comprises a data construction module which is used for designing a sensor network and a building information model, and collecting a standardized multi-dimensional data set; the behavior recognition module is used for analyzing a user moving track and a space behavior mode, constructing a three-dimensional thermodynamic diagram and predicting a people number density sequence; the energy consumption prediction module is used for predicting short-term energy consumption, medium-term energy consumption and long-term energy consumption; the energy-saving control module is used for designing a target function and obtaining a control strategy set by using an NSGA-III algorithm and a digital twinborn simulation platform; the distributed execution module is used for performing control adjustment and real-time feedback on the building equipment by utilizing an edge controller; and the maintenance management module is used for visual interaction, control strategy execution and building equipment maintenance management. According to the invention, a personalized energy-saving strategy, active control adjustment, multi-dimensional energy consumption prediction and intelligent optimization can be realized, and refined and informatization management and maintenance of energy-saving control can be realized.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of intelligent buildings, in particular to an intelligent building energy-saving control system based on AI energy consumption prediction. BACKGROUND

[0002] Building energy-saving control refers to precisely regulating and controlling the equipment in the building, such as air conditioners, lighting, elevators and the like, through an intelligent system to reduce energy waste. It is not only an important embodiment of the intelligent management of smart buildings, but also a direct realization path of the energy-saving goal of green buildings. In the future, the development trend of buildings will inevitably be the deep integration of "green" and "smart", and building energy-saving control technology will play an increasingly important role in it.

[0003] The intelligent building energy-saving control technology reduces energy consumption by actively intervening in the operation of building equipment. The existing methods have formed a multi-dimensional technical system, but there are still problems such as precision and adaptability in actual application, specifically as follows: 1) Low system integration: information islands among multiple systems, limited data interconnection, and insufficient collaborative control capability, resulting in that the energy-saving potential is not fully tapped. 2) Limited intelligence: most places still mainly rely on preset logic and single-point automatic / semi-automatic control, lack self-learning and adaptive functions, and are difficult to handle complex and changing actual scenarios. 3) Contradiction between user comfort and energy saving: excessive pursuit of energy saving sacrifices the comfort of part of the space or causes user dissatisfaction. 4) Lack of prediction and active adjustment capability: many systems rely on fixed value control or simple feedback adjustment, and do not effectively use meteorological prediction, occupancy, behavior patterns and other feedforward information. SUMMARY

[0004] In order to overcome the deficiencies of the prior art, the purpose of the present application is to provide an intelligent building energy-saving control system based on AI energy consumption prediction, which can realize personalized energy-saving strategies and active control adjustment, multi-dimensional energy consumption prediction and intelligent optimization, realize fine and information management and maintenance of energy-saving control, and promote the sustainable development of smart buildings and green buildings.

[0005] To achieve the above purpose, the present application provides the following scheme: an intelligent building energy-saving control system based on AI energy consumption prediction, comprising:

[0006] A data construction module is used for designing a sensor network and a building information model according to a building structure to perform unified data collection and preliminary processing, and obtain a standardized multi-dimensional data set;

[0007] A behavior recognition module is used for extracting spatio-temporal behavior-related data and spatio-temporal behavior features in the standardized multi-dimensional data set to analyze user moving tracks and spatial behavior patterns, construct a three-dimensional heat map, and predict a number density sequence of each space within the next 24 hours;

[0008] an energy consumption prediction module, configured to perform feature extraction and factor analysis based on the standardized multi-dimensional data set, the three-dimensional thermal diagram and the crowd density sequence to predict short-term energy consumption, medium-term energy consumption and long-term energy consumption, and obtain multi-level energy consumption prediction results;

[0009] an energy saving control module, configured to design a target function based on the multi-level energy consumption prediction results, the three-dimensional thermal diagram and the standardized multi-dimensional data set, and obtain a control strategy set by using an NSGA-III algorithm and a digital twin simulation platform;

[0010] a distributed execution module, configured to control and adjust building equipment by using an edge controller according to the control strategy set and perform real-time feedback;

[0011] a maintenance management module, configured to design a visual interactive interface and perform maintenance management on the building equipment according to an execution result of the control strategy set;

[0012] The data construction module, the behavior recognition module, the energy consumption prediction module, the energy saving control module, the distributed execution module and the maintenance management module are connected with each other.

[0013] Optionally, the data construction module comprises:

[0014] a sensor network unit, configured to deploy environment sensors, energy consumption meters and personnel sensors according to a building structure to obtain collection equipment, and install a multi-protocol adaptive intelligent gateway and a current transformer on the collection equipment;

[0015] a unified collection unit, configured to perform data collection by using the collection equipment to obtain data packets, add standardized time stamps and spatial labels to all the data packets for time and space alignment, and perform preliminary data cleaning and abnormal alarm by using an edge computing gateway, and perform format conversion and real-time uploading by using middleware services and standardized interfaces to obtain an original data set;

[0016] an information model unit, configured to construct a building information model, upload the original data set to the building information model for processing to obtain a standardized multi-dimensional data set.

[0017] Optionally, the information model unit comprises:

[0018] a building structure subunit, configured to generate a mapping relationship of each collection point by using a BIM platform according to categories, lines and three-dimensional spatial positions of the collection equipment to obtain a building space topology structure;

[0019] The equipment topology relationship sub-unit is used to attach the time-series data of the acquisition device to the BIM platform based on the building space topology, and map the acquisition number to the BIM object attribute to obtain the equipment topology relationship diagram;

[0020] The graph construction sub-unit is used to organize the energy flow paths, equipment hierarchy relationships and control logic within the building using a graph database, and to obtain an energy flow map.

[0021] The model building subunit is used to add the equipment topology diagram and the energy flow map to the building space topology to obtain a building information model. The original dataset is uploaded to the building information model for processing to obtain a standardized multidimensional dataset.

[0022] Optionally, the behavior recognition module includes:

[0023] The data processing unit is used to filter environmental parameters, personnel information and equipment data in the standardized multidimensional dataset to obtain spatiotemporal behavior-related data, and extract time features, spatial features, behavioral features, environmental features and user tags from the spatiotemporal behavior-related data to obtain spatiotemporal behavior features;

[0024] The individual trajectory prediction unit is used to bind the personnel information with spatial ID to obtain personnel dynamic data. Based on the personnel dynamic data, supervised learning is performed using an LSTM network to predict the user's movement trajectory and obtain the user trajectory sequence.

[0025] The spatial pattern recognition unit is used to extract personnel activity features based on the user trajectory sequence under spatial-temporal slices, and based on the personnel activity features, use an unsupervised clustering algorithm, combined with the environmental parameters and BIM spatial attributes to define semantic labels for the clusters, and obtain behavioral pattern labels for each space.

[0026] The spatiotemporal behavior mining unit is used to count the real-time number of people in each space based on the user trajectory sequence and the behavior pattern label, with a time granularity of 15 minutes, to obtain a three-dimensional matrix including time, space and number of people. Then, based on the three-dimensional matrix, Gaussian smoothing is used to eliminate extreme anomalies to obtain a three-dimensional heat map including time, space and number of people.

[0027] The crowd prediction unit is used to construct a crowd flow prediction model based on the three-dimensional heat map and the behavioral pattern labels, using a spatiotemporal convolutional network, to predict the population density sequence of each space in the next 24 hours.

[0028] Optionally, the energy consumption prediction module includes:

[0029] The impact factor analysis unit is used to extract energy consumption impact factors, trend features, periodic features, and node features based on the standardized multidimensional dataset, the three-dimensional heat map, and the population density sequence. Based on the energy consumption impact factors, it constructs interaction features and then aggregates the energy consumption impact factors, trend features, periodic features, node features, and interaction features to obtain a feature tensor.

[0030] Model architecture unit, used to build a model architecture that outputs partial energy consumption prediction, total energy consumption prediction and load curve prediction in parallel;

[0031] The hybrid prediction unit is used to perform short-term energy consumption prediction, medium-term energy consumption prediction and long-term energy consumption prediction based on the model architecture and the feature tensor, so as to obtain multi-level energy consumption prediction results.

[0032] Optionally, the energy consumption influencing factors include external meteorological factors, indoor environmental conditions, behavioral factors, equipment operating conditions, and static attributes; the trend characteristics include time-series moving averages and abnormal fluctuations; the periodic characteristics include daily cycles and specific activities; and the node characteristics include peak and off-peak periods and instantaneous intervention behaviors.

[0033] Optionally, the hybrid prediction unit includes:

[0034] The short-term energy consumption prediction subunit is used to input the feature tensor into the XGBoost model based on the model architecture, process the nonlinear features and categorical variables in the feature tensor, and output the sub-item energy consumption, total energy consumption and load risk warning for the next 0-6 hours;

[0035] The medium-term energy consumption prediction subunit is used to input the feature tensor into the Transformer based on the model architecture and output the sub-item energy consumption and total energy consumption for each period in the next 6-24 hours.

[0036] The long-term energy consumption prediction subunit is used to integrate SARIMA and neural networks to obtain a long-term prediction model. Based on the model architecture, the feature tensor is input into the long-term prediction model to output the sub-item energy consumption, total energy consumption and load corridor prediction for the next 1-7 days.

[0037] Optionally, the energy-saving control module includes:

[0038] A multi-objective optimization unit is used to design a comfort function and a total energy consumption function based on the multi-level energy consumption prediction results, the three-dimensional heat map, and the standardized multidimensional dataset, to obtain an objective function. The unit then uses the NSGA-III algorithm to perform multi-objective optimization on the objective function, resulting in a Pareto front solution set including multiple energy-saving control strategies.

[0039] The expression for calculating the comfort function is as follows:

[0040] ;

[0041] in, For temperature comfort, For lighting comfort, For air quality comfort, For other noise or wind speed comfort, All are weights;

[0042] The expression for calculating the total energy consumption function is as follows:

[0043] ;

[0044] Where U is the vector of control decision variables. Let i be the energy consumption of device i over time t under the control decision U. Assign weights to energy consumption items;

[0045] The strategy simulation unit is used to perform strategy pre-simulation on the Pareto front solution set using a BIM-based digital twin simulation platform, and then perform strategy evaluation to obtain a comprehensive utility score for each energy-saving control strategy, so as to select the final recommended strategy; wherein:

[0046] The formula for calculating the comprehensive utility score is as follows:

[0047] ;

[0048] in, Let j be the overall utility index of the strategy group. Let j be the energy consumption of the strategy group. For the user comfort of the j-th strategy, Let the operating cost of strategy group j be... All are weights;

[0049] The hierarchical control unit is used to organize the final recommended strategy into a global strategy and a regional strategy based on the overall building area and local areas, thus obtaining a control strategy set.

[0050] Optionally, the distributed execution module includes:

[0051] The equipment linkage unit is used to deploy edge controllers on each floor, and then interface the edge controllers with the building's air conditioning system, lighting system, shading system and elevator system to perform equipment linkage control;

[0052] An edge deployment unit is used to receive, verify, and issue control policy set instructions using the edge controller to control building equipment;

[0053] The real-time feedback unit is used to monitor the actual energy consumption data of building equipment, equipment operating status and user feedback in real time, obtain comprehensive execution results, and adjust the control strategy set based on the comprehensive execution results.

[0054] Optionally, the maintenance management module includes:

[0055] The performance evaluation unit is used to calculate the energy saving rate, comfort compliance rate and equipment operating efficiency of building equipment based on the real-time monitoring data of the acquisition device, obtain performance KPIs, and then perform maintenance and management of the building equipment based on the performance KPIs.

[0056] The visualization management unit is used to design a large visualization screen, a PC management terminal, and a mobile APP to perform visual management of the energy consumption prediction module, the energy-saving control module, and the distributed execution module.

[0057] This invention discloses the following technical effects by providing an intelligent building energy-saving control system based on AI energy consumption prediction:

[0058] 1. High-quality spatiotemporal data and equipment modeling: By integrating with the BIM platform, multi-type data is deeply merged with building space and equipment topology, and energy flow maps and multi-dimensional relationships are constructed using a graph database. This ensures the standardization, multi-dimensionality, and consistency of the collected data, laying the foundation for intelligent analysis throughout the entire process. It significantly improves the efficiency of data acquisition, uploading, and processing, and reduces the impact of noise and anomalies on the system. It supports subsequent precise mapping to spatial scenes and equipment levels, improving the accuracy of system response.

[0059] 2. Intelligent recognition of user behavior: By predicting individual trajectories, accurately modeling user space activities, functional status of each space, and predicting future pedestrian flow distribution in each space, it can dynamically observe the activities and usage patterns of people in each space, realize on-demand allocation of resources such as air conditioning and lighting, significantly improve the adaptability and personalization of energy-saving operations, reduce resource waste, and optimize user experience.

[0060] 3. Accurate Energy Consumption Prediction: By comprehensively integrating multiple factors such as environment, equipment, and behavior, it can achieve time-sharing, item-by-item, and multi-cycle energy consumption prediction; it can know the energy consumption trend and high-risk periods in advance, making energy-saving decisions more forward-looking, obtaining high-precision energy consumption prediction results, and providing data support for energy dispatch.

[0061] 4. Intelligent and precise energy-saving control strategies: By designing multi-objective functions and calculating the Pareto front solution of the optimal strategy, the final strategy is selected based on the comprehensive score. The building is controlled in a global-regional hierarchical manner to achieve a multi-dimensional balance between energy consumption and comfort. The strategy is intelligently and dynamically adjusted to adapt to complex scenarios and multi-dimensional needs, thereby improving the adaptability and intelligence of the control strategy, maximizing energy-saving potential, and ensuring spatial comfort.

[0062] 5. Automated control and adjustment: Real-time linkage at the device level is achieved through edge controllers, which can ensure the implementation of energy-saving control strategies and meet the requirements of local response and system linkage; it can also collect and monitor the execution status in real time and make adaptive adjustments to the control strategies.

[0063] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0064] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 This is a schematic diagram of the system architecture provided in an embodiment of the present invention;

[0066] Figure 2 This is a schematic diagram of the behavior prediction process provided in an embodiment of the present invention;

[0067] Figure 3 This is a schematic diagram of the energy consumption prediction and regulation process provided in an embodiment of the present invention. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0070] like Figure 1 As shown, this invention provides an intelligent building energy-saving control system based on AI energy consumption prediction, comprising:

[0071] 1. Data Construction Module

[0072] This module is used to design sensor networks and building information models based on building structures, for unified data acquisition and preliminary processing, to obtain standardized multidimensional datasets; the data construction module includes:

[0073] 1.1 Sensor Network Unit

[0074] Determine the types of data to be collected, the number of data points, the installation locations, and the collection frequency.

[0075] Static information requirements: building structure, equipment layout, pipeline routing, etc., for BIM modeling and basic comparison.

[0076] Dynamic Data Requirements: Environmental Parameters: Data points will be deployed according to functional zones / typical spaces, taking into account pedestrian flow and the distribution of heat and cold sources. Energy Consumption Meters: Priority will be given to deploying data points for important energy consumption items and large circuits directly related to air conditioning terminals, lighting, and sockets. Personnel Sensing: Dense sensing will be implemented in key areas such as entrances and exits, core corridors, and conference rooms. Equipment Monitoring: The operating parameters and status of all key energy-consuming and environmental control equipment will be collected.

[0077] Based on the building structure, environmental sensors, energy consumption meters, and personnel sensors are deployed to obtain data acquisition equipment, and a multi-protocol compatible smart gateway and current transformer are installed on the data acquisition equipment.

[0078] Environmental sensors: Select industrial or commercial grade devices with high accuracy and strong network compatibility for temperature, humidity, CO2, light, and PM2.5. At least one set should be installed in each main space and public area.

[0079] Energy consumption meters: Select sub-meters for electricity, water and gas with communication capabilities, preferably using Modbus or Ethernet protocols for easy unified access later.

[0080] Personnel sensor: Combining infrared pyroelectric, millimeter wave, video AI analysis module and Wi-Fi probe to improve the authenticity and accuracy of occupancy status.

[0081] Equipment status monitoring: Utilize the existing data acquisition capabilities of the building control / BAS system, or add smart gateways and current transformers to the equipment.

[0082] Network connectivity: wired connections are prioritized, supplemented by wireless connections, forming a stable network structure and ensuring real-time data transmission.

[0083] 1.2 Unified Data Acquisition Unit

[0084] The data acquisition device is used to collect data packets. Standardized timestamps and spatial tags are added to all the data packets for spatiotemporal alignment. Preliminary data cleaning and anomaly alarms are performed using an edge computing gateway. Then, middleware services and standardized interfaces are used for format conversion and real-time uploading to obtain the original dataset.

[0085] 1.3 Information Model Unit

[0086] To construct a building information model (BIM), the original dataset is uploaded to the BIM for processing to obtain a standardized multidimensional dataset. The information model unit includes:

[0087] 1.3.1 Building Structural Subunit

[0088] Based on the type, line, and three-dimensional spatial location of the acquisition device, the BIM platform is used to generate a mapping relationship for each acquisition point, resulting in the building space topology: room-area-floor-building.

[0089] 1.3.2 Equipment Topology Relationship Subunit

[0090] This is used to attach the time-series data of the acquisition device to the BIM platform based on the building space topology, and map the acquisition number to the BIM object attributes to obtain the device topology diagram.

[0091] 1.3.3 Map Construction Subunit

[0092] It is used to organize energy flow paths, equipment hierarchy relationships and control logic within a building using a graph database, to obtain an energy flow map, supporting subsequent simulation and control optimization.

[0093] 1.3.4 Model Building Subunit

[0094] The device topology diagram and the energy flow map are added to the building space topology to obtain a building information model. The original dataset is uploaded to the building information model for processing to obtain a standardized multidimensional dataset.

[0095] Standardized multidimensional datasets: environmental data, energy consumption data, personnel data, equipment data, and building static data, serving as the data foundation for intelligent algorithms and energy consumption analysis.

[0096] 2. Behavior recognition module

[0097] like Figure 2 As shown, the system is used to extract spatiotemporal behavior-related data and features from the standardized multidimensional dataset to analyze user movement trajectories and spatial behavior patterns, construct a three-dimensional heatmap, and predict the population density sequence of each space within the next 24 hours; the behavior recognition module includes:

[0098] 2.1 Data Processing Unit

[0099] This is used to filter environmental parameters, personnel information, and equipment data in the standardized multidimensional dataset to obtain spatiotemporal behavior-related data. Then, time features, spatial features, behavioral features, environmental features, and user tags are extracted from the spatiotemporal behavior-related data to obtain spatiotemporal behavior features.

[0100] Environmental parameters: including temperature, humidity, CO2, light intensity, etc., aggregated by spatiotemporal tags.

[0101] Personnel information: location, identity tags, activity trajectory, and linked to spatial ID.

[0102] Device data: air conditioning settings, lighting switches, terminal comfort settings, which are linked to user accounts / space IDs / time.

[0103] Time features: hour / week, holiday tags, special event calendar.

[0104] Spatial characteristics: space number, functional type: meeting room / office / rest room, etc., adjacent relationship.

[0105] Behavioral characteristics: dwell time, movement speed, switching frequency, number of manual adjustments.

[0106] Environmental characteristics: local environmental mean / rate of change, extreme values, etc.

[0107] User tags: user category, historical feedback level, etc.

[0108] 2.2 Individual Trajectory Prediction Unit

[0109] This is used to bind the personnel information with a spatial ID to obtain personnel dynamic data. Based on the personnel dynamic data, a supervised learning process is performed using an LSTM network to predict the user's movement trajectory and obtain the user trajectory sequence.

[0110] 2.3 Spatial Pattern Recognition Unit

[0111] This is used to extract personnel activity features, such as the number of people at the same time, average stay duration, and entry / exit frequency, based on the user trajectory sequence under spatial-temporal slices. Based on these personnel activity features, an unsupervised clustering algorithm is used, combined with the environmental parameters and BIM spatial attributes, to define semantic labels for the clusters, such as "high-frequency office area," "intermittent meeting area," and "low-occupancy negotiation area," to obtain behavioral pattern labels for each space, reflecting characteristics such as activity intensity and periodicity.

[0112] 2.4 Spatiotemporal Behavior Mining Unit

[0113] Based on the user trajectory sequence and the behavior pattern tags, the system counts the real-time number of people in each space with a time granularity of 15 minutes, obtaining a three-dimensional matrix including time, space, and number of people. Then, based on the three-dimensional matrix, Gaussian smoothing is used to eliminate extreme anomalies and improve the thermal consistency, resulting in a three-dimensional heatmap including time, space, and number of people.

[0114] 2.5 Population Prediction Unit

[0115] Based on the three-dimensional heat map and the behavioral pattern labels, this method utilizes a spatiotemporal convolutional network to fully leverage the spatial relationships of building topology, models spatial nodes and thermal data as a spatial adjacency matrix plus temporal input, constructs a crowd flow prediction model, and predicts the population density sequence of each space in the next 24 hours.

[0116] Among them, the people density sequence is a key spatiotemporal statistical indicator in intelligent building behavior modeling. It is used to describe the number of people actually observed or predicted at a series of consecutive time points in a spatial unit, such as a room, area, or floor, organized into an ordered list according to time windows. Essentially, it expresses the dynamic trend of people changing over time, that is, the flow of people changing under a two-dimensional slice of time and space. For example, suppose that in conference room A, the number of people is counted every 15 minutes throughout the day. The result of the people density sequence for the whole day is as follows: [0,0,2,5,8,10,9,7,2,0,0,0,...]. The above sequence represents the number of people in conference room A in a certain day, divided into 96 frames over 24 hours.

[0117] 3. Energy consumption prediction module

[0118] like Figure 3 As shown, the system is used to perform feature extraction and factor analysis based on the standardized multidimensional dataset, the three-dimensional heatmap, and the population density sequence to predict short-term, medium-term, and long-term energy consumption, obtaining multi-level energy consumption prediction results; the energy consumption prediction module includes:

[0119] 3.1 Impact Factor Analysis Unit

[0120] Based on the standardized multidimensional dataset, the three-dimensional heatmap, and the population density sequence, this method extracts energy consumption influencing factors, trend features, periodic features, and node features. Based on the energy consumption influencing factors, it constructs interactive features and then aggregates the energy consumption influencing factors, trend features, periodic features, node features, and interactive features to obtain a feature tensor.

[0121] 1) Energy consumption influencing factors include:

[0122] External meteorological factors: outdoor temperature, humidity, solar radiation, wind speed, etc.

[0123] Indoor environmental conditions: CO2, room temperature, light, etc.

[0124] Behavioral factors: personnel density, space occupancy rate, and usage type.

[0125] Equipment operating conditions: equipment start / stop status, load rate, set temperature / lighting level, etc.

[0126] Static attributes: space area, orientation, floor level, thermal insulation performance.

[0127] 2) Interaction features:

[0128] Temperature difference - number of people: When the temperature difference is high, the number of people increases the energy consumption, the heating is faster, and the cooling is stronger.

[0129] Lighting - Space Orientation: Sunlight in south-facing rooms affects lighting and air conditioning load.

[0130] Population change rate - set temperature adjustment: the impact of behavioral sensitivity on energy consumption.

[0131] 3) Trend characteristics: sequential moving average and abnormal fluctuations.

[0132] 4) Periodic characteristics include daily cycles and specific activities.

[0133] 5) Node characteristics include peak and off-peak periods and instantaneous intervention behaviors.

[0134] 3.2 Model Architecture Unit

[0135] This is used to construct a model architecture for parallel output of component energy consumption prediction, total energy consumption prediction, and load curve prediction.

[0136] Energy consumption prediction by component: air conditioning, lighting, sockets, etc.

[0137] Total energy consumption forecast: whole building, zoned.

[0138] Load curve forecast: for each cycle in the next 0-7 days.

[0139] 3.3 Hybrid Prediction Unit

[0140] This unit is used to perform short-term, medium-term, and long-term energy consumption predictions based on the model architecture and utilizing the feature tensor, thereby obtaining multi-level energy consumption prediction results. The hybrid prediction unit includes:

[0141] 3.3.1 Short-term energy consumption forecasting subunit

[0142] Based on the model architecture, the feature tensor is input into the XGBoost model to process the nonlinear features and categorical variables in the feature tensor, and output the sub-item energy consumption, total energy consumption and load risk warning for the next 0-6 hours.

[0143] 3.3.2 Medium-term energy consumption forecasting subunit

[0144] Based on the model architecture, the feature tensor is input into the Transformer to output the energy consumption of each time period and the total energy consumption for the next 6-24 hours.

[0145] 3.3.3 Long-term energy consumption prediction subunit

[0146] The model integrates SARIMA and neural networks to obtain a long-term prediction model. Based on the model architecture, the feature tensor is input into the long-term prediction model to output the sub-item energy consumption, total energy consumption and load corridor prediction for the next 1-7 days.

[0147] 4. Energy-saving control module

[0148] like Figure 3 As shown, the system is used to design an objective function based on the multi-level energy consumption prediction results, the three-dimensional heat map, and the standardized multidimensional dataset, and then obtain a control strategy set using the NSGA-III algorithm and a digital twin simulation platform; the energy-saving control module includes:

[0149] 4.1 Multi-objective optimization unit

[0150] Based on the multi-level energy consumption prediction results, the three-dimensional heat map, and the standardized multidimensional dataset, a comfort function and a total energy consumption function are designed to obtain an objective function. The NSGA-III algorithm is then used to perform multi-objective optimal solution on the objective function to obtain a Pareto front solution set that includes multiple energy-saving control strategies.

[0151] NSGA-III: Non-dominated Sorting Genetic Algorithm III, is a genetic algorithm in the field of multi-objective optimization. It mainly solves high-dimensional objective optimization problems. Its core advantage lies in the introduction of a "reference point" mechanism, which improves the stability and diversity of the search in high-dimensional objective space, allowing the algorithm to find the optimal solution that takes into account multiple objectives more efficiently.

[0152] The expression for calculating the comfort function is as follows:

[0153] ;

[0154] in, For temperature comfort, For lighting comfort, For air quality comfort, For other noise or wind speed comfort, All are weights.

[0155] The expression for calculating the total energy consumption function is as follows:

[0156] ;

[0157] Where U is the vector of control decision variables. Let i be the energy consumption of device i over time t under the control decision U. The weighting is for the energy consumption component.

[0158] 4.2 Strategy Simulation Unit

[0159] This is used to perform strategy pre-simulation on the Pareto front solution set using a BIM-based digital twin simulation platform, followed by strategy evaluation to obtain a comprehensive utility score for each energy-saving control strategy, in order to select the final recommended strategy; wherein:

[0160] The formula for calculating the comprehensive utility score is as follows:

[0161] ;

[0162] in, Let j be the overall utility index of the strategy group. Let j be the energy consumption of the strategy group. For the user comfort of the j-th strategy, Let the operating cost of strategy group j be... All are weights.

[0163] 4.3 Hierarchical control unit

[0164] Based on the overall and local areas of the building, the final recommended strategies are organized into global and regional strategies to obtain a set of control strategies.

[0165] Global strategy: For the building as a whole, plan the start-up and shutdown of cold / heat sources, group control of main units, energy price response, total load constraints, etc.

[0166] Regional strategy: The floor / functional area space is divided into zones, and the settings are adaptively adjusted according to predicted pedestrian flow, heat / light load, etc., such as flexible control of the zoning of office area, meeting area and public area.

[0167] 5. Distributed Execution Module

[0168] like Figure 3 As shown, the distributed execution module is used to control and adjust building equipment and provide real-time feedback based on the control strategy set using an edge controller; the distributed execution module includes:

[0169] 5.1 Equipment Interaction Unit

[0170] This is used to deploy edge controllers on each floor, and then interface the edge controllers with the building's air conditioning system, lighting system, shading system and elevator system to perform equipment linkage control.

[0171] Air conditioning system: issues commands for temperature and humidity settings, frequency conversion rate, pre-cooling and preheating, and air volume.

[0172] Lighting system: dynamically adjusts light brightness, combines natural light with BIM space orientation, automatic zone switching and daylight compensation.

[0173] Shading system: Automatically adjusts the opening angle of curtains or blinds according to sunlight, season, and room temperature to prevent glare and reduce cooling load.

[0174] Elevator system: Linked with passenger flow prediction, optimized group control algorithm, intelligent allocation of car operation, and intelligent hibernation or priority energy saving in off-peak mode.

[0175] 5.2 Edge Deployment Unit

[0176] The edge controller is used to receive, verify, and issue control policy set instructions to control building equipment; the edge controller has built-in local storage, data buffering, and decision reasoning capabilities.

[0177] 5.3 Real-time Feedback Unit

[0178] This is used to monitor the actual energy consumption data, equipment operating status, and user feedback of building equipment in real time, obtain comprehensive execution results, and adjust the control strategy set based on the comprehensive execution results.

[0179] 6. Maintenance and Management Module

[0180] This module is used to design a visual interactive interface and to perform maintenance management of building equipment based on the execution results of the control strategy set; the maintenance management module includes:

[0181] 6.1 Performance Evaluation Unit

[0182] This is used to calculate the energy saving rate, comfort compliance rate, and equipment operating efficiency of building equipment based on the real-time monitoring data of the acquisition device, obtain performance KPIs, and then perform maintenance and management of the building equipment based on the performance KPIs.

[0183] 6.2 Visual Management Unit

[0184] This is used to design a large visualization screen, a PC management terminal, and a mobile APP to perform visual management of the energy consumption prediction module, the energy-saving control module, and the distributed execution module.

[0185] Therefore, this invention provides an intelligent building energy-saving control system based on AI energy consumption prediction, which can realize personalized energy-saving strategies and active control adjustment, multi-dimensional energy consumption prediction and intelligent optimization, achieve refined and information-based management and maintenance of energy-saving control, and promote the sustainable development of smart buildings and green buildings.

[0186] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0187] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. An intelligent building energy-saving control system based on AI energy consumption prediction, characterized in that, include: The data construction module is used to design sensor networks and building information models based on building structures, so as to collect and process data in a unified manner and obtain standardized multidimensional datasets. The behavior recognition module is used to extract spatiotemporal behavior-related data and spatiotemporal behavior features from the standardized multidimensional dataset to analyze user movement trajectories and spatial behavior patterns, construct a three-dimensional heat map, and predict the population density sequence of each space in the next 24 hours. The energy consumption prediction module is used to perform feature extraction and factor analysis based on the standardized multidimensional dataset, the three-dimensional heat map and the population density sequence to predict short-term energy consumption, medium-term energy consumption and long-term energy consumption, and obtain multi-level energy consumption prediction results. The energy-saving control module is used to design an objective function based on the multi-level energy consumption prediction results, the three-dimensional heat map, and the standardized multidimensional dataset, and then use the NSGA-III algorithm and the digital twin simulation platform to obtain a set of control strategies. The distributed execution module is used to control and adjust the building equipment and provide real-time feedback using the edge controller according to the control strategy set; The maintenance management module is used to design a visual interactive interface and to perform maintenance management on building equipment based on the execution results of the control strategy set. The data construction module, the behavior recognition module, the energy consumption prediction module, the energy-saving control module, the distributed execution module, and the maintenance management module are interconnected.

2. The intelligent building energy-saving control system based on AI energy consumption prediction according to claim 1, characterized in that, The data construction module includes: The sensor network unit is used to deploy environmental sensors, energy consumption meters and personnel sensors according to the building structure to obtain data acquisition devices, and to install multi-protocol compatible smart gateways and current transformers on the data acquisition devices. A unified acquisition unit is used to acquire data using the acquisition device to obtain data packets. For all the data packets, standardized timestamps and spatial tags are added for spatiotemporal alignment. The edge computing gateway is used for preliminary data cleaning and anomaly alarms. Then, middleware services and standardized interfaces are used for format conversion and real-time uploading to obtain the original dataset. The information model unit is used to construct a building information model. The original dataset is uploaded to the building information model for processing to obtain a standardized multidimensional dataset.

3. The intelligent building energy-saving control system based on AI energy consumption prediction according to claim 2, characterized in that, The information model unit includes: The building structure sub-unit is used to generate a mapping relationship for each acquisition point using the BIM platform based on the type, line and three-dimensional spatial location of the acquisition device, so as to obtain the building space topology. The equipment topology relationship sub-unit is used to attach the time-series data of the acquisition device to the BIM platform based on the building space topology, and map the acquisition number to the BIM object attribute to obtain the equipment topology relationship diagram; The graph construction sub-unit is used to organize the energy flow paths, equipment hierarchy relationships and control logic within the building using a graph database, and to obtain an energy flow map. The model building subunit is used to add the equipment topology diagram and the energy flow map to the building space topology to obtain a building information model. The original dataset is uploaded to the building information model for processing to obtain a standardized multidimensional dataset.

4. The intelligent building energy-saving control system based on AI energy consumption prediction according to claim 3, characterized in that, The behavior recognition module includes: The data processing unit is used to filter environmental parameters, personnel information and equipment data in the standardized multidimensional dataset to obtain spatiotemporal behavior-related data, and extract time features, spatial features, behavioral features, environmental features and user tags from the spatiotemporal behavior-related data to obtain spatiotemporal behavior features; The individual trajectory prediction unit is used to bind the personnel information with spatial ID to obtain personnel dynamic data. Based on the personnel dynamic data, supervised learning is performed using an LSTM network to predict the user's movement trajectory and obtain the user trajectory sequence. The spatial pattern recognition unit is used to extract personnel activity features based on the user trajectory sequence under spatial-temporal slices, and based on the personnel activity features, use an unsupervised clustering algorithm, combined with the environmental parameters and BIM spatial attributes to define semantic labels for the clusters, and obtain behavioral pattern labels for each space. The spatiotemporal behavior mining unit is used to count the real-time number of people in each space based on the user trajectory sequence and the behavior pattern label, with a time granularity of 15 minutes, to obtain a three-dimensional matrix including time, space and number of people. Then, based on the three-dimensional matrix, Gaussian smoothing is used to eliminate extreme anomalies to obtain a three-dimensional heat map including time, space and number of people. The crowd prediction unit is used to construct a crowd flow prediction model based on the three-dimensional heat map and the behavioral pattern labels, using a spatiotemporal convolutional network, to predict the population density sequence of each space in the next 24 hours.

5. The intelligent building energy-saving control system based on AI energy consumption prediction according to claim 4, characterized in that, The energy consumption prediction module includes: The impact factor analysis unit is used to extract energy consumption impact factors, trend features, periodic features, and node features based on the standardized multidimensional dataset, the three-dimensional heat map, and the population density sequence. Based on the energy consumption impact factors, it constructs interaction features and then aggregates the energy consumption impact factors, trend features, periodic features, node features, and interaction features to obtain a feature tensor. Model architecture unit, used to build a model architecture that outputs partial energy consumption prediction, total energy consumption prediction and load curve prediction in parallel; The hybrid prediction unit is used to perform short-term energy consumption prediction, medium-term energy consumption prediction and long-term energy consumption prediction based on the model architecture and the feature tensor, so as to obtain multi-level energy consumption prediction results.

6. The intelligent building energy-saving control system based on AI energy consumption prediction according to claim 5, characterized in that, The energy consumption influencing factors include external meteorological factors, indoor environmental conditions, behavioral factors, equipment operating conditions, and static attributes. The trend characteristics include time-series moving averages and abnormal fluctuations. The periodic characteristics include daily cycles and specific activities. The node characteristics include peak and off-peak periods and instantaneous intervention behaviors.

7. The intelligent building energy-saving control system based on AI energy consumption prediction according to claim 6, characterized in that, The hybrid prediction unit includes: The short-term energy consumption prediction subunit is used to input the feature tensor into the XGBoost model based on the model architecture, process the nonlinear features and categorical variables in the feature tensor, and output the sub-item energy consumption, total energy consumption and load risk warning for the next 0-6 hours; The medium-term energy consumption prediction subunit is used to input the feature tensor into the Transformer based on the model architecture and output the sub-item energy consumption and total energy consumption for each period in the next 6-24 hours. The long-term energy consumption prediction subunit is used to integrate SARIMA and neural networks to obtain a long-term prediction model. Based on the model architecture, the feature tensor is input into the long-term prediction model to output the sub-item energy consumption, total energy consumption and load corridor prediction for the next 1-7 days.

8. The intelligent building energy-saving control system based on AI energy consumption prediction according to claim 7, characterized in that, The energy-saving control module includes: A multi-objective optimization unit is used to design a comfort function and a total energy consumption function based on the multi-level energy consumption prediction results, the three-dimensional heat map, and the standardized multidimensional dataset, to obtain an objective function. The unit then uses the NSGA-III algorithm to perform multi-objective optimization on the objective function, resulting in a Pareto front solution set including multiple energy-saving control strategies. The expression for calculating the comfort function is as follows: ； in, For temperature comfort, For lighting comfort, For air quality comfort, For other noise or wind speed comfort, All are weights; The expression for calculating the total energy consumption function is as follows: ； Where U is the vector of control decision variables. Let i be the energy consumption of device i over time t under the control decision U. Assign weights to energy consumption items; The strategy simulation unit is used to perform strategy pre-simulation on the Pareto front solution set using a BIM-based digital twin simulation platform, and then perform strategy evaluation to obtain a comprehensive utility score for each energy-saving control strategy, so as to select the final recommended strategy; wherein: The formula for calculating the comprehensive utility score is as follows: ； in, Let j be the overall utility index of the strategy group. Let j be the energy consumption of the strategy group. For the user comfort of the j-th strategy, Let the operating cost of strategy group j be... All are weights; The hierarchical control unit is used to organize the final recommended strategy into a global strategy and a regional strategy based on the overall building area and local areas, thus obtaining a control strategy set.

9. The intelligent building energy-saving control system based on AI energy consumption prediction according to claim 8, characterized in that, The distributed execution module includes: The equipment linkage unit is used to deploy edge controllers on each floor, and then interface the edge controllers with the building's air conditioning system, lighting system, shading system and elevator system to perform equipment linkage control; An edge deployment unit is used to receive, verify, and issue control policy set instructions using the edge controller to control building equipment; The real-time feedback unit is used to monitor the actual energy consumption data of building equipment, equipment operating status and user feedback in real time, obtain comprehensive execution results, and adjust the control strategy set based on the comprehensive execution results.

10. The intelligent building energy-saving control system based on AI energy consumption prediction according to claim 9, characterized in that, The maintenance management module includes: The performance evaluation unit is used to calculate the energy saving rate, comfort compliance rate and equipment operating efficiency of building equipment based on the real-time monitoring data of the acquisition device, obtain performance KPIs, and then perform maintenance and management of the building equipment based on the performance KPIs. The visualization management unit is used to design a large visualization screen, a PC management terminal, and a mobile APP to perform visual management of the energy consumption prediction module, the energy-saving control module, and the distributed execution module.

Images