Green building-oriented energy consumption self-adaptive operation and maintenance method and system

Abstract

The application relates to a green building-oriented energy consumption self-adaptive operation and maintenance method and system. Historical operation and maintenance data of each building partition in a target green building are acquired by a cloud management platform to train a model, and an operation and maintenance state prediction model is obtained. In combination with historical operation and maintenance data of each partition and preset professional regulation and control rules, targeted equipment regulation and control strategy data are generated. The trained prediction model and the regulation and control strategy data special for each partition are packaged and distributed to corresponding edge computing nodes. The edge computing nodes complete the localization deployment of the model and the loading of the strategy data, realize the capability migration from 'cloud dependence' to 'edge autonomy', continuously monitor real-time operation and maintenance characteristic data in the owned partition, call the local model to quickly predict the operation and maintenance state, combine the equipment regulation and control strategy data, automatically generate partition regulation and control instructions adaptive to current and future short-time working conditions, and issue the instructions to regulation and control execution equipment in the region in the first time to complete closed-loop control.

Description

Technical Field

[0001] This application relates to the field of intelligent building management technology, and in particular to an energy consumption adaptive operation and maintenance method and system for green buildings. Background Technology

[0002] With the deepening implementation of the "dual-carbon" strategy in the construction field, green buildings have become an important carrier for urban energy conservation and emission reduction, and their energy consumption management level is directly related to the achievement of building energy conservation goals. Currently, energy consumption during the building operation and maintenance phase accounts for more than 70% of the entire life cycle. How to achieve refined energy consumption management through intelligent means has always been a focus of industry attention. In existing technologies, building energy consumption management mainly develops along two paths: one is a management and control model based on cloud-based centralized computing, which relies on a wide area network to aggregate sensor data from the entire building to the cloud, where the cloud uniformly completes the calculation and issues control commands. This type of solution is prone to cloud computing power bottlenecks when facing massive amounts of building data, and the latency caused by data round-trip transmission often makes the adjustment of equipment such as air conditioning and lighting lag behind actual environmental changes; the other is the local offline control mode adopted by traditional building automation systems, which usually relies on fixed logic rules pre-written by engineers, lacks the ability to autonomously learn from the long-term operating characteristics of buildings, and is difficult to cope with dynamic conditions such as personnel flow and sudden weather changes. Although the building AI energy-saving solutions that have emerged in recent years have introduced predictive models, most of these models are still deployed in the cloud. Local devices only act as data transmission nodes and do not truly have real-time inference capabilities at the edge. The effectiveness of intelligent control is still limited by the quality of network communication, and the system reliability drops significantly in scenarios with network outages or high latency. Summary of the Invention

[0003] Based on this, the purpose of this application is to provide an energy consumption adaptive operation and maintenance method and system for green buildings, which reduces cloud computing load and network transmission dependence, while giving the building's local nodes real-time intelligent prediction and adaptive control capabilities, thereby achieving precise energy consumption management with low latency across operating conditions.

[0004] The energy consumption adaptive operation and maintenance method for green buildings described in this application includes the following steps: Historical operation and maintenance data of each building zone of the target green building are obtained through a cloud management platform, and an operation and maintenance status prediction model is trained and generated. The historical operation and maintenance data includes historical operation and maintenance feature data as training features and historical operation and maintenance status data as training labels. Based on the historical operation and maintenance data of each building zone and preset professional control rules, equipment control strategy data corresponding to each building zone is generated. Based on the operation and maintenance status prediction model and the equipment control strategy data of each building zone, the cloud management platform generates edge control deployment data packages for each building zone; and sends the edge control deployment data packages to the corresponding edge computing nodes. Each edge computing node receives the edge control deployment data packet, deploys the operation and maintenance status prediction model and the equipment control strategy data; monitors the real-time operation and maintenance characteristic data of the building partitions respectively, inputs the real-time operation and maintenance characteristic data into the locally deployed operation and maintenance status prediction model, and obtains the operation and maintenance status prediction data output by the model. By combining the device control strategy data deployed locally by each edge computing node with the operation and maintenance status prediction data, the control instructions for the corresponding building zone are generated. The control instructions are then sent to the control execution devices of the corresponding building zone to complete the control of local energy consumption devices in the zone.

[0005] This application also provides an energy-adaptive operation and maintenance system for green buildings, including a cloud management platform and several edge computing nodes deployed in various building zones of the green building: A cloud-based management platform is used to acquire historical operation and maintenance data of each building zone of the target green building and train an operation and maintenance status prediction model. The historical operation and maintenance data includes historical operation and maintenance feature data as training features and historical operation and maintenance status data as training labels. Based on the historical operation and maintenance data of each building zone and preset professional control rules, equipment control strategy data corresponding to each building zone is generated. Based on the operation and maintenance status prediction model and the equipment control strategy data of each building zone, an edge control deployment data package for each building zone is generated. The edge control deployment data package is sent to the corresponding edge computing node. An edge computing node is used to receive the edge control deployment data packet, deploy the operation and maintenance status prediction model and the device control strategy data; monitor the real-time operation and maintenance characteristic data of its respective building zone, input the real-time operation and maintenance characteristic data into the locally deployed operation and maintenance status prediction model, and obtain the operation and maintenance status prediction data output by the model; based on the locally deployed device control strategy data and combined with the operation and maintenance status prediction data, generate the zone control instruction for its respective building zone; and send the zone control instruction to the control execution device of the corresponding building zone to complete the local energy consumption device control of the zone.

[0006] This application's embodiments construct an execution chain of "cloud-based training and deployment—edge-based local inference—partition-based closed-loop control." First, the cloud management platform aggregates all historical operation and maintenance data to train the operation and maintenance status prediction model, enabling the model to fully learn the dynamic characteristics of the building during long-term operation. Next, the cloud combines historical data with preset professional control rules to generate equipment control strategy data, and packages the model and strategy into an edge control deployment data package, which is then deployed to the edge computing nodes of the corresponding building partition. This shortens the physical distance between the decision-making unit and the execution unit, eliminating the architectural dependence on real-time network transmission and avoiding control delays caused by excessively long data transmission links in centralized cloud management. After deployment at the edge, the edge computing nodes monitor local real-time operation and maintenance characteristic data and input it into the local model. The model inference process is completed entirely in the local computing environment, without real-time interaction with the cloud. This not only significantly reduces the cloud's computing load and data bandwidth consumption but also ensures that operation and maintenance status prediction data can be generated within milliseconds, meeting the timeliness requirements of immediate device response. Finally, the edge nodes generate partition control instructions based on the locally generated prediction data and preset strategy data and directly deploy them for execution. This dual-drive mechanism of "prediction + rules" enables the control logic to move beyond static rules preset by humans. Instead, it can dynamically adjust the equipment operating parameters based on predicted operating conditions. This achieves low-latency response while also providing adaptive energy-saving effects that adapt to the dynamic operating conditions of the building. It ensures that the building zones can still maintain local autonomous intelligent control capabilities in scenarios with network interruptions or high latency.

[0007] To better understand and implement this application, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating an energy-adaptive operation and maintenance method for green buildings, as described in an embodiment of this application. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings. Wherein, when the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements.

[0010] It should be understood that the embodiments described below do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application.

[0011] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application are also intended to include the plural forms unless the context clearly indicates otherwise. Furthermore, in the description of this application, unless otherwise stated, “a plurality” means two or more. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more associated listed items, for example, A and / or B, which can represent: A alone, A and B together, and B alone; the character “ / ” generally indicates that the preceding and following objects are in an “or” relationship.

[0012] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, this information should not be limited to these terms, and these terms are only used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence, nor should they be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances. Depending on the context, the word "if" as used in this application can be interpreted as "when," "when," or "in response to determination."

[0013] Please refer to Figure 1 The energy consumption adaptive operation and maintenance method for green buildings described in this application embodiment is applied to an energy consumption adaptive operation and maintenance system, which includes a cloud management platform and several edge computing nodes deployed in various building zones of the green building; the method includes the following steps: S101: Obtain historical operation and maintenance data of each building zone of the target green building through the cloud management platform, and train and generate an operation and maintenance status prediction model; the historical operation and maintenance data includes historical operation and maintenance feature data as training features and historical operation and maintenance status data as training labels; based on the historical operation and maintenance data of each building zone and the preset professional control rules, generate equipment control strategy data corresponding to each building zone. S102: Based on the operation and maintenance status prediction model and the equipment control strategy data of each building partition, the cloud management platform generates edge control deployment data packets for each building partition; the edge control deployment data packets are sent to the corresponding edge computing nodes. S103: Receive the edge control deployment data packet through each edge computing node, deploy the operation and maintenance status prediction model and the equipment control strategy data; monitor the real-time operation and maintenance characteristic data of the building partition respectively, input the real-time operation and maintenance characteristic data into the locally deployed operation and maintenance status prediction model, and obtain the operation and maintenance status prediction data output by the model; S104: Based on the device control strategy data deployed locally by each edge computing node and combined with the operation and maintenance status prediction data, generate the partition control instructions for the building partition; send the partition control instructions to the control execution devices of the corresponding building partition to complete the local energy consumption device control of the partition.

[0014] This application's embodiments construct an execution chain of "cloud-based training and deployment—edge-based local inference—partition-based closed-loop control." First, the cloud management platform aggregates all historical operation and maintenance data to train the operation and maintenance status prediction model, enabling the model to fully learn the dynamic characteristics of the building during long-term operation. Next, the cloud combines historical data with preset professional control rules to generate equipment control strategy data, and packages the model and strategy into an edge control deployment data package, which is then deployed to the edge computing nodes of the corresponding building partition. This shortens the physical distance between the decision-making unit and the execution unit, eliminating the architectural dependence on real-time network transmission and avoiding control delays caused by excessively long data transmission links in centralized cloud management. After deployment at the edge, the edge computing nodes monitor local real-time operation and maintenance characteristic data and input it into the local model. The model inference process is completed entirely in the local computing environment, without real-time interaction with the cloud. This not only significantly reduces the cloud's computing load and data bandwidth consumption but also ensures that operation and maintenance status prediction data can be generated within milliseconds, meeting the timeliness requirements of immediate device response. Finally, the edge nodes generate partition control instructions based on the locally generated prediction data and preset strategy data and directly deploy them for execution. This dual-drive mechanism of "prediction + rules" enables the control logic to move beyond static rules preset by humans. Instead, it can dynamically adjust the equipment operating parameters based on predicted operating conditions. This achieves low-latency response while also providing adaptive energy-saving effects that adapt to the dynamic operating conditions of the building. It ensures that the building zones can still maintain local autonomous intelligent control capabilities in scenarios with network interruptions or high latency.

[0015] The energy-adaptive operation and maintenance method for green buildings described in this application uses computer equipment as the execution subject and is applied to an energy-adaptive operation and maintenance system. The energy-adaptive operation and maintenance system refers to a distributed control system consisting of a cloud management platform and edge computing nodes distributed throughout the physical zones of the green building via wired or wireless networks, used to perform closed-loop management of data acquisition, model calculation, and equipment control.

[0016] The following provides a detailed explanation of each step.

[0017] For step S101, historical operation and maintenance data of each building zone of the target green building are obtained through the cloud management platform, and an operation and maintenance status prediction model is trained and generated; the historical operation and maintenance data includes historical operation and maintenance feature data as training features and historical operation and maintenance status data as training labels; based on the historical operation and maintenance data of each building zone and the preset professional control rules, equipment control strategy data corresponding to each building zone is generated.

[0018] Among them, the cloud management platform refers to the comprehensive management server or server cluster deployed in the cloud computing center. It has the capabilities of massive data storage, high-performance parallel computing and model training. It is used to manage the full life cycle operation and maintenance data of one or more target green buildings in a unified manner, and to perform big data analysis, artificial intelligence model training and global strategy generation tasks. It interacts with edge computing nodes distributed in each building through a wide area network.

[0019] The target green building refers to the specific building entity that needs to be subject to energy consumption control. Its interior is divided into multiple relatively independent building zones according to function or physical structure, such as office area, conference area, podium commercial area or standard floor independent unit.

[0020] Edge computing nodes refer to hardware devices with embedded computing capabilities deployed on-site in green buildings (such as low-voltage rooms, high-voltage shafts, or IoT gateways within building zones on each floor). They run lightweight inference engines and real-time control logic, and are used to be close to sensing terminals and control execution devices in physical space. They are responsible for local data preprocessing, model inference calculations, and real-time control command issuance, and only need to perform non-real-time model updates and policy synchronization with the cloud.

[0021] Historical operation and maintenance data refers to the structured dataset generated during the past operation of the target green building, which has been cleaned and labeled. It is the basic corpus for model training in the cloud. Specifically, it consists of feature vectors (i.e., historical operation and maintenance feature data) used to describe the operating status of the environment and equipment and labeled values ​​(i.e., historical operation and maintenance status data) used to characterize the actual operating results or targets under that status.

[0022] An operation and maintenance status prediction model refers to a mathematical model trained by a cloud management platform using historical operation and maintenance data through machine learning or deep learning algorithms. This model can fit the nonlinear mapping relationship between building environmental parameters, equipment operating parameters, and operating status. After being deployed to edge computing nodes, it is used to infer and predict the building's operating status based on current real-time data. In one embodiment, a time-series prediction model can be used to predict the building's operating status within a short future time window (such as indoor temperature, occupancy load, or energy consumption trends in the next 10 minutes).

[0023] Historical operation and maintenance characteristic data refers to multi-dimensional feature vectors extracted from historical time series to characterize the building's operating environment and equipment status, including but not limited to outdoor meteorological parameters (temperature, humidity, irradiance), indoor environmental parameters (temperature, humidity, CO2 concentration, personnel density data), timestamp features (weekdays, holidays, time), and historical equipment operating parameters (such as chilled water outlet temperature, fan frequency).

[0024] Historical operation and maintenance status data refers to real historical status records that serve as supervision labels for model training. These records are used to characterize the actual response status of a building under specific feature data inputs, such as the actual indoor thermal comfort index (PMV), the actual energy consumption value for the next period, or the actual personnel density level.

[0025] Pre-set professional control rules refer to a set of logical constraints pre-set based on building HVAC and electrical professional design specifications and energy-saving operation experience, such as "the indoor temperature setpoint shall not be lower than 26℃", "the illuminance shall not be lower than 300 lux when people are present", or "free cooling of fresh air shall be given priority during the transition season", which are used to ensure the safety and compliance of the control strategy.

[0026] Equipment control strategy data refers to the configuration parameter table generated by the cloud based on the historical operation and maintenance characteristic data of each building zone and the preset professional control rules, which is used to guide edge nodes to control equipment. It not only includes the specific equipment operation setting value range, but also includes the constraints and priority logic to ensure the safe operation of the equipment.

[0027] In this step, the cloud management platform retrieves past time period operation logs for each building zone from the Building Management System (BMS) or historical database, and cleans and extracts historical operation and maintenance data. Taking a standard office area as an example, the platform selects time-series data from 8:00 to 18:00 daily over the past year, using the indoor and outdoor environmental parameters and personnel density of the previous hour as historical operation and maintenance characteristic data, and the indoor temperature change and energy consumption value of the next hour as historical operation and maintenance status data. A Long Short-Term Memory (LSTM) network or gradient boosting decision tree algorithm is used to train and generate an operation and maintenance status prediction model. Simultaneously, based on the "Energy-Saving Design Standard for Public Buildings" and the building's equipment ledger, the cloud platform analyzes preset professional control rules, such as defining the logic of "starting cooling when the predicted indoor temperature is higher than 26℃," and combines this with the operation and maintenance characteristics of each zone to generate differentiated equipment control strategy data.

[0028] In one embodiment, the operation and maintenance status prediction model is an energy consumption status and personnel status prediction model; the historical operation and maintenance characteristic data includes personnel activity characteristic data, environmental characteristic data, and equipment energy consumption operation characteristic data; the historical operation and maintenance status data includes personnel activity status data and energy consumption status data; the operation and maintenance status prediction data includes personnel activity status prediction data and energy consumption status prediction data.

[0029] Personnel activity characteristic data refers to characteristic variables used to describe the behavioral patterns of people in building spaces and their interaction with the environment. These may include, but are not limited to, access control card swipe records, the number of terminal connections detected by WiFi, head count data extracted from video surveillance, infrared sensor status of workstations, and probability distribution of work and rest times based on historical patterns.

[0030] Environmental characteristic data refers to parameters that characterize the physical environment of a building's interior and exterior, and may include temperature, humidity, solar radiation intensity, CO2 concentration, light intensity, and PM2.5 concentration.

[0031] Equipment energy consumption operation characteristic data refers to parameters that characterize the real-time operating conditions of energy-consuming equipment in a building. Specifically, these may include chilled water supply and return temperatures of air conditioning units, cooling water flow rate, compressor operating frequency, fan current, switching status and dimming level of lighting circuits, and active power of socket circuits.

[0032] Personnel activity status data refers to the actual labeled values ​​used as training labels for the personnel status prediction sub-model. It is used to characterize the actual presence of personnel in a building zone within a specific time period, and may include instantaneous personnel density (persons / square meter), personnel dwell time, or personnel turnover rate, etc.

[0033] Energy status data refers to the actual labeled values ​​used as training labels for the energy status prediction sub-model. It is used to characterize the actual energy consumption level within a building zone during a specific time period. Specifically, it may include minute-by-minute predicted electricity consumption, peak cooling consumption, or energy consumption per unit area for short future periods.

[0034] Operation and maintenance status prediction data refers to the output results obtained by forward reasoning of the operation and maintenance status prediction model deployed locally on the edge computing node based on real-time operation and maintenance feature data. In this embodiment, it specifically includes personnel activity status prediction data (such as predicted personnel density) and energy consumption status prediction data (such as predicted cooling load demand).

[0035] Step S101, which involves training and generating an operation and maintenance status prediction model through a cloud management platform, includes: Step S1011: Use personnel activity feature data and environmental feature data as training features and personnel activity status data as training labels to train the personnel status prediction sub-model.

[0036] The personnel status prediction sub-model refers to a machine learning model trained in the cloud that is specifically designed to establish a mapping relationship between personnel activity characteristic data, environmental characteristic data, and personnel activity status data. It is used to predict the distribution of personnel in future periods and to provide a basis for accurate prediction of the demand for fresh air and the operating demand of energy-consuming equipment such as air conditioning cooling load.

[0037] In this step, the cloud extracts personnel activity characteristic data and environmental characteristic data from the full historical operation and maintenance data as training feature sets. For example, it selects the hourly access control card swipe records and the number of WiFi connection terminals over the past six months as personnel activity characteristics, and simultaneously extracts the outdoor temperature and indoor CO2 concentration during the same period as environmental characteristics. Subsequently, the cloud uses the actual personnel density data confirmed within the same time period as training labels (i.e., personnel activity status data), and uses a hybrid neural network architecture combining convolutional neural networks and time series models to train a personnel status prediction sub-model. This sub-model is able to learn non-linear laws such as "personnel density reaches its peak one hour after the peak work period" and "CO2 concentration increases linearly with increasing personnel density".

[0038] Step S1012: Use equipment energy consumption operation characteristic data as training features and energy consumption status data as training labels to train the energy consumption status prediction sub-model.

[0039] The energy consumption status prediction sub-model refers to a machine learning model trained in the cloud that is specifically designed to establish a mapping relationship between equipment energy consumption operation characteristic data and energy consumption status data. It is used to predict the energy consumption trend of equipment in the future and support the formulation of peak shaving and valley filling strategies.

[0040] In this step, the cloud extracts equipment energy consumption operation characteristic data, such as the operating frequency of the chiller unit, the temperature difference and pressure difference of the chilled water supply and return water, and other historical operating parameters. The corresponding building zone total electricity meter readings or cooling capacity metering values ​​are used as energy consumption status data. Random forest or XGBoost algorithms are used to train the energy consumption status prediction sub-model to establish the mapping relationship between equipment operating conditions and future energy consumption.

[0041] Step S1013: Perform quantization and pruning on the personnel status prediction sub-model and the energy consumption status prediction sub-model to obtain the operation and maintenance status prediction model.

[0042] Quantization and pruning refers to a model compression technique. "Quantization" means converting model weights and activation values ​​from high-precision 32-bit floating-point numbers to low-precision 8-bit integers (INT8) or 16-bit floating-point numbers (FP16) to reduce model storage volume and computational complexity. "Pruning" means removing redundant neurons or connections in the neural network whose weights are close to zero, reducing the amount of computation without significantly reducing model accuracy, so that it can be adapted to the limited computing power of edge computing nodes.

[0043] To overcome the limitations of limited storage space and insufficient computing power in edge computing nodes, the cloud performs quantization pruning on the trained personnel status prediction sub-model and energy consumption status prediction sub-model. Specifically, the cloud uses the trained floating-point model as a baseline, performs quantization-aware training (QAT) through a calibration dataset, quantizes the model weights into INT8 format, and uses L1 regularization to prune neural connections in the model whose contribution is below a preset threshold, ultimately obtaining a lightweight operation and maintenance status prediction model suitable for edge deployment.

[0044] This embodiment decouples the operation and maintenance status prediction model into a personnel status prediction sub-model and an energy consumption status prediction sub-model. These sub-models are then trained specifically using personnel activity characteristic data, environmental characteristic data, and equipment energy consumption operation characteristic data, respectively. This significantly improves the model's ability to characterize the complex mechanisms of building operation, enabling it not only to predict energy consumption trends but also to accurately anticipate thermal disturbances caused by personnel movement. This provides more refined data support for edge node decision-making. Secondly, by performing quantization pruning on the two sub-models, the challenge of deploying high-precision deep learning models on resource-constrained edge computing nodes is effectively solved. This significantly reduces the model's computational complexity and storage requirements with almost no loss of prediction accuracy, allowing edge nodes to achieve millisecond-level real-time inference with extremely low power consumption, ensuring the immediacy of control commands.

[0045] In one embodiment, the personnel state prediction sub-model is constructed using a temporal network structure, and the energy consumption state prediction sub-model is constructed using a single-class detection network structure; the temporal network structure is a temporal convolutional network or a long short-term memory network; the single-class detection network structure is an isolated forest model or an autoencoder model.

[0046] Temporal-series network structures refer to neural network architectures specifically designed to process time-dependent sequential data, capable of capturing the correlation and trends of data over time. These networks identify patterns such as population movement and environmental temperature fluctuations over time through sliding computation of memory units or convolutional kernels. In this embodiment, a temporal-series network structure is used to construct a sub-model for predicting population status, in order to process continuous time-series data.

[0047] Temporal convolutional networks refer to a type of deep learning model built on causal convolution and dilated convolution. By expanding the receptive field, they can capture long-distance temporal dependencies. Compared with recurrent neural networks, they have the advantages of strong parallel computing capabilities, stable training, and low risk of gradient vanishing. They are suitable for processing long-sequence data on human activity features.

[0048] Long Short-Term Memory (LSTM) networks are a special type of recurrent neural network (RNN) that uses input gates, forget gates, and output gates to control the accumulation and forgetting of information. They can effectively solve the gradient vanishing problem in long sequence training and are good at capturing long-term patterns in time series, such as weekly or daily attendance patterns.

[0049] A single-class classification detection network structure refers to an anomaly detection or semi-supervised learning model that only needs to be trained using samples from a single class (normal state). Its core idea is to learn the boundaries or distribution of normal data, and to determine an anomaly when input data deviates from these boundaries. In this embodiment, it is used to construct an energy consumption state prediction sub-model, which predicts future energy consumption baselines by learning historical normal energy consumption patterns.

[0050] The Isolation Forest model refers to a single-class anomaly detection algorithm based on ensemble learning. It constructs multiple binary trees by randomly selecting features and randomly splitting feature values, thus isolating data points. Because outliers typically have unique properties, they are easier to isolate (shorter path length). In energy consumption prediction scenarios, this model can be used to establish confidence intervals for normal energy consumption, thereby predicting energy consumption trends.

[0051] An autoencoder model is a neural network consisting of an encoder and a decoder that captures the intrinsic features of data by forcing the network to learn and reconstruct the input data. The encoder compresses the high-dimensional input into a low-dimensional latent vector, and the decoder then reconstructs the data from the latent vector. In energy consumption prediction, the autoencoder is trained using historical normal energy consumption data. When real-time features are input, the reconstructed value output by the model is the predicted value for the normal energy consumption state. If the reconstruction error is large, it indicates that abnormal energy consumption may exist.

[0052] This embodiment employs a time-series network structure for the personnel status prediction sub-model, fully leveraging the advantages of this type of structure in processing time-series data. It effectively captures the periodicity, trends, and correlations of personnel movement, solving the problem that traditional static models cannot handle dynamic changes over time, thus significantly improving the accuracy of personnel density prediction. Secondly, the energy consumption status prediction sub-model uses a single-classification detection network structure, cleverly avoiding the challenges of scarce and complex distribution of abnormal samples in energy consumption data. By learning normal energy consumption patterns to define the energy consumption baseline, it not only reduces dependence on data quality but also enhances the generalization ability to unknown operating conditions, making energy consumption prediction more robust.

[0053] For step S102, the cloud management platform generates edge control deployment data packets for each building partition based on the operation and maintenance status prediction model and the equipment control strategy data for each building partition; and sends the edge control deployment data packets to the corresponding edge computing nodes.

[0054] Edge control deployment data packages refer to data sets generated by cloud encapsulation for edge-side empowerment, which at least include serialized operation and maintenance status prediction model files, device control strategy data configuration files for each partition, and metadata indexes for model-strategy matching.

[0055] In this step, the cloud management platform performs edge encapsulation of the model and policies. The cloud converts the trained operation and maintenance status prediction model into a format suitable for lightweight inference engines at the edge (such as TensorFlow Lite or ONNX Runtime), and serializes and packages it with the corresponding partition's device control policy data in JSON or XML format, adding a version number and MD5 checksum to form an edge control deployment data package. Based on the building partition's IP address or device identifier, the platform pushes these data packages to the edge computing nodes located in each partition via HTTPS protocol or MQTT message queue.

[0056] For step S103, each edge computing node receives the edge control deployment data packet, deploys the operation and maintenance status prediction model and the device control strategy data; monitors the real-time operation and maintenance characteristic data of the building partition to which it belongs, inputs the real-time operation and maintenance characteristic data into the locally deployed operation and maintenance status prediction model, and obtains the operation and maintenance status prediction data output by the model.

[0057] Real-time operation and maintenance characteristic data refers to the current data that edge computing nodes collect in real time from sensors, smart meters, or device controllers deployed within their respective building zones via Modbus, BACnet, or MQTT protocols, and that is consistent with the dimensions of the aforementioned historical operation and maintenance characteristic data.

[0058] Operation and maintenance status prediction data refers to the output results obtained by the operation and maintenance status prediction model deployed locally on the edge side based on forward inference of real-time operation and maintenance characteristic data. It is used to characterize the building operation status, such as predicted indoor temperature trends, predicted changes in personnel load, or predicted energy consumption peaks.

[0059] In this step, after receiving the data packet, each edge computing node first performs integrity verification and decompression, then loads the operation and maintenance status prediction model into its local memory, and writes the device control strategy data into the local embedded database. During the operation phase, the edge computing node periodically polls the sensing terminals within its assigned partition, such as CO2 sensors, temperature and humidity sensors, and smart meters, to obtain real-time operation and maintenance characteristic data. The node uses the current real-time data (e.g., current indoor temperature 25℃, personnel density 0.1 people / square meter) as an input tensor, inputting it into the locally deployed operation and maintenance status prediction model. After forward propagation calculation, the model outputs operation and maintenance status prediction data, for example, predicting that the indoor temperature will rise to 27℃ in 10 minutes due to personnel entering.

[0060] In one embodiment, after step S103, which involves inputting the real-time operation and maintenance feature data into the locally deployed operation and maintenance status prediction model through each edge computing node to obtain the operation and maintenance status prediction data output by the model, the following steps are included: Step S1031: Send the corresponding operation and maintenance status prediction data to the cloud management platform through each edge computing node.

[0061] In this step, each edge computing node uploads the calculated operational status prediction data (including predicted personnel density and energy consumption trends) to the cloud management platform in real time. For example, the edge node in the office area on the east side uploads "predicts that the personnel density will reach 0.2 people / square meter in 10 minutes and there will be a cooling load shortfall of 5kW", while the edge node in the meeting area on the west side uploads "predicts that the meeting will end in 10 minutes, the personnel density will return to zero and there will be a large amount of surplus cooling capacity".

[0062] Step S1032: Based on the operation and maintenance status prediction data of each edge computing node and the preset cross-zone collaborative control rules, the cloud management platform determines whether there is a need for cross-zone collaborative control. When it is determined that there is a need for cross-zone collaborative control, the real-time operation and maintenance characteristic data of all building zones of the target green building are obtained. Based on the real-time operation and maintenance characteristic data of all building zones and the operation and maintenance status prediction data, a collaborative control instruction is generated and the corresponding edge computing node is determined. The collaborative control instruction is sent to the corresponding edge computing node.

[0063] Preset cross-zone collaborative control rules refer to the logical judgment criteria built into the cloud management platform used to identify and handle energy coupling or resource conflicts between different functional areas of a building. For example, the rules can be defined as "when the core office area is predicted to experience a surge in cooling load, and the commercial area is predicted to have redundant cooling capacity or exhaust heat recovery potential, trigger cross-zone linkage"; or "when there is a significant temperature gradient difference between adjacent meeting areas and office areas and ventilation connectivity is available, trigger natural wind diversion scheduling", aiming to break down physical zoning barriers and achieve optimal allocation of resources throughout the building.

[0064] Cross-zone collaborative control needs refer to situations where, based on predicted data from local edge nodes or global real-time data analysis, it is determined that independent control of a single zone cannot meet the conditions for optimal overall building energy efficiency or balanced comfort. This typically occurs in scenarios involving heat conduction interference, shared fresh air systems, hydraulic coupling, or large-scale personnel movement, requiring cloud-based intervention for a global perspective.

[0065] Collaborative control commands refer to globally binding control commands generated by the cloud management platform after determining that there is a need for cross-zone collaborative control. Unlike intra-zone commands generated by edge nodes, these commands typically involve the linkage of devices in multiple zones. For example, "reduce the opening of the chilled water valve in zone A to suppress excessive cooling, while increasing the opening of the chilled water valve in zone B to compensate for insufficient cooling," or "open the air valve connecting zones C and D to balance air pressure," used to guide relevant edge nodes to correct their local control logic.

[0066] In this step, after receiving the predicted data from each edge computing node, the cloud management platform analyzes it according to preset cross-zone collaborative control rules. For example, the rule is set as "when the predicted temperature difference between adjacent zones is greater than 2℃ and there is a possibility of hydraulic or wind tunnel connection, it is determined that there is a need for cross-zone collaborative control." The cloud retrieves real-time operation and maintenance characteristic data (such as duct pressure, water valve opening, and partition status) of all building zones of the target green building and finds that the east office area and the west meeting area are connected by fresh air ducts. Based on this, the cloud generates a collaborative control command: "Open the fresh air linkage valve between the east and west sides to divert excess cold air from the west side to the east side." The cloud determines that this command needs to be sent to both the east and west edge computing nodes simultaneously and pushes it out through a message queue.

[0067] Step S1033: When the edge computing node receives the collaborative control instruction, it corrects the zonal control instruction in conjunction with the collaborative control instruction, and sends the corrected zonal control instruction to the control execution device of the corresponding building zone to complete the control of the local energy consumption device of the zone.

[0068] The revised partition control command refers to the final execution command generated by the edge computing node after receiving the collaborative control command from the cloud and combining it with the original partition control command generated based on local prediction, through logical fusion or overriding. This revision process follows a preset priority logic (usually cloud commands have higher priority than local prediction commands) to ensure that local execution actions are consistent with the global optimization goal, while retaining the low latency characteristics of local execution.

[0069] In this step, edge nodes do not relinquish local control due to cloud intervention; instead, they execute fused commands. For example, the eastern node, which initially generated a zonal control command to "reduce the chilled water valve opening" based on local predictions, corrects this command upon receiving a "diversion" command from the cloud, changing it to "maintain the chilled water valve opening unchanged and open the fresh air valve," thus utilizing natural cooling sources while avoiding frequent equipment start-ups and shutdowns. The western node, on the other hand, corrects its original "shut down the fan" command to "reduce the fan frequency to maintain air pressure balance." The corrected zonal control commands are immediately sent to the corresponding building zonal control execution devices, completing the final closed-loop control. During this process, even if there is a brief delay in communication between the cloud and the edge, the edge nodes can still maintain basic operation based on their local models, only making strategy corrections upon receiving collaborative commands from the cloud, thus balancing global optimization and local stability.

[0070] This embodiment constructs a technical logic of "edge prediction and reporting—cloud-based global collaboration—edge command correction." First, by enabling edge nodes to proactively report operational status prediction data, the cloud can gain insight into the future operational status of the building, identifying cross-regional energy conflicts or complementary opportunities in advance, thus solving the drawbacks of traditional building automation systems where subsystems operate independently and lack coordination. Second, by generating collaborative control commands based on global data in the cloud, the problem of a single edge node being unable to handle cross-regional coupling issues due to limited visibility is effectively solved, further unlocking energy-saving potential. Furthermore, this hierarchical and progressive control architecture enhances the system's robustness. The edge side can maintain autonomy during network fluctuations, while the cloud can make fine-tuning when the network is unobstructed, thereby maximizing the overall energy efficiency and environmental comfort of the building while ensuring system stability.

[0071] For step S104, the control policy data of each edge computing node based on the locally deployed device control strategy data is combined with the operation and maintenance status prediction data to generate the control instruction for the building zone; the control instruction is sent to the control execution device of the corresponding building zone to complete the control of local energy consumption devices in the zone.

[0072] Zone control commands refer to standardized control signals generated by edge computing nodes after fusing operation and maintenance status prediction data with equipment control strategy data. These signals are typically analog setpoints (such as air conditioning supply temperature setpoints) or digital switch commands (such as lighting circuit on / off).

[0073] Control and control equipment refers to the controlled electromechanical equipment terminals within a building zone, including but not limited to HVAC control equipment for controlling HVAC equipment, lighting control equipment for controlling lighting equipment, and fresh air control equipment for controlling fresh air equipment.

[0074] Based on the above prediction results, this step involves the edge computing node initiating its local decision-making logic. The node reads the locally stored device control strategy data, identifies the rule "temperature above 26℃ requires cooling," and calculates the optimal control parameters based on the constraint "the reduction should not exceed 2℃." For example, the air conditioning supply temperature setpoint is adjusted from 20℃ to 18℃, and a corresponding instruction is generated as a zone control instruction. This instruction is directly sent to the air conditioning unit controller and VAV terminal within the zone via the edge node's I / O interface, without needing to send a response back to the cloud. When the network is interrupted, the edge node can still independently complete the above perception, prediction, and control closed loop based on the latest locally cached model and strategy until the network is restored and the operation log is reported to the cloud.

[0075] In one embodiment, the zone control instructions include conventional energy-saving control instructions and / or energy consumption emergency protection control instructions.

[0076] Conventional energy-saving control commands refer to flexible control commands generated by edge computing nodes based on energy-saving targets when they determine that a building zone is in a low-load or inactive state (such as sparsely populated or unoccupied for extended periods). These commands aim to tap into energy-saving potential by reducing equipment operating power, increasing set temperatures, or shutting down unnecessary loads. They typically do not immediately affect the safe operating boundaries of equipment and belong to optimization control.

[0077] Energy consumption emergency protection and control commands refer to mandatory control commands generated by edge computing nodes to ensure the safety of equipment and systems when they detect abnormalities in predicted energy consumption data (such as predicted short-circuit overload, equipment stall, or extreme energy consumption spikes). These commands have the highest priority and typically involve operations such as cutting off power, forced shutdown, or limiting the maximum operating frequency; they belong to safety protection control.

[0078] Step S104 involves generating zoning control instructions for the building zone based on the device control strategy data deployed locally by each edge computing node and the operation and maintenance status prediction data. This includes: Step S1041: Obtain a preset personnel activity threshold and a preset duration; when it is determined that the predicted personnel activity status data is less than the preset personnel activity threshold and the duration is greater than or equal to the preset duration, generate a corresponding conventional energy-saving control instruction based on the predicted personnel activity status data and the locally deployed equipment control strategy data.

[0079] The preset personnel activity threshold refers to the critical value of personnel density or personnel flow defined in the equipment control strategy data to distinguish the "occupied / unoccupied" or "active / quiet" state of a space. For example, 0.05 people / square meter can be set as the threshold; below this value, the area is considered to be unoccupied or in a state of extremely low personnel activity.

[0080] The preset duration refers to the time window set to avoid the deep energy-saving strategy being accidentally triggered due to people leaving briefly (such as going to the restroom). Only when the activity level of personnel is below the preset threshold for a sustained period of time (such as 30 minutes) will the system determine that it has entered a stable energy-saving operating condition, thereby triggering the regular energy-saving control command.

[0081] In this step, the edge computing node first reads the preset personnel activity threshold (e.g., 0.02 people / square meter) and preset duration (e.g., 20 minutes) from the locally stored device control strategy data. The node monitors the personnel activity prediction data output by the personnel status prediction sub-model in real time. When the prediction data shows that the predicted personnel density in the area will be below 0.02 people / square meter for a period of time, and this low-load state is expected to last for more than 20 minutes, the node determines that the area has entered a stable "unattended" or "dormant" mode. At this time, the node generates routine energy-saving control commands based on the personnel activity prediction data and strategy data, such as adjusting the air conditioner setting temperature back to energy-saving mode (e.g., 28℃), turning off 50% of the lighting circuits in the area, or reducing the fresh air unit frequency to the lowest setting. By setting a duration filter, sensor misjudgments caused by brief personnel movement or obstruction are effectively avoided, preventing frequent device start-ups and shutdowns.

[0082] Step S1042: Obtain a preset energy consumption anomaly confidence threshold; when the anomaly confidence level corresponding to the energy consumption status prediction data is greater than the preset energy consumption anomaly confidence threshold, determine that the energy consumption status prediction data is abnormal; based on the energy consumption status prediction data and the locally deployed device control strategy data, generate a corresponding energy consumption emergency protection control command.

[0083] The preset energy consumption anomaly confidence threshold refers to the critical probability value or score limit used in a single-class classification detection network structure to determine whether energy consumption data deviates from the normal pattern. For example, in the isolated forest model, if the anomaly score of a sample point exceeds 0.8 (out of 1), or if the confidence level of its predicted value falling outside the normal range exceeds 95%, it is judged as an anomaly.

[0084] Anomaly confidence refers to the probability, in the output of the energy consumption state prediction sub-model, that the current or predicted operating condition deviates from the distribution of historical normal training data. A higher confidence level indicates a greater likelihood of anomalies and a stronger urgency for the system to take emergency protection measures.

[0085] In this step, the edge computing nodes perform security monitoring on the output of the energy consumption state prediction sub-model. The nodes read a preset energy consumption anomaly confidence threshold (e.g., 95%) and obtain the anomaly confidence level corresponding to the energy consumption state prediction data. Taking the isolated forest model as an example, the anomaly score output by the model represents the energy consumption anomaly confidence level. When the predicted data shows that the energy consumption trajectory in the short term exceeds the distribution range of historical normal data, and the anomaly confidence level is greater than 95%, the node determines that an energy consumption anomaly is about to occur. At this time, the node no longer executes the conventional energy-saving optimization logic, but immediately generates energy consumption emergency protection and control instructions based on the energy consumption state prediction data and strategy data.

[0086] This embodiment introduces a dual judgment mechanism of preset personnel activity threshold and preset duration, ensuring that regular energy-saving control commands are only triggered after the operating conditions are indeed stable. This achieves deep energy consumption reduction while avoiding mechanical wear and personnel discomfort caused by frequent equipment adjustments, thus improving the practicality of the energy-saving strategy and user experience. Secondly, by introducing a preset energy consumption anomaly confidence threshold and utilizing the single-class detection capability of the energy consumption state prediction sub-model, edge nodes can identify risks and generate energy consumption emergency protection control commands within seconds or even minutes before energy consumption anomalies actually occur, greatly shortening response time. Finally, since the generation and execution of emergency protection commands are completed entirely locally in a closed loop, without relying on remote decision-making in the cloud, even in extreme cases of network interruption, building equipment still possesses local "self-protection" capabilities, significantly enhancing the security, reliability, and autonomy of the green building operation and maintenance system.

[0087] In one embodiment, after the edge computing node determines that the energy consumption state prediction data is abnormal, step S1042 further includes: Step S10421: The edge computing node determines the energy consumption anomaly event corresponding to the energy consumption status prediction data based on the preset energy consumption anomaly judgment rule; and sends the energy consumption anomaly event to the cloud management platform.

[0088] Among them, the preset energy consumption anomaly judgment rules refer to the logical mapping table pre-set inside the edge computing node, which is used to transform the abstract anomaly confidence level into specific anomaly types. For example, the rule can be defined as "when the predicted current exceeds the rated value by 120% and continues to rise, it is judged as 'motor stall'; when the variance of the predicted chilled water flow exceeds the threshold, it is judged as 'pipeline cavitation'", which is used to translate numerical anomalies into machine-readable semantic events.

[0089] Energy consumption anomaly events refer to structured data packets output by edge computing nodes according to preset energy consumption anomaly judgment rules, which include specific anomaly type, occurrence partition, predicted occurrence time, and confidence level.

[0090] In this step, when an edge computing node determines that the predicted energy consumption data is abnormal based on the aforementioned steps (e.g., the predicted current of the chilled water pump is abnormally high based on the isolated forest model), the node does not simply execute a shutdown operation directly. Instead, it calls the preset energy consumption anomaly judgment rules stored locally for semantic parsing. For example, the rule base stipulates that "if the predicted current exceeds the limit and is accompanied by an increase in the predicted vibration value, it is defined as a 'bearing wear warning'." Based on this, the node transforms the numerical anomaly into a specific energy consumption anomaly event. Subsequently, the node immediately reports the energy consumption anomaly event to the cloud management platform via the MQTT protocol, while continuing to execute locally generated energy consumption emergency protection and control commands (such as frequency reduction operation) to prevent the fault from escalating.

[0091] Step S10422: When the cloud management platform receives the energy consumption anomaly event, it obtains the historical operation and maintenance data of all building zones of the target green building; based on the energy consumption anomaly event and the historical operation and maintenance data, it determines the root cause of the fault, generates an operation and maintenance work order, and pushes it to the operation and maintenance terminal to prompt the operation and maintenance personnel to carry out maintenance.

[0092] Root cause analysis refers to the process by which the cloud management platform, upon receiving an energy consumption anomaly event, performs in-depth logical reasoning using historical operational data from across the building. Unlike edge computing, which only focuses on localized anomalies, the cloud platform traces the initial source of the anomaly by correlating and analyzing historical data from different partitions and systems, rather than merely addressing superficial symptoms.

[0093] A maintenance work order is an electronic document automatically generated in the cloud based on the root cause analysis of a fault, containing details of the maintenance task. Its content typically includes the faulty device number, a description of the fault symptom, suspected root cause analysis, recommended maintenance measures, and suggested priorities, used to guide on-site maintenance personnel in performing accurate repairs.

[0094] Operation and maintenance terminals refer to mobile smart devices (such as smartphones and tablets) held by on-site operation and maintenance personnel. They are used to receive operation and maintenance work orders pushed from the cloud, support work order confirmation, navigation and positioning, maintenance process recording and result feedback, and realize digital closed-loop management of the operation and maintenance process.

[0095] In this step, upon receiving the energy consumption anomaly, the cloud management platform immediately initiates the fault diagnosis process. The cloud first retrieves historical operation and maintenance data for all building zones of the target green building, using fault tree analysis or knowledge graph reasoning to determine the root cause of the fault. It concludes that, for example, the root cause of the anomaly is not pump damage itself, but rather reduced heat exchange efficiency due to condenser scaling, leading to increased pump load. Based on this, the cloud automatically generates a maintenance work order and pushes it to the on-site maintenance personnel's terminal via API. After receiving a pop-up notification on their terminal, maintenance personnel can follow the work order instructions to perform targeted cleaning and maintenance on-site, thereby eliminating potential hazards. The entire process achieves seamless integration from edge prediction and cloud-based diagnosis to manual intervention.

[0096] This embodiment constructs a technical chain of "edge alarm - cloud diagnostics - work order push". By pre-setting energy consumption anomaly judgment rules, numerical anomalies are transformed into specific energy consumption anomaly events and reported to the cloud. The cloud, based on historical operation and maintenance data of the entire building, determines the root cause of the fault, overcoming the limitation of edge nodes only being able to perform local single-point judgments. It can identify complex causal relationships across systems and devices, thereby avoiding misjudgments and invalid maintenance work, and significantly improving the accuracy of fault location. Furthermore, the mechanism of automatically generating and pushing operation and maintenance work orders allows operation and maintenance personnel to intervene before or in the early stages of equipment failure, effectively extending equipment lifespan and reducing the risk of sudden downtime. This ensures both the immediate safety of equipment operation and improves the intelligence level and efficiency of the overall building operation and maintenance management.

[0097] In one embodiment, each edge computing node constrains the zoning control instructions of its building zone based on its local device health profile; the device health profile includes real-time device operation constraint information, which is used as an additional criterion for the edge computing node to generate zoning control instructions.

[0098] Equipment health records refer to dynamically updated digital history files created by the cloud management platform for each critical control and execution device within a building. These records not only contain static information such as factory parameters and installation date, but more importantly, they include dynamic data such as real-time operational constraints, including percentage of remaining service life, recommended maximum operating frequency, vibration tolerance threshold, recent maintenance records, and maintenance cycle countdowns, reflecting the actual health status of the equipment.

[0099] Real-time operating constraint information refers to the dynamic parameters in the equipment health record used to limit the current allowable operating range of the equipment. For example, for a wind turbine that has just experienced a minor fault, its constraint information may be updated to "maximum operating frequency shall not exceed 80%" or "continuous operating time shall not exceed 4 hours". These parameters serve as hard boundaries to limit the adjustment range of edge nodes and prevent the equipment from operating under overload while malfunctioning.

[0100] After step S10422, which generates an operation and maintenance work order through the cloud management platform and pushes it to the operation and maintenance terminal, the following steps are included: Step S104221: Receive the work order processing completion instruction from the operation and maintenance terminal through the cloud management platform and update the equipment health record.

[0101] A work order completion instruction refers to an electronic receipt sent from the maintenance terminal to the cloud management platform after maintenance personnel have completed repair, maintenance, or replacement operations on-site. This receipt is sent via interface confirmation or scanning of the device's QR code. This instruction marks the end of a maintenance intervention process and includes a completion code (such as "bearing replaced," "filter cleaned," or "no abnormalities found"), triggering a reassessment of the device's health status by the cloud.

[0102] In this step, after the maintenance personnel arrive on-site and complete the repair, they click the "Complete" button through the maintenance terminal to send a work order completion instruction to the cloud management platform. Upon receiving this instruction, the cloud automatically triggers the update logic of the equipment health record. For example, the "health level" of the chilled pump is reset from "sub-healthy" to "good", the original "vibration exceeding the standard" mark is cleared, the "recommended maximum operating frequency" is restored from 80% to 100%, and the "next maintenance mileage" is updated to the new cycle.

[0103] Step S104222: The updated device health record is sent to the edge computing node associated with the collaborative control command.

[0104] Edge computing nodes associated with coordinated control commands refer to those involved in generating coordinated control commands in the cloud, or those that need to receive updated information due to changes in the status of equipment within their jurisdiction. For example, if the cloud issues a coordinated command for chiller units in zone A, and the health status of these units changes after maintenance, not only do the edge nodes in zone A need to update their files, but the edge computing nodes in downstream zones such as zone B and zone C, which depend on this chiller system, may also need to receive updates synchronously to ensure the consistency of the global control strategy.

[0105] In this step, the cloud encapsulates the updated device health profile and proactively distributes it to the edge computing nodes associated with the device. Specifically, in addition to distributing the profile to the edge computing nodes that directly manage the malfunctioning device, if previous collaborative control commands involved other partitions that depend on the malfunctioning device (such as the office area on the second floor), the cloud will also synchronously distribute the updated profile to the edge computing nodes of those other partitions. Upon receiving the updated profile, each edge node immediately refreshes its local constraint parameters, removes the previously imposed frequency restrictions, and restores the device to full-performance operation under optimal conditions. For example, if an edge node detects that the device has been repaired, it immediately removes the 70% frequency restriction, allowing the regeneration of frequency commands up to 90% based on predicted load, thereby maximizing the system's control capabilities and energy-saving potential without sacrificing device lifespan.

[0106] This embodiment introduces real-time operational constraints of equipment at edge nodes as additional criteria for command generation. This allows control commands to balance energy efficiency targets with equipment safety, effectively preventing equipment from overloading or operating while malfunctioning due to excessive pursuit of energy conservation, thus extending equipment lifespan and reducing total lifecycle costs. Secondly, by receiving work orders and processing commands and updating equipment health records via the cloud, real-time linkage between maintenance operations and system control strategies is achieved. This ensures that changes in equipment status are immediately reflected in the control logic, eliminating control blind spots caused by untimely manual reporting. Furthermore, the updated equipment health records are distributed to the edge computing nodes associated with the collaborative control commands, ensuring consistency in equipment status perception across all relevant zones within the building. This avoids cross-zone control conflicts caused by information asymmetry (e.g., upstream equipment has been repaired but downstream flow is still restricted), improving the stability and effectiveness of global system collaboration.

[0107] In one embodiment, the equipment control strategy data includes the preset professional control rules and basic control parameters adapted to the operating characteristics of the corresponding building zone; the basic control parameters are determined based on the historical operation and maintenance data of the corresponding building zone and the preset professional control rules. The preset professional control rules include building comfort constraint parameters, time-of-use electricity price adjustment coefficients, and zone energy consumption upper limit constraint parameters for each building zone; the basic control parameters, the building comfort constraint parameters, the time-of-use electricity price adjustment coefficients, and the zone energy consumption upper limit constraint parameters are respectively used to constrain the zone control instructions to meet the equipment benchmark operating conditions, human comfort conditions, optimal peak-valley electricity price costs, and energy consumption limits.

[0108] Among them, the basic control parameters refer to the set of benchmark setpoints that characterize the normalized operation patterns of the corresponding building zone, determined by the cloud management platform based on historical operation and maintenance characteristic data (including personnel activity characteristic data, environmental characteristic data, and equipment energy consumption operation characteristic data) and pre-set professional control rules, through data mining or lightweight optimization algorithms. The basic control parameters are derived from the statistical patterns of the actual operation data of the zone, such as the "optimal air conditioning water supply temperature setpoint from 9:00 to 18:00 on weekdays" or the "optimal fresh air ratio under typical personnel density" obtained from historical data regression analysis, and have the characteristics of data-driven and adaptive adjustment.

[0109] Building comfort constraints refer to a set of rigid indicators in pre-defined professional control rules used to ensure the physiological and psychological comfort of indoor occupants. These typically include temperature setpoint range (e.g., 26℃±2℃), relative humidity range (e.g., 40%-60%), upper limit of CO2 concentration (e.g., 1000ppm), minimum fresh air volume (e.g., 30m³ / h·person), and lower limit of illuminance (e.g., 300lux). These parameters define inviolable comfort red lines at critical points, ensuring that any energy-saving or emergency control commands will not compromise the user experience.

[0110] The time-of-use electricity price adjustment coefficient refers to an economic leverage coefficient preset in the control rules based on the peak-valley-flat electricity price policy issued by the local power supply bureau. This coefficient is typically assigned a value based on time periods. For example, during peak electricity price periods (such as 10:00-14:00), the coefficient is 1.2, encouraging edge nodes to prioritize reducing unnecessary loads; during off-peak electricity price periods (such as 23:00-7:00), the coefficient is 0.8, allowing edge nodes to appropriately store cold or heat. This coefficient guides edge nodes to automatically find the control time with the lowest operating costs while meeting comfort constraints.

[0111] The zonal energy consumption limit constraint parameter refers to the maximum allowable energy consumption threshold (e.g., kW·h / ㎡·day) set for each building zone to ensure the overall carbon emission target of the building or the energy consumption budget set by the owner. This parameter serves as a decomposition indicator of the energy consumption limit. When an edge node predicts that the energy consumption of its zone is about to reach this limit, it will automatically trigger a mandatory load reduction strategy to ensure that local operation conforms to the global energy-saving goal.

[0112] The equipment baseline operating condition refers to the stable operating state of the equipment under normal operating conditions (such as extreme weather or full-capacity meetings), based on historical operation and maintenance characteristic data. The basic control parameters are set to maintain this data-driven normal operating condition. For example, the average operating frequency of the air conditioning unit and the typical head of the water pump are derived from historical data. These parameters serve as a reference for edge nodes to determine whether the equipment is operating "healthily".

[0113] Specifically, the cloud platform extracts historical operation and maintenance characteristic data for each building section within the target green building over the past quarter. Taking a core office area as an example, the cloud platform extracts historical personnel density data, indoor and outdoor temperature data, and air conditioning unit operating frequency data for that area from 9:00 AM to 6:00 PM on weekdays. Using correlation analysis algorithms, the cloud platform discovered that when the indoor personnel density is maintained at 0.1 people / square meter and the outdoor temperature is 30°C, setting the air conditioning unit's water supply temperature to 7.5°C and coordinating it with a fan frequency of 45Hz can achieve the lowest energy consumption while meeting indoor temperature requirements. Based on this historical data statistical pattern, the cloud platform determines "water supply temperature 7.5°C, fan frequency 45Hz" as the basic control parameters for this section during this period. This process is entirely based on actual operating data, rather than theoretical calculations, and therefore better reflects the actual energy consumption characteristics of this section.

[0114] The cloud platform encapsulates the basic control parameters determined therefrom along with the various constraint parameters in the preset professional control rules. Among them, the building comfort constraint parameter is set as "temperature 24-26℃, CO2 <1000ppm"; the time-of-use electricity price adjustment coefficient is set as "peak period (10:00-12:00) coefficient 1.5, valley period (00:00-8:00) coefficient 0.7" according to the local peak-valley electricity price policy; and the zoned energy consumption upper limit constraint parameter is set as "daily average energy consumption not exceeding 20kW·h / ㎡".

[0115] During the edge computing node's operation, when a node generates a zone control command, it uses the basic control parameters issued by the cloud as the benchmark. For example, in the core office area case mentioned above, the edge node receives real-time personnel density data (0.09 people / square meter) and temperature data (25.5℃) at 10:30 AM (peak electricity price period). The node first retrieves the locally stored basic control parameters (water supply temperature 7.5℃, fan frequency 45Hz). Subsequently, the node introduces the time-of-use electricity price adjustment coefficient (1.5). Due to the high electricity price period, the node attempts to make energy-saving offsets based on the basic control parameters, such as trying to increase the water supply temperature to 7.8℃. However, before execution, the node must verify the building comfort constraint parameters to confirm that the predicted indoor temperature will not exceed the upper limit of 26℃ after the temperature increase. At the same time, the node calculates the current energy consumption rate to ensure that it will not touch the zone's energy consumption upper limit constraint parameter (20kW·h / ㎡). Finally, if the predicted temperature is still within the comfort range and the energy consumption does not exceed the standard, the node generates the command: "Set water supply temperature 7.8℃, fan frequency 42Hz". This instruction inherits the best operating benchmark derived from historical data, responds to real-time electricity price signals, and operates safely within the constraints of comfort and energy consumption.

[0116] This embodiment determines basic control parameters based on historical operation and maintenance characteristic data, truly reflecting the dynamic characteristics and implicit patterns of building zones in actual operation. This makes the control benchmark of edge nodes more closely match actual operating conditions, improving control accuracy and energy-saving potential from the source. Correspondingly, the generation process of zone control instructions is a multi-objective optimization process based on established experience benchmarks: it respects the best practices of historical operation (basic control parameters) while meeting current comfort requirements (comfort constraints), economic demands (electricity price coefficient), and management indicators (energy consumption ceiling). Finally, since the basic control parameters are derived from historical data, the cloud can periodically update them using the latest historical data, enabling the entire system to continuously evolve and adapt to long-term changes such as building aging and changes in personnel behavior, maintaining long-term operation and maintenance efficiency.

[0117] In one embodiment, each building zone is equipped with energy consumption devices, sensing terminals, and control execution devices; the energy consumption devices establish communication connections with the control execution devices; and the sensing terminals and control execution devices establish communication connections with the edge computing nodes of their respective building zones. The energy-consuming equipment includes at least one of heating and ventilation equipment, lighting equipment, and fresh air equipment; the control and execution equipment includes heating and ventilation control equipment for controlling heating and ventilation equipment, lighting control equipment for controlling lighting equipment, and fresh air control equipment for controlling fresh air equipment; the conventional energy-saving control command is used to control at least one of reducing the cooling load of heating and ventilation equipment, reducing the output brightness of lighting equipment, or reducing the valve opening of fresh air equipment; the energy consumption emergency protection control command is used to execute at least one of forcibly reducing the operating load of heating and ventilation equipment, shutting down unnecessary lighting equipment pre-defined in the zone, or reducing the output power of fresh air equipment.

[0118] Energy-consuming equipment refers to the general term for terminal equipment within a building zone that directly consumes electrical energy, heat energy, or other forms of energy to perform specific functions. In this embodiment, it specifically includes HVAC equipment (such as air conditioning units and fan coil units), lighting equipment (such as LED lamps and light strips), and fresh air equipment (such as fresh air handling units, total heat exchangers, and damper actuators), which are the direct targets of energy consumption control.

[0119] Sensing terminals refer to sensors or smart meters deployed within building zones to collect environmental and equipment operating status data. These include, but are not limited to, temperature and humidity sensors, CO2 sensors, illuminance sensors, human infrared sensors, smart meters, water flow meters, and vibration sensors, which are responsible for converting analog signals from the physical world into digital signals that can be recognized by edge computing nodes.

[0120] Control execution equipment refers to electromechanical devices that receive zone control commands issued by edge computing nodes and directly drive energy-consuming devices to change their operating states. In this embodiment, its type corresponds one-to-one with the energy-consuming devices, including HVAC control equipment, lighting control equipment, and fresh air control equipment.

[0121] HVAC control equipment specifically refers to actuators used to control the operating parameters of HVAC equipment. For example, after receiving instructions from an edge node, it can control the outlet air temperature of the air conditioning unit by adjusting the opening of the chilled water electric regulating valve, or change the fan speed by adjusting the output frequency of the inverter, thereby achieving precise control of the indoor thermal and humidity environment.

[0122] Lighting control equipment specifically refers to actuators used to control the luminous intensity of lighting equipment. For example, after receiving instructions from an edge node, it can adjust the brightness by changing the current flowing through the LED lamp via a DALI dimming module, or achieve switching control by cutting off / connecting the power supply to the lighting circuit via a relay.

[0123] Fresh air control equipment specifically refers to actuators used to control the air volume and energy recovery efficiency of fresh air systems. For example, after receiving instructions from edge nodes, it controls the amount of fresh air supplied to the room by adjusting the opening ratio of the fresh air valve, return air valve, and exhaust air valve, or it changes the fan speed by adjusting the frequency of the fresh air unit's inverter.

[0124] This embodiment specifies conventional energy-saving control commands as reducing HVAC cooling load, lowering lighting brightness, and reducing the opening of fresh air valves. It also specifies emergency energy protection control commands as forcibly reducing load, shutting off unnecessary lighting, and reducing output power, establishing an effective equipment safety defense system. By employing differentiated control methods for different subsystems such as HVAC, lighting, and fresh air, the energy-saving potential and safety assurance capabilities of each system are fully utilized, achieving a dynamic balance between safety, comfort, and economy in building energy consumption management.

[0125] In one embodiment, the sensing terminal includes an environmental acquisition terminal, a personnel sensing terminal, and a power acquisition terminal; The environmental data acquisition terminal is used to collect environmental characteristic data, including temperature, humidity, CO2 concentration, and light intensity. The personnel sensing terminal is used to collect personnel activity characteristic data, and the personnel activity status data includes personnel presence status and pedestrian flow. The power acquisition terminal is used to collect equipment energy consumption operation characteristic data, which includes active power of lighting circuit, active power of air conditioning circuit, and real-time current data.

[0126] Among them, the environmental data acquisition terminal refers to sensing devices specifically designed for monitoring the physical parameters of the indoor and outdoor environment of buildings, including but not limited to high-precision temperature and humidity sensors, CO2 sensors, and silicon photovoltaic light meters. This terminal is primarily used to collect environmental characteristic data, providing environmental context for the personnel status prediction sub-model and providing data benchmarks for comfort verification.

[0127] A people sensing terminal refers to a sensing device specifically designed to detect the presence and movement of people within a building space. Specific devices may include passive infrared detectors, millimeter-wave radar, cameras, or WiFi probes. The personnel activity characteristic data output by this terminal is the core input driving the personnel status prediction sub-model, used to determine space occupancy and personnel movement trajectories.

[0128] A power data acquisition terminal refers to an intelligent instrument used to monitor the electrical parameters of energy-consuming equipment. This terminal can acquire electrical parameters such as voltage, current, active power, and power factor with high precision. It is a key device for obtaining data on the energy consumption operation characteristics of equipment and provides real-time load data for the energy consumption status prediction sub-model.

[0129] The presence status of personnel refers to the Boolean quantity or probability value output by the personnel sensing terminal, which characterizes whether there is personnel activity in a specific space. For example, "personnel" (True / 1) or "no one" (False / 0), or the occupancy probability expressed as a percentage (such as 85%), is the direct basis for edge nodes to determine whether to enter the unmanned energy-saving mode.

[0130] People flow refers to the number of people passing through a certain cross-section or entering a certain area per unit of time, which is usually calculated by people sensing terminals by statistically analyzing people's movement trajectories. For example, "5 people entering the door / minute, 3 people leaving the door / minute" is an important feature for the people status prediction sub-model to capture people gathering trends and predict short-term load changes.

[0131] The active power of a lighting circuit refers to the actual electrical energy consumed by lighting equipment per unit time (unit: kW), obtained by sampling and calculating the lighting distribution branch from the power acquisition terminal. This data directly reflects the real-time energy consumption level of the lighting system and is an important basis for the energy consumption status prediction sub-model to determine whether the lighting load is normal.

[0132] The active power of an air conditioning circuit refers to the real-time energy consumption value (unit: kW) obtained by the power acquisition terminal from sampling and calculating the air conditioning power distribution branches (including fans, compressors, etc.). This data not only reflects the energy consumption level of the air conditioning system, but is also directly related to the cooling / heating load, and is a core parameter for energy consumption prediction and fault diagnosis.

[0133] Real-time current data refers to the instantaneous effective value of the current (unit: A) obtained by the power acquisition terminal from sampling the power supply line of the equipment. Compared with active power, current data is more sensitive to mechanical faults in equipment (such as a surge in stall current caused by bearing jamming) and is a key triggering factor for generating energy consumption emergency protection and control commands.

[0134] This embodiment constructs a panoramic building operation status perception system by deploying sensing terminals covering three dimensions: environment, personnel, and electricity. First, high-precision temperature, humidity, CO2, and light data acquired through environmental acquisition terminals not only provide crucial contextual features for the personnel status prediction model but also enable edge nodes to verify in real time whether the control results conform to building comfort constraints, forming a negative feedback adjustment based on measured data and effectively avoiding drift errors that may arise from solely relying on model predictions. Second, the accurate acquisition of personnel presence and traffic data through personnel sensing terminals solves the problem of inaccurate "personnel presence" judgments in traditional building automation systems. This allows the personnel status prediction model to capture subtle patterns of personnel activity, supporting edge nodes in decisively implementing deep energy-saving strategies when no one is present and in advance pre-cooling and pre-heating when there is a surge in personnel, significantly improving the balance between user experience and energy efficiency. Third, real-time monitoring of active power in lighting circuits, active power in air conditioning circuits, and real-time current data through electricity acquisition terminals provides high-quality training and inference samples for the energy consumption status prediction model, enabling the model to keenly detect early signs of abnormal energy consumption. This fusion of multi-source data enables edge nodes to conduct comprehensive analysis based on the physical environment, human behavior, and electrical load simultaneously. It overcomes the limitations of a single data source, significantly improves the perception sensitivity, prediction accuracy, and control security of the entire energy consumption adaptive operation and maintenance system, and provides a reliable data foundation for the refined management of green buildings.

[0135] In one embodiment, the cloud management platform carries the corresponding version information each time it generates an edge control deployment data package.

[0136] Version information refers to the globally unique identifier assigned to the edge control deployment data package by the cloud management platform when generating the package. It is typically represented by an incrementing sequence number (e.g., V1.0, V1.1), a timestamp (e.g., 20260115_1430), or a hash checksum (e.g., MD5, SHA-1). This information is used to identify the current iteration batch of model and policy data, serving as the basis for the cloud to determine the age of edge-side software assets, and also preventing data packages from being tampered with or corrupted during transmission.

[0137] Step S1051: When the cloud management platform receives version information sent by the edge computing node, it determines whether the version information of the edge computing node is the latest version information; when the version information of the edge computing node is not the latest version information, it generates an update data packet based on the latest operation and maintenance status prediction model and device control strategy data; and sends the update data packet to the corresponding edge computing node; wherein, the update data packet is used to update the operation and maintenance status prediction model and device control strategy data corresponding to the version information of the edge computing node to the latest version.

[0138] An update data package refers to an incremental or full data file generated by the cloud management platform when it detects that the edge node version is outdated. Unlike the full data package used during initial deployment, an update data package can be an incremental file containing only differences in model parameters, or a complete replacement file containing the latest model and policies. This data package also carries the latest version information and a checksum, used to perform hot updates or cold restarts at the edge, enabling seamless upgrades of models and policies.

[0139] The "version information is not the latest version" message means that after receiving version information proactively reported by edge nodes, the cloud management platform compares the reported version number and finds that it is outdated compared to the version number currently managed by the cloud. This usually occurs after the cloud has completed a new round of model training (e.g., using last month's data to optimize prediction accuracy) or after manual revision of preset professional control rules. This indicates that the decision-making basis currently used by the edge nodes is outdated and needs to be updated synchronously to maintain optimal system operation.

[0140] In this step, after receiving the version information of the edge computing node, the cloud immediately queries the latest version library maintained locally. Assuming the latest version on the cloud has been iterated to "Model_V2.3_Strategy_V1.5", the cloud determines that the edge node's version information is outdated. At this point, the cloud does not immediately distribute the large, complete model file. Instead, based on the latest operational status prediction model (V2.3) and device control strategy data (V1.5), it generates a streamlined update data package (e.g., a binary file containing only model weight differences). The cloud then distributes this update data package along with the latest version number to the corresponding edge computing node. In one embodiment, each edge computing node proactively reports its locally deployed version information after power-on, ensuring that the cloud promptly distributes the latest version of the model and strategy data to the edge computing node.

[0141] Step S1052: When the edge computing node receives the update data packet and passes the verification, it updates the local operation and maintenance status prediction model and equipment control strategy data based on the update data packet.

[0142] Verification success refers to the series of integrity and legality checks performed by the edge computing node after receiving the update data packet from the cloud. Specifically, this includes verifying the digital signature of the data packet (anti-forgery), calculating whether the file hash value is consistent with that on the cloud (anti-tampering), and checking whether the model file format is supported by the current inference engine (compatibility check). Only after all checks pass will the edge node perform the update operation; otherwise, it will discard the packet and report an error to the cloud, ensuring the stability of the system.

[0143] This embodiment effectively solves the challenge of iterative maintenance of edge-side models and policies by introducing version information and update data packet mechanisms. By proactively comparing version information in the cloud and sending update data packets instead of complete data packets, network bandwidth consumption is significantly reduced, lowering the risk of network congestion during the update process. This allows for smooth updates even in narrowband IoT (NB-IoT) or unstable network environments in buildings. A rigorous verification mechanism at the edge nodes ensures system security and stability during the update process, preventing edge node downtime due to unauthorized intrusion or data corruption, and safeguarding the network security of the building automation system.

[0144] This application also provides an energy-adaptive operation and maintenance system for green buildings, including a cloud management platform and several edge computing nodes deployed in various building zones of the green building: A cloud-based management platform is used to acquire historical operation and maintenance data of each building zone of the target green building and train an operation and maintenance status prediction model. The historical operation and maintenance data includes historical operation and maintenance feature data as training features and historical operation and maintenance status data as training labels. Based on the historical operation and maintenance data of each building zone and preset professional control rules, equipment control strategy data corresponding to each building zone is generated. Based on the operation and maintenance status prediction model and the equipment control strategy data of each building zone, an edge control deployment data package for each building zone is generated. The edge control deployment data package is sent to the corresponding edge computing node. An edge computing node is used to receive the edge control deployment data packet, deploy the operation and maintenance status prediction model and the device control strategy data; monitor the real-time operation and maintenance characteristic data of its respective building zone, input the real-time operation and maintenance characteristic data into the locally deployed operation and maintenance status prediction model, and obtain the operation and maintenance status prediction data output by the model; based on the locally deployed device control strategy data and combined with the operation and maintenance status prediction data, generate the zone control instruction for its respective building zone; and send the zone control instruction to the control execution device of the corresponding building zone to complete the local energy consumption device control of the zone.

[0145] It should be noted that the energy-adaptive operation and maintenance system for green buildings provided in the above embodiments is only illustrated by the division of the above functional modules when executing the energy-adaptive operation and maintenance method for green buildings. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the equipment can be divided into different functional modules to complete all or part of the functions described above. In addition, the energy-adaptive operation and maintenance system for green buildings and the energy-adaptive operation and maintenance method for green buildings provided in the above embodiments belong to the same concept, and the implementation process is detailed in the above embodiments of the energy-adaptive operation and maintenance method for green buildings, which will not be repeated here.

[0146] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of this application, and this application also intends to include these modifications and variations.

Claims

1. An energy consumption adaptive operation and maintenance method for green buildings, characterized in that, Includes the following steps: Historical operation and maintenance data of each building zone of the target green building are obtained through a cloud management platform, and an operation and maintenance status prediction model is trained and generated. The historical operation and maintenance data includes historical operation and maintenance feature data as training features and historical operation and maintenance status data as training labels. Based on the historical operation and maintenance data of each building zone and the preset professional control rules, generate equipment control strategy data for each building zone. Based on the operation and maintenance status prediction model and the equipment control strategy data of each building zone, the cloud management platform generates edge control deployment data packages for each building zone. The edge control deployment data packet is sent to the corresponding edge computing node; Each edge computing node receives the edge control deployment data packet, deploys the operation and maintenance status prediction model and the equipment control strategy data; monitors the real-time operation and maintenance characteristic data of the building partitions respectively, inputs the real-time operation and maintenance characteristic data into the locally deployed operation and maintenance status prediction model, and obtains the operation and maintenance status prediction data output by the model. By combining the device control strategy data based on local deployment of each edge computing node with the operation and maintenance status prediction data, the zone control instructions for the building zone are generated. The zone control command is sent to the control execution equipment of the corresponding building zone to complete the control of local energy consumption equipment in the zone.

2. The energy consumption adaptive operation and maintenance method for green buildings according to claim 1, characterized in that, The operation and maintenance status prediction model is an energy consumption status and personnel status prediction model; the operation and maintenance status prediction data includes personnel activity status prediction data and energy consumption status prediction data; the zone control instructions include regular energy-saving control instructions and / or energy consumption emergency protection control instructions. The steps for generating zone control instructions for a building zone based on the device control strategy data deployed locally by each edge computing node and the operation and maintenance status prediction data include: Obtain a preset personnel activity threshold and a preset duration; when it is determined that the predicted personnel activity status data is less than the preset personnel activity threshold and the duration is greater than or equal to the preset duration, generate a corresponding conventional energy-saving control instruction based on the predicted personnel activity status data and the locally deployed equipment control strategy data. Obtain a preset energy consumption anomaly confidence threshold; when the anomaly confidence level corresponding to the energy consumption status prediction data is greater than the preset energy consumption anomaly confidence threshold, determine that the energy consumption status prediction data is abnormal; based on the energy consumption status prediction data and the locally deployed device control strategy data, generate corresponding energy consumption emergency protection control instructions.

3. The energy consumption adaptive operation and maintenance method for green buildings according to claim 2, characterized in that, The equipment control strategy data includes the preset professional control rules and basic control parameters adapted to the operating characteristics of the corresponding building zones. The basic control parameters are determined based on the historical operation and maintenance data of the corresponding building zone and the preset professional control rules. The preset professional control rules include building comfort constraint parameters, time-of-use electricity price adjustment coefficients, and zoned energy consumption upper limit constraint parameters for each building zone. The basic control parameters, the building comfort constraint parameters, the time-of-use electricity price adjustment coefficient, and the zoned energy consumption upper limit constraint parameters are used to constrain the zoned control instructions to meet the equipment's baseline operating conditions, human comfort conditions, optimal peak-valley electricity price costs, and energy consumption limits, respectively.

4. The energy consumption adaptive operation and maintenance method for green buildings according to claim 2, characterized in that, The historical operation and maintenance characteristic data includes personnel activity characteristic data, environmental characteristic data, and equipment energy consumption operation characteristic data; the historical operation and maintenance status data includes personnel activity status data and energy consumption status data. The steps for training and generating an operation and maintenance status prediction model through a cloud management platform include: A personnel state prediction sub-model is trained using personnel activity characteristic data and environmental characteristic data as training features and personnel activity state data as training labels. The energy consumption operation characteristic data of the equipment is used as the training feature and the energy consumption status data is used as the training label to train the energy consumption status prediction sub-model. The personnel status prediction sub-model and the energy consumption status prediction sub-model are subjected to quantization and pruning to obtain the operation and maintenance status prediction model.

5. The energy consumption adaptive operation and maintenance method for green buildings according to any one of claims 2 to 4, characterized in that, Each building zone is equipped with energy consumption devices, sensing terminals, and control execution devices; the energy consumption devices establish communication connections with the control execution devices; the sensing terminals and control execution devices each establish communication connections with the edge computing nodes of their respective building zones; The energy-consuming equipment includes at least one of heating and ventilation equipment, lighting equipment, and fresh air equipment; The control execution equipment includes HVAC control equipment for controlling HVAC equipment, lighting control equipment for controlling lighting equipment, and fresh air control equipment for controlling fresh air equipment. The conventional energy-saving control command is used to control at least one of the following: reducing the cooling load of HVAC equipment, lowering the output brightness of lighting equipment, or reducing the valve opening of fresh air equipment. The energy consumption emergency protection control command is used to execute at least one of the following: forcibly reducing the operating load of HVAC equipment, shutting down unnecessary lighting equipment pre-defined in the zone, or reducing the output power of fresh air equipment.

6. The energy consumption adaptive operation and maintenance method for green buildings according to any one of claims 2 to 4, characterized in that, After the step of inputting the real-time operation and maintenance feature data into the locally deployed operation and maintenance status prediction model through each edge computing node to obtain the operation and maintenance status prediction data output by the model, the following steps are included: Each edge computing node sends the corresponding operation and maintenance status prediction data to the cloud management platform. Based on the operation and maintenance status prediction data of each edge computing node and the preset cross-partition collaborative control rules, the cloud management platform determines whether there is a need for cross-partition collaborative control. When it is determined that there is a need for cross-zone collaborative control, real-time operation and maintenance characteristic data of all building zones of the target green building are obtained; Based on the real-time operation and maintenance characteristic data of all building zones and the operation and maintenance status prediction data, a collaborative control instruction is generated and the corresponding edge computing node is determined; the collaborative control instruction is then sent to the corresponding edge computing node. When an edge computing node receives a collaborative control instruction, it modifies the zonal control instruction based on the collaborative control instruction and sends the modified zonal control instruction to the control execution device of the corresponding building zone to complete the control of local energy consumption devices in the zone.

7. The energy consumption adaptive operation and maintenance method for green buildings according to any one of claims 2 to 4, characterized in that, After the edge computing node determines that the energy consumption state prediction data is abnormal, the following steps are also included: The edge computing node determines the energy consumption anomaly event corresponding to the energy consumption status prediction data based on the preset energy consumption anomaly judgment rules; and sends the energy consumption anomaly event to the cloud management platform; When the cloud management platform receives the energy consumption anomaly event, it obtains the historical operation and maintenance data of all building zones of the target green building; based on the energy consumption anomaly event and the historical operation and maintenance data, it determines the root cause of the fault, generates an operation and maintenance work order, and pushes it to the operation and maintenance terminal to prompt the operation and maintenance personnel to carry out maintenance.

8. The energy consumption adaptive operation and maintenance method for green buildings according to claim 7, characterized in that, Each edge computing node constrains the zoning control instructions of its building zone based on its local device health records; the device health records include real-time device operation constraint information, which is used as an additional judgment condition for the edge computing node to generate zoning control instructions; After generating maintenance work orders through the cloud management platform and pushing them to the maintenance terminal, the following steps are included: The system receives work order completion instructions from the maintenance terminal via a cloud management platform and updates the equipment health records. The updated device health profile is sent to the edge computing node associated with the collaborative control command.

9. The energy consumption adaptive operation and maintenance method for green buildings according to claim 1, characterized in that, The cloud management platform carries the corresponding version information each time it generates an edge control deployment data package; The method further includes the following steps: When the cloud management platform receives version information sent by the edge computing node, it determines whether the version information of the edge computing node is the latest version. If the version information of the edge computing node is not the latest version, it generates an update data packet based on the latest operation and maintenance status prediction model and device control strategy data. The update data packet is then sent to the corresponding edge computing node. The update data packet is used to update the operation and maintenance status prediction model and device control strategy data corresponding to the version information of the edge computing node to the latest version. When the edge computing node receives the update data packet and passes the verification, it updates the local operation and maintenance status prediction model and equipment control strategy data based on the update data packet.

10. An energy consumption adaptive operation and maintenance system for green buildings, characterized in that, This includes a cloud management platform and several edge computing nodes deployed in various building zones of the green building: A cloud-based management platform is used to acquire historical operation and maintenance data of each building zone of the target green building and train and generate an operation and maintenance status prediction model; the historical operation and maintenance data includes historical operation and maintenance feature data as training features and historical operation and maintenance status data as training labels. Based on the historical operation and maintenance data of each building zone and the preset professional control rules, generate equipment control strategy data for each building zone. Based on the operation and maintenance status prediction model and the equipment control strategy data of each building zone, an edge control deployment data package for each building zone is generated. The edge control deployment data packet is sent to the corresponding edge computing node; An edge computing node is used to receive the edge control deployment data packet and deploy the operation and maintenance status prediction model and the device control strategy data; Real-time operation and maintenance characteristic data of the respective building zones are monitored, and the real-time operation and maintenance characteristic data are input into the locally deployed operation and maintenance status prediction model to obtain the operation and maintenance status prediction data output by the model. Based on locally deployed device control strategy data and combined with operation and maintenance status prediction data, generate zone control instructions for the building zone. The zone control command is sent to the control execution equipment of the corresponding building zone to complete the control of local energy consumption equipment in the zone.

Images