Building energy-saving control method, system, equipment and medium

Abstract

The present invention discloses a building energy-saving control method, system, device and medium, which relate to the field of energy-saving control technology, including the construction of a self-organizing multi-agent, wherein the first-level agent predicts future power demand values ​​through an LSTM deep learning model, uses a quadratic programming QP method to calculate the optimal set temperature value and the optimal heating and cooling equipment power value according to the energy-saving target, and the second-level agent uses a fuzzy logic algorithm to optimize power for different areas of the building. The method of the present invention combines the first-level agent and the second-level agent to enable global optimization and local fine-tuning of building energy-saving control, and further refines the building energy-saving control through the third-level agent, while ensuring the comfort of each room and minimizing energy waste. The building energy-saving control is further refined from the first-level agent and the second-level agent to the room level, realizing personalized energy-saving optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of energy-saving control, and in particular to a building energy-saving control method, system, equipment and medium. Background Art

[0002] Against the backdrop of increasingly stringent carbon emission regulations, optimizing building energy consumption has become an important research direction in the field of intelligent buildings. Traditional building energy-saving control methods mainly rely on schedule control, rule-based control based on set thresholds, or centralized energy management systems to optimize the energy consumption of heating and ventilation systems.

[0003] However, although the current deep learning-based load forecasting methods can provide relatively accurate energy demand forecasts, they usually use static models and are difficult to adapt to the dynamic changes in building energy consumption, especially in multi-region and multi-room environments. They lack precise personalized control capabilities. Secondly, in terms of regional energy consumption optimization, existing fuzzy logic algorithms usually rely on fixed membership functions and rules, which are difficult to cope with real-time environmental changes, resulting in insufficient flexibility in control strategies. In addition, building energy consumption anomaly detection in existing systems still relies on simple threshold judgments or anomaly detection based on traditional statistical methods, and lacks an efficient intelligent prediction mechanism, which makes it difficult to detect and deal with abnormal energy consumption problems in a timely manner. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a building energy-saving control method to solve the problem that although the current deep learning-based load forecasting method can provide relatively accurate energy demand forecasts, it usually adopts a static model and is difficult to adapt to the dynamic changes in building energy consumption, especially in a multi-area and multi-room environment, it lacks precise personalized control capabilities. Secondly, in terms of regional energy consumption optimization, existing fuzzy logic algorithms usually rely on fixed membership functions and rules, which are difficult to cope with real-time environmental changes, resulting in insufficient flexibility in the control strategy. In addition, the building energy consumption anomaly detection in the existing system still relies on simple threshold judgment or anomaly detection based on traditional statistical methods, and lacks an efficient intelligent prediction mechanism, which makes it difficult to detect and deal with abnormal energy consumption problems in a timely manner.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a building energy-saving control method, comprising:

[0008] A self-organizing multi-agent system was constructed. The first-level agent used an LSTM deep learning model to predict future power demand values ​​and a quadratic programming (QP) method to calculate the optimal set temperature and heating and cooling equipment power values ​​based on energy-saving goals. The second-level agent used a fuzzy logic algorithm to optimize power consumption for different building areas.

[0009] The three-level agent uses variational autoencoders (VAE) to process room data, and uses adaptive time series decomposition (ASTD) to perform short-term predictions of the collected data and optimize power for different rooms.

[0010] Construct an intelligent adjacency matrix with adaptive topology, calculate the interaction weights between agents, and dynamically adjust the agent connection weights based on the agent's optimization contribution. Update the adjacency matrix and optimization contribution to optimize the agent's communication frequency.

[0011] Use the Q-learning algorithm to update the strategy and calculate the similarity of the environment state data for the three-level intelligent agent to perform strategy optimization;

[0012] The equipment adjustment data is encrypted for transmission and storage, and identity verification is performed to detect energy consumption anomalies and issue anomaly warnings.

[0013] As a preferred solution of the building energy-saving control method of the present invention, wherein: the self-organizing multi-agent is constructed, wherein the first-level agent predicts the future power demand value through the LSTM deep learning model, uses the quadratic programming QP method to calculate the optimal set temperature value and the optimal heating and cooling equipment power value according to the energy-saving target, and the second-level agent uses the fuzzy logic algorithm to optimize the power for different areas of the building, including:

[0014] The self-organizing multi-agent includes a first-level agent, a second-level agent, and a third-level agent;

[0015] The first-level intelligent agent collects real-time energy consumption data of the building, including the energy consumption of heating, lighting, and ventilation systems, and collects the total electricity consumption of the building through smart meters and current sensors;

[0016] Use a pre-trained LSTM deep learning model to predict the building's future power load values ​​based on historical power load values;

[0017] Using a building thermodynamic model, the future power demand of heating and cooling equipment is predicted based on the specific heat capacity of air and the air mass flow rate.

[0018] The optimization objective is to minimize the power demand of heating and cooling equipment using the quadratic programming QP method;

[0019] Set the temperature range and power range of the heating and cooling equipment as constraints, and use the QP solver to calculate the optimal set temperature value and the optimal heating and cooling equipment power value;

[0020] Based on the camera device, the YOLOv5 target detection algorithm is used to detect the area and calculate the building area occupancy value;

[0021] Calculate the comprehensive lighting power based on area occupancy and lighting equipment startup power;

[0022] The secondary agent constructs a regional state model and uses a fuzzy logic algorithm to calculate temperature deviation values, humidity deviation values, and carbon dioxide concentration deviation values ​​for different areas of the building based on the differences between the optimal set temperature values, historical suitable humidity values, and historical suitable carbon dioxide concentration values ​​and the average temperature detection values, average humidity detection values, and average carbon dioxide concentration values ​​of different areas;

[0023] The fuzzy rules are set using a trapezoidal membership function: if the temperature deviation is high and the area occupancy is high, the heating and cooling equipment power value is increased; if the humidity deviation is high and the area occupancy is high, the heating and cooling equipment power value is increased; if the temperature deviation is low and the area occupancy is low, the ventilation flow rate is reduced; if the carbon dioxide concentration deviation is high and the area occupancy is high, the ventilation flow rate is increased;

[0024] The membership degree is calculated by Mamdani fuzzy reasoning method, the fuzzy center method is used to solve the fuzziness, and the control coefficient is calculated;

[0025] The power values ​​of regional heating and cooling equipment and regional ventilation flow values ​​are adjusted according to the control coefficients and transmitted to the third-level intelligent entities.

[0026] As a preferred solution of the building energy-saving control method of the present invention, the three-level intelligent agent uses a variational autoencoder (VAE) to process room data collection, and uses adaptive time series decomposition (ASTD) to perform short-term prediction of the collected data, and optimizes the power of different rooms, including:

[0027] The three-level intelligent agent collects room data based on the building rooms in the building area through room sensor data, including room temperature data, room humidity data, and room carbon dioxide concentration data;

[0028] Variational autoencoder (VAE) is used for data dimensionality reduction. Adaptive time series decomposition (ASTD) is used to construct the time series data of the room acquisition data. Empirical mode decomposition (EMD) is used to decompose the room acquisition data to obtain IMF components and residual terms. Wavelet denoising is performed on the IMF components of the room acquisition data, and the time series data of the acquisition data is reconstructed with the residual terms.

[0029] The denoised data of different collected data are used to form a denoised time series and a state space vector. The Kalman filter is applied for short-term prediction, and the collected data, including the predicted values ​​of room temperature data, room humidity data, and room carbon dioxide concentration data, are calculated separately. The room ventilation flow adjustment value is determined based on the fuzzy logic algorithm, and the temperature compensation value is calculated based on the heat loss of the room due to ventilation. Combined with the building thermodynamic model, the final room heating and cooling equipment power value is determined.

[0030] As a preferred solution of the building energy-saving control method of the present invention, wherein: the construction of the intelligent adjacency matrix of the adaptive topology, the calculation of the interaction weights between the intelligent agents, the dynamic adjustment of the intelligent agent connection weights according to the intelligent agent optimization contribution, the updating of the adjacency matrix and the optimization contribution, and the optimization of the communication frequency of the intelligent agents include:

[0031] Based on the three agents, an adaptive topology is constructed. The total number of secondary agents and the total number of tertiary agents are determined based on the three agents in the building. The interaction relationship between the agents is represented by an adjacency matrix A. In the adjacency matrix A, if two agents are directly connected, it is represented by 1, otherwise it is represented by 0.

[0032] Calculate the interaction weights between agents based on the temperature values ​​of adjacent rooms;

[0033] Calculate the agent optimization contribution based on the difference in energy consumption data optimized by the agent, and dynamically adjust the agent connection weight. Based on the historical threshold of the agent optimization contribution, distinguish the size of the agent's contribution to energy saving, and dynamically optimize the agent topology based on the optimization contribution of the corresponding agent.

[0034] Based on the sum of the mean and standard deviation of the agent connection weight as the weight threshold, if the agent connection weight is less than or equal to the weight threshold, the agent connection is disconnected and the adjacency matrix A is updated;

[0035] Recalculate the optimization contribution of the agent, calculate the interaction weight based on the ratio of the optimization contribution of a single agent to all agents, and calculate the communication frequency optimization value based on the product of the agent's baseline communication frequency and the interaction weight.

[0036] As a preferred solution of the building energy-saving control method of the present invention, wherein: the Q-learning algorithm is used to update the strategy, and the similarity of the environmental state data calculated by the three-level intelligent agent is used to optimize the strategy, including:

[0037] Using the Q-learning algorithm, the agent's control decisions are made by adjusting heating / cooling equipment and controlling ventilation. Temperature, humidity, and energy consumption are used as environmental states. The optimization contribution value is used as a calculation reward. Decision updates are made based on the control decisions and environmental states. A ϵ-greedy strategy is used for exploration to update the Q-learning value table.

[0038] For the three-level intelligent agents, the difference in environmental state data between adjacent intelligent agents is calculated as the similarity value, and the similarity threshold is determined based on historical data. If the similarity value of each environmental state data is less than or equal to the similarity threshold of the corresponding environmental state data, it is judged that the environmental states of the intelligent agents are similar, and the Q-learning value table is shared to obtain the optimized intelligent agent control strategy.

[0039] As a preferred solution of the building energy-saving control method described in the present invention, the encrypted transmission and storage of the equipment adjustment data and the authentication of the adjustment of the power value of the building heating and cooling equipment and the adjustment of the regional ventilation flow value by the intelligent control strategy adopt the AES-256 encryption algorithm for data encryption and storage, the TLS protocol for transmission encryption, the HMAC calculation of the identity authentication token, the identity authentication, and the generation of a log file.

[0040] As a preferred solution of the building energy-saving control method described in the present invention, the energy consumption anomaly detection and anomaly warning refers to using a pre-trained LSTM-autoencoder to predict the energy consumption data of the heating, lighting, and ventilation systems, and performing anomaly detection based on historical anomaly thresholds. If the data prediction value is greater than or equal to the anomaly threshold, it is judged as an energy consumption anomaly, and an early warning signal is sent to the maintenance terminal, and the predicted data of the energy consumption anomaly is logged.

[0041] In a second aspect, the present invention provides a building energy-saving control system, comprising:

[0042] Energy consumption data collection module, which collects energy consumption data of building heating, lighting and ventilation systems;

[0043] The power demand forecasting module uses the LSTM deep learning model to predict future power load demand, calculate the future power demand of heating and cooling equipment, and use the quadratic programming method to calculate the optimal set temperature value and the optimal heating and cooling equipment power value;

[0044] The regional energy consumption optimization module optimizes the power and ventilation flow of regional heating and cooling equipment based on fuzzy logic algorithms, calculates the deviations of temperature, humidity, and CO2 concentration, and uses trapezoidal membership functions to set fuzzy rules. The membership degree is calculated using the Mamdani fuzzy inference method, and the center of gravity method is used to defuzzify and calculate the control coefficient to optimize energy consumption.

[0045] The short-term prediction module uses adaptive time series decomposition to construct a time series of room-collected data and performs short-term predictions. It also uses Kalman filtering to perform short-term predictions of temperature, humidity, and carbon dioxide concentrations, optimize room-level ventilation control, calculate heat loss caused by ventilation, perform temperature compensation calculations, and ultimately determine the power value of room-level heating and cooling equipment.

[0046] The agent topology optimization module uses the adjacency matrix A to represent the interaction relationship between agents and dynamically adjusts the agent connection structure. It calculates the temperature changes of adjacent rooms, determines the agent interaction weights, improves the accuracy of data sharing, and dynamically adjusts the agent connection weights based on the optimization contribution to optimize the agent network topology.

[0047] The strategy optimization module uses the Q-learning reinforcement learning algorithm to enable the intelligent agent to autonomously learn the optimal heating, cooling, and ventilation control strategy, calculate the similarity of environmental states, and determine whether to share the Q-learning value table based on the similarity of temperature, humidity, and energy consumption data between adjacent intelligent agents;

[0048] Data security module, using AES-256 to encrypt storage devices to adjust data and TLS protocol to encrypt communication between agents;

[0049] The anomaly detection module uses LSTM-autoencoder to predict data, determines whether to trigger an early warning based on historical anomaly thresholds, and records anomaly prediction data.

[0050] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the building energy-saving control method described in the first aspect of the present invention is implemented.

[0051] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the building energy-saving control method as described in the first aspect of the present invention is implemented.

[0052] The beneficial effects of the present invention are: through the combination of the first-level intelligent agent and the second-level intelligent agent, the building energy-saving control has global optimization and local fine adjustment, and the third-level intelligent agent makes the building energy-saving control more refined, while ensuring the comfort of each room, minimizing energy waste, and making the building energy-saving control further refined from the first-level intelligent agent and the second-level intelligent agent to the room level, realizing personalized energy-saving optimization, and improving the energy utilization efficiency of the overall building management system. Through the dynamic adjustment strategy of the intelligent agent optimization contribution, the computing load of energy consumption optimization is more reasonably distributed. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0054] Figure 1 This is a flow chart of the building energy-saving control method in Example 1.

[0055] Figure 2 This is a schematic diagram of the structure of the building energy-saving control system in Example 2. DETAILED DESCRIPTION

[0056] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0057] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0058] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0059] Example 1, with reference to Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a building energy-saving control method, comprising the following steps:

[0060] S1: Build a self-organizing multi-agent system. The first-level agent uses an LSTM deep learning model to predict future power demand values ​​and a quadratic programming (QP) method to calculate the optimal set temperature and optimal heating and cooling equipment power values ​​based on energy-saving goals. The second-level agent uses a fuzzy logic algorithm to optimize power for different areas of the building.

[0061] Preferably, a self-organizing multi-agent is constructed, wherein the first-level agent uses the LSTM deep learning model to predict the future power demand value, and uses the quadratic programming QP method to calculate the optimal set temperature value and the optimal heating and cooling equipment power value according to the energy saving target. The second-level agent uses the fuzzy logic algorithm to optimize the power for different areas of the building, including:

[0062] Self-organizing multi-agents include first-level agents, second-level agents, and third-level agents;

[0063] The first-level intelligent agent collects real-time energy consumption data of the building, including the energy consumption of heating, lighting, and ventilation systems. It collects the total electricity consumption of the building through smart meters (measuring the energy consumption of each area) and current sensors (detecting the power load of equipment);

[0064] Use a pre-trained LSTM deep learning model to predict the building's future power load values ​​based on historical power load values;

[0065] The future power demand of heating and cooling equipment is predicted based on the specific heat capacity of air and the air mass flow rate through the building thermodynamic model, which is expressed as:

[0066] ;

[0067] in represents the power demand value of heating and cooling equipment at the predicted time t, is the specific heat capacity of air, Indicates the air flow rate, Indicates the set temperature. represents the room temperature at time t; the optimization objective is to minimize the power demand of heating and cooling equipment using the quadratic programming QP method, which is expressed as:

[0068] ;

[0069] in represents the optimization goal, T represents the total time, represents the power demand value of heating and cooling equipment at time t, express, Represents the human comfort index at time t (determined based on the Fanger thermal comfort model), Indicates the target PMV value (determined based on historical data);

[0070] Set the temperature range and power range of the heating and cooling equipment as constraints, and use the QP solver to calculate the optimal set temperature value and the optimal heating and cooling equipment power value;

[0071] Based on the camera device, the YOLOv5 target detection algorithm is used to detect the area and calculate the building area occupancy value;

[0072] Calculate the comprehensive lighting power based on area occupancy and lighting equipment startup power;

[0073] The secondary agent constructs a regional state model and uses a fuzzy logic algorithm to calculate the temperature deviation, humidity deviation, and carbon dioxide concentration deviation for different areas of the building based on the difference between the optimal set temperature value, historical suitable humidity value, and historical suitable carbon dioxide concentration value and the average temperature detection value, average humidity detection value, and average carbon dioxide concentration value of different areas;

[0074] The fuzzy rules are set using a trapezoidal membership function: if the temperature deviation is high and the area occupancy is high, the heating and cooling equipment power value is increased; if the humidity deviation is high and the area occupancy is high, the heating and cooling equipment power value is increased; if the temperature deviation is low and the area occupancy is low, the ventilation flow rate is reduced; if the carbon dioxide concentration deviation is high and the area occupancy is high, the ventilation flow rate is increased;

[0075] The membership degree is calculated by Mamdani fuzzy reasoning method, the fuzzy center method is used to solve the fuzziness, and the control coefficient is calculated;

[0076] The power values ​​of regional heating and cooling equipment and regional ventilation flow values ​​are adjusted according to the control coefficients and transmitted to the third-level intelligent entities.

[0077] Using LSTM to predict future power demand, heating / cooling strategies can be adjusted in advance, reducing energy waste. The QP method is used to calculate the optimal heating / cooling equipment power, ensuring that heating / cooling equipment meets comfort requirements with minimal energy consumption. Power load forecasting prevents equipment from overloading during peak hours, thereby reducing peak power demand and lowering building operating costs. The PMV thermal comfort model is used to calculate the human comfort index, ensuring that heating / cooling system adjustments do not cause excessively low or high temperatures, improving user experience. Fuzzy logic algorithms are used to calculate regional temperature, humidity, and CO2 deviations, enabling more reasonable heating / cooling power allocation for each area and avoiding unnecessary high-energy consumption. Mamdani fuzzy reasoning is used to automatically calculate the optimal heating / cooling power and ventilation flow without manual intervention, significantly reducing management costs and enhancing system intelligence. CO2 concentration detection and ventilation control strategies ensure that carbon dioxide concentrations do not exceed standards, improving air quality and reducing discomfort caused by excessive CO2. Cameras automatically detect occupant movement and dynamically adjust lighting power and HVAC equipment operation, reducing inaccuracies in human settings.

[0078] The combination of the first-level intelligent agent and the second-level intelligent agent enables building energy-saving control to have the dual capabilities of global optimization and local fine-tuning. The first-level intelligent agent provides prediction and control strategies for global building energy consumption optimization to ensure that the power demand of heating and cooling equipment is within a reasonable range, while the second-level intelligent agent performs fine-grained regulation of heating, ventilation and lighting systems based on the actual conditions of specific areas to improve regional energy utilization efficiency. Through this two-layer optimization architecture, the energy management of the entire building can not only achieve overall energy saving at the macro level, but also dynamically adjust each area at the micro level to adapt to the actual needs of different times, different occupancy conditions, and different environmental conditions. This method effectively reduces energy waste and reduces the overall energy consumption of the building, while ensuring the comfort of the indoor environment and improving the efficiency and accuracy of intelligent building management.

[0079] In S2, the three-level agent uses variational autoencoder (VAE) to process room data, and uses adaptive time series decomposition (ASTD) to perform short-term prediction of the collected data and optimize power for different rooms.

[0080] Preferably, the three-level agent uses variational autoencoder VAE to process room data collection, and uses adaptive time series decomposition ASTD to perform short-term prediction of the collected data, and optimizes the power of different rooms, including:

[0081] The third-level intelligent agent collects room data based on the building rooms in the building area through room sensor data, including room temperature data, room humidity data, and room carbon dioxide concentration data;

[0082] Variational autoencoder (VAE) is used for data dimensionality reduction. Adaptive time series decomposition (ASTD) is used to construct the time series data of the room acquisition data. Empirical mode decomposition (EMD) is used to decompose the room acquisition data to obtain IMF components and residual terms. Wavelet denoising is performed on the IMF components of the room acquisition data, and the time series data of the acquisition data is reconstructed with the residual terms, which can be expressed as:

[0083] ;

[0084] in represents the denoised data of the time t of the rth collected data, m and k represent the ending component index and the starting component index of the IMF component respectively, Represents the i-th component index of the r-th type of collected data, represents the residual term at time t of the rth type of collected data;

[0085] The denoised data of different collected data are used to form a denoised time series and a state space vector. The Kalman filter is applied for short-term prediction, and the collected data, including the predicted values ​​of room temperature data, room humidity data, and room carbon dioxide concentration data, are calculated separately. The room ventilation flow adjustment value is determined based on the fuzzy logic algorithm, and the temperature compensation value is calculated based on the heat loss of the room due to ventilation. Combined with the building thermodynamic model, the final room heating and cooling equipment power value is determined.

[0086] Through the intelligent agent through room-level sensor data collection, the internal environment of the building can be finely monitored, and the specific environmental conditions of different rooms can be independently optimized. By using variational autoencoders (VAE) for data dimensionality reduction, the system can effectively reduce the redundant information of high-dimensional data, while extracting key features, improving data processing efficiency and prediction stability. Through adaptive time series decomposition (ASTD), the three-level intelligent agent can construct time series data, making the collected room environmental variable data more predictable. In addition, empirical mode decomposition (EMD) is used to ensure the accuracy of short-term predictions. Through this series of data preprocessing methods, the three-level intelligent agent can effectively remove random interference and abnormal fluctuations in environmental monitoring data, making subsequent predictions and controls more accurate.

[0087] The three-level intelligent agent makes building energy-saving control more refined, while ensuring the comfort of each room, minimizing energy waste. Compared with the traditional fixed set temperature or regional average-based control method, the three-level intelligent agent can make personalized adjustments based on the real-time status of each room, which not only improves energy utilization, but also avoids local overcooling or overheating problems that may be caused by regional-level control. The three-level intelligent agent provides a set of efficient and intelligent solutions for data processing, prediction and room-level energy consumption optimization, which enables building energy-saving control to be further refined from the global (first-level intelligent agent) and regional (second-level intelligent agent) to the room level, realizing personalized energy-saving optimization and improving the energy utilization efficiency of the overall building management system.

[0088] S3, builds an intelligent adjacency matrix with adaptive topology, calculates the interaction weights between agents, and dynamically adjusts the agent connection weights based on the agent's optimization contribution. It also updates the adjacency matrix and optimization contribution to optimize the agent's communication frequency.

[0089] Preferably, an intelligent adjacency matrix with adaptive topology is constructed, the interaction weights between agents are calculated, and the agent connection weights are dynamically adjusted according to the agent optimization contribution, and the adjacency matrix and optimization contribution are updated to optimize the communication frequency of the agents, including:

[0090] Based on the three agents, an adaptive topology is constructed. The total number of secondary agents and the total number of tertiary agents are determined based on the three agents in the building. The interaction relationship between the agents is represented by an adjacency matrix A. In the adjacency matrix A, if two agents are directly connected, it is represented by 1, otherwise it is represented by 0.

[0091] The interaction weights between agents are calculated based on the temperature values ​​of adjacent rooms and are expressed as:

[0092] ;

[0093] in represents the interaction weight between the i-th and j-th agents, is the normalized parameter representing the temperature change, and denote the temperatures of the i-th and j-th adjacent rooms respectively;

[0094] The agent optimization contribution is calculated based on the difference in energy consumption data of the agent optimization, and the agent connection weight is dynamically adjusted. According to the historical threshold of the agent optimization contribution, the contribution of the agent to energy saving is distinguished, and the agent topology is dynamically optimized according to the optimization contribution of the corresponding agent, which is expressed as:

[0095] ;

[0096] ;

[0097] ;

[0098] in represents the optimization contribution of the i-th agent, represents the energy consumption value before optimization, Indicates the optimized energy consumption value, and They represent the connection weights of the agent ij at time t and time t+1 when the contribution is judged to be large, and They represent the connection weights of the agent ij at time t and time t+1 when the contribution is judged to be small, and Represents the learning rates for which the contribution is judged to be large and small, respectively (set based on empirical data);

[0099] Based on the sum of the mean and standard deviation of the agent connection weight as the weight threshold, if the agent connection weight is less than or equal to the weight threshold, the agent connection is disconnected and the adjacency matrix A is updated;

[0100] Recalculate the optimization contribution of the agent, calculate the interaction weight based on the ratio of the optimization contribution of a single agent to all agents, and calculate the communication frequency optimization value based on the product of the agent's baseline communication frequency and the interaction weight.

[0101] By constructing an intelligent adjacency matrix with adaptive topology, the interaction structure of the agents can be dynamically adjusted as the optimization contribution changes, enabling the system to adaptively optimize the energy consumption control strategy under different operating conditions and avoid the inefficient interaction problem caused by fixed topology. In the process of calculating the optimization contribution, the agents can be dynamically evaluated according to the energy consumption optimization effect, ensuring that resources are mainly allocated to the agents that can truly optimize the building energy consumption, reducing invalid calculations and data transmission, thereby improving the computing efficiency and real-time response capabilities of the entire system. At the same time, since the connections of low-contribution agents will be disconnected when the weight is lower than the threshold, the system can automatically streamline the network structure, reduce redundant computing overhead, and make building energy-saving control more efficient and flexible.

[0102] The dynamic adjustment strategy of the agent's optimization contribution enables a more reasonable distribution of the computing load for energy optimization. High-contribution agents undertake a larger proportion of computing tasks, while low-contribution agents reduce unnecessary calculations and data exchanges, thereby avoiding waste of computing resources. In addition, by optimizing the communication frequency, the data exchange frequency of each agent is no longer fixed, but is proportional to its optimization contribution. High-contribution agents can exchange information with neighbors more frequently to ensure the effectiveness of the optimization strategy, while low-contribution agents reduce the communication load, further improving the computing efficiency and stability of the system.

[0103] S4, uses the Q-learning algorithm to update the strategy and calculates the similarity of the environment state data for the three-level intelligent agents to perform strategy optimization;

[0104] Preferably, the Q-learning algorithm is used to update the strategy, and the similarity of the environment state data is calculated for the three-level intelligent agent to perform strategy optimization, including:

[0105] Using the Q-learning algorithm, the agent's control decisions are made by adjusting heating / cooling equipment and controlling ventilation. Temperature, humidity, and energy consumption are used as environmental states. The optimization contribution value is used as a calculation reward. Decision updates are made based on the control decisions and environmental states. A ϵ-greedy strategy is used for exploration to update the Q-learning value table.

[0106] For the three-level intelligent agents, the difference in environmental state data between adjacent intelligent agents is calculated as the similarity value, and the similarity threshold is determined based on historical data. If the similarity value of each environmental state data is less than or equal to the similarity threshold of the corresponding environmental state data, it is judged that the environmental states of the intelligent agents are similar, and the Q-learning value table is shared to obtain the optimized intelligent agent control strategy.

[0107] By updating its strategies through the Q-learning algorithm, the system enables the agent to autonomously learn optimal heating, cooling, and ventilation control strategies without relying on fixed rules or manual intervention. Because Q-learning uses temperature, humidity, and energy consumption as environmental conditions, and heating / cooling equipment adjustments and ventilation control as control decisions, the agent can continuously optimize its control strategies over the long term, enabling the system to adapt to changes in different building environments and improving the accuracy and adaptability of energy regulation.

[0108] By using the optimization contribution as a calculation reward, the intelligent agent can find the best balance between energy saving and comfort, rather than simply pursuing minimization of energy consumption or maximization of comfort, thereby avoiding the problems of over-cooling or over-heating that may be caused by traditional energy-saving control strategies. For the three-level intelligent agents, intelligent agents with similar environmental states can share the Q-learning value table, improving the stability and reliability of the optimization control strategy. At the same time, since the sharing of the Q-learning value table is limited to intelligent agents with similar environmental states, it avoids erroneous learning caused by irrelevant environmental states and improves the effectiveness of the Q-learning learning process.

[0109] S5, encrypts the transmission and storage of device adjustment data, performs identity authentication, detects abnormal energy consumption, and issues abnormal warnings;

[0110] Preferably, the device adjustment data is encrypted for transmission and storage, and authentication is performed. The adjustment of the power value of the building heating and cooling equipment and the adjustment of the regional ventilation flow value by the intelligent control strategy are encrypted and stored using the AES-256 encryption algorithm;

[0111] The TLS protocol is used for transmission encryption, and HMAC is used to calculate the authentication token for identity authentication and generate log files.

[0112] By encrypting the transmission and storage of device adjustment data, the system effectively prevents unauthorized access and data tampering, ensuring the integrity and reliability of the intelligent control strategy. Since adjustments to heating and cooling equipment power and zone ventilation flow rates directly impact building energy optimization and comfort control, data security is crucial. Using AES-256 encrypted storage ensures that even if data is intercepted or leaked, unauthorized access remains undecipherable. This prevents malicious tampering with device parameters that could lead to abnormal building system operation, ensuring the stability and security of the heating and cooling system.

[0113] Furthermore, energy consumption anomaly detection and anomaly warning refer to using pre-trained LSTM-autoencoders to predict the energy consumption data of heating, lighting, and ventilation systems, and performing anomaly detection based on historical anomaly thresholds. If the data prediction value is greater than or equal to the anomaly threshold, it is judged as energy consumption anomaly, and an early warning signal is issued to the maintenance terminal, and the predicted data of energy consumption anomaly is logged.

[0114] The energy consumption anomaly detection and early warning system can effectively improve the safety and reliability of building energy consumption management, enabling the system to promptly detect and respond to abnormal energy consumption, avoiding energy waste and potential losses caused by abnormal equipment operation. By using pre-trained LSTM-autoencoders for data prediction, the system can establish accurate prediction models based on long-term energy consumption patterns and improve the accuracy of anomaly detection.

[0115] This embodiment also provides a building energy-saving control system, including:

[0116] Energy consumption data collection module, which collects energy consumption data of building heating, lighting and ventilation systems;

[0117] The power demand forecasting module uses the LSTM deep learning model to predict future power load demand, calculate the future power demand of heating and cooling equipment, and use the quadratic programming method to calculate the optimal set temperature value and the optimal heating and cooling equipment power value;

[0118] The regional energy consumption optimization module optimizes the power and ventilation flow of regional heating and cooling equipment based on fuzzy logic algorithms, calculates the deviations of temperature, humidity, and CO2 concentration, and uses trapezoidal membership functions to set fuzzy rules. The membership degree is calculated using the Mamdani fuzzy inference method, and the center of gravity method is used to defuzzify and calculate the control coefficient to optimize energy consumption.

[0119] The short-term prediction module uses adaptive time series decomposition to construct a time series of room-collected data and performs short-term predictions. It also uses Kalman filtering to perform short-term predictions of temperature, humidity, and carbon dioxide concentrations, optimize room-level ventilation control, calculate heat loss caused by ventilation, perform temperature compensation calculations, and ultimately determine the power value of room-level heating and cooling equipment.

[0120] The agent topology optimization module uses the adjacency matrix A to represent the interaction relationship between agents and dynamically adjusts the agent connection structure. It calculates the temperature changes of adjacent rooms, determines the agent interaction weights, improves the accuracy of data sharing, and dynamically adjusts the agent connection weights based on the optimization contribution to optimize the agent network topology.

[0121] The strategy optimization module uses the Q-learning reinforcement learning algorithm to enable the intelligent agent to autonomously learn the optimal heating, cooling, and ventilation control strategy, calculate the similarity of environmental states, and determine whether to share the Q-learning value table based on the similarity of temperature, humidity, and energy consumption data between adjacent intelligent agents;

[0122] Data security module, using AES-256 to encrypt storage devices to adjust data and TLS protocol to encrypt communication between agents;

[0123] The anomaly detection module uses LSTM-autoencoder to predict data, determines whether to trigger an early warning based on historical anomaly thresholds, and records anomaly prediction data.

[0124] This embodiment also provides a computer device suitable for the building energy-saving control method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the building energy-saving control method proposed in the above embodiment.

[0125] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0126] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the building energy-saving control method proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0127] In summary, the present invention enables global optimization and local fine-tuning of building energy-saving control through the combination of first-level intelligent agents and second-level intelligent agents, and further refines building energy-saving control through third-level intelligent agents, while ensuring the comfort of each room and minimizing energy waste, so that building energy-saving control is further refined from first-level intelligent agents and second-level intelligent agents to the room level, realizing personalized energy-saving optimization and improving the energy utilization efficiency of the overall building management system. Through the dynamic adjustment strategy of the intelligent agent optimization contribution, the computing load of energy consumption optimization is more reasonably distributed.

[0128] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A building energy-saving control method, characterized in that: include: A self-organizing multi-agent system was constructed. The first-level agent used an LSTM deep learning model to predict future power demand values ​​and a quadratic programming (QP) method to calculate the optimal set temperature and heating and cooling equipment power values ​​based on energy-saving goals. The second-level agent used a fuzzy logic algorithm to optimize power consumption for different building areas. The three-level agent uses variational autoencoders (VAE) to process room data, and uses adaptive time series decomposition (ASTD) to perform short-term predictions of the collected data and optimize power for different rooms. Construct an intelligent adjacency matrix with adaptive topology, calculate the interaction weights between agents, and dynamically adjust the agent connection weights based on the agent's optimization contribution. Update the adjacency matrix and optimization contribution to optimize the agent's communication frequency. Use the Q-learning algorithm to update the strategy and calculate the similarity of the environment state data for the three-level intelligent agent to perform strategy optimization; Encrypted transmission and storage of equipment adjustment data, identity verification, abnormal energy consumption detection and abnormal warning; The intelligent adjacency matrix of the adaptive topology is constructed, the interaction weights between the intelligent agents are calculated, and the connection weights of the intelligent agents are dynamically adjusted according to the optimization contribution of the intelligent agents, and the adjacency matrix and the optimization contribution are updated to optimize the communication frequency of the intelligent agents. include, Based on the three agents, an adaptive topology is constructed. The total number of secondary agents and the total number of tertiary agents are determined based on the three agents in the building. The interaction relationship between the agents is represented by an adjacency matrix A. In the adjacency matrix A, if two agents are directly connected, it is represented by 1, otherwise it is represented by 0. Calculate the interaction weights between agents based on the temperature values ​​of adjacent rooms; Calculate the agent optimization contribution based on the difference in energy consumption data optimized by the agent, and dynamically adjust the agent connection weight. Based on the historical threshold of the agent optimization contribution, distinguish the size of the agent's contribution to energy saving, and dynamically optimize the agent topology based on the optimization contribution of the corresponding agent. Based on the sum of the mean and standard deviation of the agent connection weight as the weight threshold, if the agent connection weight is less than or equal to the weight threshold, the agent connection is disconnected and the adjacency matrix A is updated; Recalculate the optimization contribution of the agent, and calculate the interaction weight based on the ratio of the optimization contribution of a single agent to all agents, and calculate the communication frequency optimization value based on the product of the agent's baseline communication frequency and the interaction weight; The Q-learning algorithm is used to update the strategy and calculate the similarity of the environment state data for the three-level intelligent agent to perform strategy optimization, including: Using the Q-learning algorithm, the agent's control decisions are made by adjusting heating / cooling equipment and controlling ventilation. Temperature, humidity, and energy consumption are used as environmental states. The optimization contribution value is used as a calculation reward. Decision updates are made based on the control decisions and environmental states. A ϵ-greedy strategy is used for exploration to update the Q-learning value table. For the three-level intelligent agents, the difference in environmental state data between adjacent intelligent agents is calculated as the similarity value, and the similarity threshold is determined based on historical data. If the similarity value of each environmental state data is less than or equal to the similarity threshold of the corresponding environmental state data, it is judged that the environmental states of the intelligent agents are similar, and the Q-learning value table is shared to obtain the optimized intelligent agent control strategy.

2. The building energy-saving control method according to claim 1, wherein: The self-organizing multi-agent system is constructed, wherein the first-level agent uses the LSTM deep learning model to predict future power demand values, and uses the quadratic programming QP method to calculate the optimal set temperature value and the optimal heating and cooling equipment power value according to the energy saving target. The second-level agent uses the fuzzy logic algorithm to optimize the power of different areas of the building, including: The self-organizing multi-agent includes a first-level agent, a second-level agent, and a third-level agent; The first-level intelligent agent collects real-time energy consumption data of the building, including the energy consumption of heating, lighting, and ventilation systems, and collects the total electricity consumption of the building through smart meters and current sensors; Use a pre-trained LSTM deep learning model to predict the building's future power load values ​​based on historical power load values; Using a building thermodynamic model, the future power demand of heating and cooling equipment is predicted based on the specific heat capacity of air and the air mass flow rate. The optimization objective is to minimize the power demand of heating and cooling equipment using the quadratic programming QP method; Set the temperature range and power range of the heating and cooling equipment as constraints, and use the QP solver to calculate the optimal set temperature value and the optimal heating and cooling equipment power value; Based on the camera device, the YOLOv5 target detection algorithm is used to detect the area and calculate the building area occupancy value; Calculate the comprehensive lighting power based on area occupancy and lighting equipment startup power; The secondary agent constructs a regional state model and uses a fuzzy logic algorithm to calculate temperature deviation values, humidity deviation values, and carbon dioxide concentration deviation values ​​for different areas of the building based on the differences between the optimal set temperature values, historical suitable humidity values, and historical suitable carbon dioxide concentration values ​​and the average temperature detection values, average humidity detection values, and average carbon dioxide concentration values ​​of different areas; The fuzzy rules are set using a trapezoidal membership function: if the temperature deviation is high and the area occupancy is high, the heating and cooling equipment power value is increased; if the humidity deviation is high and the area occupancy is high, the heating and cooling equipment power value is increased; if the temperature deviation is low and the area occupancy is low, the ventilation flow rate is reduced; if the carbon dioxide concentration deviation is high and the area occupancy is high, the ventilation flow rate is increased; The membership degree is calculated by Mamdani fuzzy reasoning method, the fuzzy center method is used to solve the fuzziness, and the control coefficient is calculated; The power values ​​of regional heating and cooling equipment and regional ventilation flow values ​​are adjusted according to the control coefficients and transmitted to the third-level intelligent entities.

3. The building energy-saving control method according to claim 2, wherein: The three-level agent uses variational autoencoder (VAE) to process room data collection, and uses adaptive time series decomposition (ASTD) to perform short-term prediction of the collected data, and optimizes the power of different rooms, including: The three-level intelligent agent collects room data based on the building rooms in the building area through room sensor data, including room temperature data, room humidity data, and room carbon dioxide concentration data; Variational autoencoder (VAE) is used for data dimensionality reduction. Adaptive time series decomposition (ASTD) is used to construct the time series data of the room acquisition data. Empirical mode decomposition (EMD) is used to decompose the room acquisition data to obtain IMF components and residual terms. Wavelet denoising is performed on the IMF components of the room acquisition data, and the time series data of the acquisition data is reconstructed with the residual terms. The denoised data of different collected data are used to form a denoised time series and a state space vector. The Kalman filter is applied for short-term prediction, and the collected data, including the predicted values ​​of room temperature data, room humidity data, and room carbon dioxide concentration data, are calculated separately. The room ventilation flow adjustment value is determined based on the fuzzy logic algorithm, and the temperature compensation value is calculated based on the heat loss of the room due to ventilation. Combined with the building thermodynamic model, the final room heating and cooling equipment power value is determined.

4. The building energy-saving control method according to claim 3, wherein: The encrypted transmission and storage of the equipment adjustment data and the authentication of the device adjustment data refer to the adjustment of the power value of the building heating and cooling equipment and the adjustment of the regional ventilation flow value by the intelligent control strategy using the AES-256 encryption algorithm for data encryption and storage, the TLS protocol for transmission encryption, the HMAC calculation of the identity authentication token, the authentication, and the generation of a log file.

5. The building energy-saving control method according to claim 4, characterized in that: The energy consumption anomaly detection and anomaly warning mentioned above refers to using a pre-trained LSTM-autoencoder to predict the energy consumption data of the heating, lighting, and ventilation systems, and performing anomaly detection based on historical anomaly thresholds. If the data prediction value is greater than or equal to the anomaly threshold, it is judged as an energy consumption anomaly, and an early warning signal is sent to the maintenance terminal, and the predicted data of the energy consumption anomaly is logged.

6. A building energy-saving control system, based on the building energy-saving control method according to any one of claims 1 to 5, characterized in that: include, Energy consumption data collection module, which collects energy consumption data of building heating, lighting and ventilation systems; The power demand forecasting module uses the LSTM deep learning model to predict future power load demand, calculate the future power demand of heating and cooling equipment, and use the quadratic programming method to calculate the optimal set temperature value and the optimal heating and cooling equipment power value; The regional energy consumption optimization module optimizes the power and ventilation flow of regional heating and cooling equipment based on fuzzy logic algorithms, calculates the deviations of temperature, humidity, and CO2 concentration, and uses trapezoidal membership functions to set fuzzy rules. The membership degree is calculated using the Mamdani fuzzy inference method, and the center of gravity method is used to defuzzify and calculate the control coefficient to optimize energy consumption. The short-term prediction module uses adaptive time series decomposition to construct a time series of room-collected data and performs short-term predictions. It also uses Kalman filtering to perform short-term predictions of temperature, humidity, and carbon dioxide concentrations, optimize room-level ventilation control, calculate heat loss caused by ventilation, perform temperature compensation calculations, and ultimately determine the power value of room-level heating and cooling equipment. The agent topology optimization module uses the adjacency matrix A to represent the interaction relationship between agents and dynamically adjusts the agent connection structure. It calculates the temperature changes of adjacent rooms, determines the agent interaction weights, improves the accuracy of data sharing, and dynamically adjusts the agent connection weights based on the optimization contribution to optimize the agent network topology. The strategy optimization module uses the Q-learning reinforcement learning algorithm to enable the intelligent agent to autonomously learn the optimal heating, cooling, and ventilation control strategy, calculate the similarity of environmental states, and determine whether to share the Q-learning value table based on the similarity of temperature, humidity, and energy consumption data between adjacent intelligent agents; Data security module, using AES-256 to encrypt storage devices to adjust data and TLS protocol to encrypt communication between agents; The anomaly detection module uses LSTM-autoencoder to predict data, determines whether to trigger an early warning based on historical anomaly thresholds, and records anomaly prediction data.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the building energy-saving control method according to any one of claims 1 to 6 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the building energy-saving control method according to any one of claims 1 to 6 are implemented.

Images