Multi-agent-based optimal control system for building equipment

Abstract

The invention discloses a multi-agent-based optimal control system for building equipment. The system comprises an equipment layer, an edge layer and a cloud platform layer, the invention relates to the technical field of building energy equipment adjustment. According to the multi-agent-based optimal control system for the building equipment, through node division and acquisition of a personnel thermal image, judgment that the current node needs to be heated, cooled or maintained at the current temperature is achieved, a reference is provided for indoor temperature control, and through small-gradient and large-gradient temperature change control, the optimal control effect is achieved. The energy consumption analysis model is used for visually analyzing the energy consumption generated by adjusting the indoor environment temperature while effectively adapting to the indoor environment temperature change and reducing the irritation of the environment temperature change to the human body, flexible energy consumption regulation and control are performed in cooperation with the electricity price time period, flexible control conditions are provided for energy conservation and comfort level balance, and the energy-saving effect is improved. And optimal control of the intelligent equipment on the building is realized.

Description

Technical Field

[0001] This invention relates to the field of building energy equipment regulation technology, specifically to an optimized control system for building equipment based on multi-agent systems, used for the optimized control of air conditioning systems and fresh air systems. Background Technology

[0002] Multi-agent systems are an important branch of distributed artificial intelligence. With the development of building intelligence, building energy regulation and control systems have become important tools for energy conservation, emission reduction, and equipment maintenance. Traditional intelligent control methods for indoor environmental temperature have a relatively one-sided view of comfortable temperature positioning. They often simply determine the comfortable temperature based on the subjective feelings of humans in a large sample, failing to make targeted adjustments based on the actual situation of people. Furthermore, the indoor comfortable environmental temperature setting is relatively fixed, which can easily lead to large temperature differences. For example, in summer when the outside temperature is high, a relatively low cooling temperature is often set indoors to ensure comfort. Although this large temperature difference can eliminate heat, the large range of temperature changes is very detrimental to people's health, especially when people frequently switch between indoor and outdoor environments, which is quite stimulating to the human body.

[0003] To address this, a multi-agent-based optimized control system for building equipment is proposed. This system can determine whether heating or cooling is needed based on thermal images of people, and provide flexible control conditions for energy-saving requirements while meeting the temperature needs of people's comfortable environment through small and large gradient adjustments. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an optimized control system for building equipment based on multi-agent systems. This system solves the problem that traditional intelligent control methods for indoor ambient temperature have a relatively one-sided view of the temperature positioning of the comfortable environment, and the indoor comfortable environment temperature setting is relatively fixed, which easily leads to large temperature differences. Such large temperature differences are highly stimulating to the human body.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an optimized control system for building equipment based on multi-agent systems, comprising: Device Layer: The device layer communicates with the edge layer through a communication protocol. The device layer is used to collect environmental data, thermal images of people, and relevant data of electrical equipment. The environmental data includes outdoor temperature and humidity, and indoor temperature and humidity. Edge layer: The edge layer communicates with the cloud platform layer through a communication protocol. The edge layer has built-in edge control strategies, which are used to divide nodes and control the electrical equipment of the corresponding nodes. Cloud platform layer: The cloud platform layer has a built-in data analysis model and a dynamic strategy adjustment module. The data analysis model is used to generate device layer control logic based on preprocessed analysis data to perform gradient adjustment of node temperature. The dynamic strategy adjustment module optimizes the control strategy in real time through a deep Q network.

[0006] The present invention is further configured such that: the edge control strategy is used to perform node partitioning to obtain several functional areas, wherein the node partitioning methods include: The indoor area is divided into functional zones, with each individual functional zone as a node, and the equipment layer under that node consisting of all electrical equipment in that individual functional zone.

[0007] The present invention is further configured such that: the data analysis model includes an energy consumption analysis model and a gradient control model; The energy consumption analysis model is used to analyze the energy consumption when indoor temperatures and humidity are reached under different outdoor temperature and humidity conditions. The gradient control model is used to determine the comfortable temperature of the environment for personnel based on thermal images, and combines it with an energy consumption analysis model to perform small-gradient or large-gradient control of electrical equipment. Small-gradient control adjusts the temperature by 0.3-0.5℃, and large-gradient control adjusts the temperature by 1.2-1.5℃. Specific control methods include: During periods of high electricity prices, implement small-gradient control of electrical equipment with energy conservation as the goal: During periods of high electricity prices, electrical equipment is controlled with a large gradient to ensure comfort. During periods of low electricity prices, electrical equipment is controlled with a large gradient to ensure comfort. During periods of low electricity prices, small-gradient control of electrical equipment is implemented with energy conservation as the goal.

[0008] The present invention is further configured such that the deep Q-network in the dynamic policy adjustment module includes: State space: includes ambient temperature and humidity, equipment energy consumption, personnel thermal image features, current control strategy parameters and electricity price time period identifiers, as well as the control effect prediction features output by the data analysis model. The control effect prediction features include energy consumption prediction deviation rate and personnel comfort dynamic sensitivity. Action space: including temperature gradient adjustment, equipment start-up and shutdown, wind speed adjustment, fresh air system linkage and thermal imaging resolution switching, as well as the output of dynamic gradient coefficients, which are used to correct the basic adjustment range of the gradient control model; Its reward function is: In the formula, To reduce energy consumption, To achieve the maximum energy consumption reduction, Rate user comfort. The user comfort score is the maximum. For equipment operating costs, To maximize equipment operating costs, All are weighting coefficients; Q network update mechanism: The historical state-action-reward triplet is stored in the experience replay pool. Batch data is extracted every 5 minutes to update the Q network parameters. The target network synchronizes the weight of the main network once every 1 hour. In addition, node collaboration features are introduced during the update process. The node collaboration features include cross-node temperature uniformity deviation and correlation between device start-up and shutdown frequency.

[0009] The present invention is further configured such that the dynamic strategy adjustment module optimizes the control strategy in real time through a deep Q-network in the following ways: A1. The data analysis model generates an initial gradient control strategy, which includes small or large gradient control signals, and simultaneously outputs the control effect prediction features to the deep Q network. A2. The deep Q network takes the environmental state, historical data of gradient control strategy, and prediction features of control effect as input, and outputs optimized control actions and cross-node resource scheduling instructions. The cross-node resource scheduling instructions are used to coordinate the startup sequence of multi-node devices. A3. Feed the optimized actions back to the data analysis model; For the gradient control model, the basic adjustment amplitude is reconstructed through dynamic gradient coefficients; The energy consumption analysis model is modified to correct energy consumption prediction based on the nonlinear loss characteristics of the injection equipment. A4: When the temperature deviation exceeds the preset temperature deviation threshold within a unit of time, and the rate of change of the deviation exceeds the preset rate of change threshold, it is determined that a sudden environmental change has occurred. At this time, the Q network parameter update is frozen, and the temperature adjustment amplitude of the current gradient is increased according to a preset ratio. After the equipment has been running for a set period of time, if the rate of increase in energy consumption exceeds the preset energy consumption rate threshold within a unit of time, it is determined that a sudden environmental change has occurred. At this time, the Q network parameter update is frozen, and the temperature adjustment range of the current gradient is reduced by a preset ratio. A5: If a sudden change in the environment is detected, the deep Q network activates the fault-tolerant action space: while maintaining the core control objectives, the temporary equipment load buffer coefficient is increased, and normal optimization is restored after the environment stabilizes; A6: In multi-node scenarios, the global deep Q network integrates the optimization results of the local deep Q networks of each node, and corrects the total energy consumption calculation logic of the cloud platform data analysis model through the cross-node energy consumption transfer coefficient, thereby realizing complementary energy scheduling between nodes.

[0010] The present invention is further configured such that the thermal imaging resolution switching method includes: The structure reparameterization technique is used to perform real-time super-resolution reconstruction of low-resolution thermal images, and the system switches to low-resolution mode in emergency control mode.

[0011] The present invention is further configured such that the method for determining the environmental comfort temperature of personnel based on thermal images includes: Infrared thermal imagers are set up in nodes to collect thermal images of people and establish a mapping relationship between the thermal images of people and their identities. Based on the thermal images of personnel, regional temperatures are compared to determine the environmental temperature requirements for personnel to cool down or heat up. Based on the determination results, combined with the personnel's previous location, current location, and the current status of the air conditioning and fresh air systems, control logic for the air conditioning and fresh air systems is generated and sent to the corresponding nodes for control of the subordinate equipment layer.

[0012] The present invention is further configured such that the method for establishing the mapping relationship between personnel thermal images and personnel identities includes: The Canny edge detection algorithm was used to extract the human body outline from the thermal image, and the aspect ratio and torso proportion were calculated as human-specific thermal image feature parameters.

[0013] The present invention is further configured such that the method for determining the ambient temperature requirement for personnel to cool down or heat up includes: when At that time, it was determined that the personnel needed to be heated; when At that time, maintain the current temperature; In the formula, The ratio of temperature difference between the limbs and the trunk. This represents the average temperature of the hand. The average temperature of the feet. The surface temperature of the chest. This refers to the abdominal surface temperature. To determine the threshold for temperature increase; when At that time, it was determined that the personnel needed to be cooled down; when At that time, maintain the current temperature; In the formula, The face and neck temperature load index. It is a function of forehead surface temperature changing over time. Set the temperature for the air conditioning system. The rate of change of neck surface temperature. The weighting coefficient for the rate of change of neck temperature. The integration time interval, The threshold for determining cooling is used.

[0014] The present invention is further configured such that: after establishing the mapping relationship between personnel thermal images and personnel identities, different priority permissions are assigned to personnel; When there is a single high-priority user in the same node, the target is the ambient temperature requirement corresponding to the high-priority user. When several personnel at the current node share the same high-priority access, the environmental temperature requirements of all high-priority personnel at the current node are represented in a ternary form. Then, the environmental temperature requirement control target is determined. The determination method includes: In the formula, To quantify the indicator value, This represents the sum of the ternary representations of the environmental temperature requirements for all high-priority users at the current node. This represents the total number of all high-priority users on the current node. when At that time, it was determined that the ambient temperature requirement was to increase the temperature. when At that time, the ambient temperature requirement is determined to be to maintain the current temperature; when At that time, it was determined that the ambient temperature required cooling.

[0015] This invention provides an optimized control system for building equipment based on multi-agent systems. It offers the following advantages: (1) This invention uses node division and acquisition of thermal images of people to determine whether the current node needs to be heated, cooled or kept at the current temperature, providing a reference for indoor temperature control. Furthermore, through small and large gradient temperature change control, people can effectively adapt to changes in indoor ambient temperature, reducing the stimulation of ambient temperature changes on the human body. At the same time, through the energy consumption analysis model, the energy consumption generated by adjusting the indoor ambient temperature is analyzed intuitively. Combined with the electricity price period, flexible energy consumption regulation is carried out, providing flexible control conditions for balancing energy saving and comfort, and realizing the intelligent equipment's optimized control of the building.

[0016] (2) By setting priority, energy-saving target and comfort target, the present invention facilitates the system to be customized and improves the system flexibility. The setting of several functional area representative nodes provides convenience for the actual deployment of the system. Combined with the acquisition of thermal images for personnel identification, it effectively ensures the privacy of personnel. On the basis of accurately matching the subjective feelings of personnel and energy saving, it provides support for the optimized control of the comfort environment of building equipment.

[0017] (3) The dynamic strategy adjustment module of the present invention continuously optimizes the temperature gradient adjustment. When environmental changes occur, it can make adaptive energy consumption and comfort adjustments, and improve the resolution by reconstructing the thermal image to facilitate personnel identification. When environmental changes occur, thermal image reconstruction is no longer performed to ensure sufficient computing power and provide a guarantee for the long-term smooth and stable operation of the system. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the system architecture of the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0020] Please see Figure 1 The present invention provides the following technical solution: an optimized control system for building equipment based on multi-agent systems, as shown in the attached figure. Figure 1 As shown, it consists of a device layer, an edge layer, and a cloud platform layer. The device layer communicates with the edge layer through a communication protocol, and the edge layer communicates with the cloud platform layer through a communication protocol.

[0021] As a preferred option, the equipment layer is used to collect environmental data, thermal images of personnel, and relevant data of electrical equipment. Specifically, the equipment layer includes temperature sensors, humidity sensors, infrared thermal imagers, air conditioning systems, and fresh air systems. Among them, several sets of temperature and humidity sensors are set up and deployed indoors and outdoors respectively to collect outdoor temperature and humidity, indoor temperature and humidity as environmental data. The infrared thermal imager is a temperature-measuring type infrared thermal imager used to acquire thermal images of personnel. The electrical equipment is the air conditioning system and the fresh air system.

[0022] As a preferred solution, to achieve coordinated control of the air conditioning system and the fresh air system and provide a comfortable environment for indoor occupants, the edge layer includes a functional area division module and an edge control module. An edge control strategy is built into the edge layer, and this strategy is implemented based on the functional area division module and the edge control module. The edge control module is used to control the electrical equipment, and the functional area division module is used to divide nodes and obtain several functional areas. The node division methods include: The indoor area is divided into functional zones, with each individual functional zone as a node, and the equipment layer under that node consisting of all electrical equipment in that individual functional zone.

[0023] To further explain, after completing the division of functional areas into corresponding nodes, infrared thermal imagers are installed in the nodes to acquire thermal images of personnel within those nodes. After adding unique identifiers to each person, a mapping relationship is established between their thermal images and their identities. Specific methods include: The Canny edge detection algorithm was used to extract the human body outline from the thermal image, and the aspect ratio and torso proportion were calculated as human-specific thermal image feature parameters.

[0024] By identifying unique personnel information based on thermal image feature data, personnel identity can be recognized. After setting different priority permissions for personnel, the permission priority of the current personnel at the node can be obtained.

[0025] As a preferred solution, the cloud platform layer incorporates a data analysis model and a dynamic strategy adjustment module. The data analysis model is used to generate device-level control logic based on pre-processed analysis data to perform gradient adjustment of node temperature. The dynamic strategy adjustment module optimizes the control strategy in real time through a deep Q network and triggers an emergency control mode when the environmental load changes abruptly. The cloud platform layer includes an energy consumption analysis model, a gradient control model, and a database.

[0026] As a preferred option, the energy consumption analysis model is used to analyze the energy consumption of indoor systems when reaching the target temperature and humidity under different outdoor temperature and humidity environments. Since indoor spaces are mostly relatively enclosed spaces, taking enclosed spaces as an example, the energy consumption of the air conditioning system and the fresh air system when reaching the target temperature and humidity indoors under different combinations of outdoor temperature and humidity environments is collected and recorded as reference data. After obtaining a large number of samples, the missing data is supplemented by linear interpolation to generate the energy consumption analysis model. After obtaining the actual energy consumption data, the supplemented data is replaced.

[0027] As a preferred option, the gradient control module offers the advantage of flexible and personalized settings. Specifically, it determines the comfortable temperature of the environment based on thermal images of personnel and combines this with an energy consumption analysis model to implement small or large gradient control of electrical equipment. Small gradient control adjusts the temperature by 0.3-0.5℃, while large gradient control adjusts it by 1.2-1.5℃. Specific control methods include: During periods of high electricity prices, implement small-gradient control of electrical equipment with energy conservation as the goal: During periods of high electricity prices, electrical equipment is controlled with a large gradient to ensure comfort. During periods of low electricity prices, electrical equipment is controlled with a large gradient to ensure comfort. During periods of low electricity prices, small-gradient control of electrical equipment is implemented with energy conservation as the goal.

[0028] The above control method can be actively determined by the system administrator. The temperature response speed in this control method is shown in Table 1. Table 1 As a detailed explanation, the energy-saving mode uses the energy efficiency ratio as a reference to control the air conditioning system and the fresh air system.

[0029] As a detailed explanation, the methods for determining the environmental comfort temperature of personnel based on thermal images include: Based on thermal images of personnel, regional temperatures are compared to determine the required ambient temperature for personnel to cool down or heat up. The determination methods include: when At that time, it was determined that the personnel needed to be heated; when At that time, maintain the current temperature; In the formula, The ratio of temperature difference between the limbs and the trunk. This represents the average temperature of the hand. The average temperature of the feet. The surface temperature of the chest. This refers to the abdominal surface temperature. To determine the threshold for temperature increase; when At that time, it was determined that the personnel needed to be cooled down; when At that time, maintain the current temperature; In the formula, The face and neck temperature load index. It is a function of forehead surface temperature changing over time. Set the temperature for the air conditioning system. The rate of change of neck surface temperature. The weighting coefficient for the rate of change of neck temperature. The integration time interval, To determine the threshold for cooling; Based on the judgment results, combined with the personnel's previous location, current location, and the current status of the air conditioning system and fresh air system, the control logic for the air conditioning system and fresh air system is generated and sent to the corresponding nodes for control of the subordinate equipment layer.

[0030] As a preferred option, the deep Q-network in the dynamic policy adjustment module includes: State space: includes ambient temperature and humidity, equipment energy consumption, personnel thermal image features, current control strategy parameters and electricity price time period identifiers, as well as the control effect prediction features output by the data analysis model. The control effect prediction features include the energy consumption prediction deviation rate for the next 15 minutes output by the energy consumption analysis model based on the deviation between the simulated energy consumption value of the current gradient strategy and the actual value of the same historical scenario, and the dynamic sensitivity of personnel comfort output by the gradient control model based on the frequency of personnel body movement and the rate of change of face and neck temperature in the thermal image. Action space: includes temperature gradient adjustment, equipment start / stop, fan speed adjustment, fresh air system linkage, and thermal imaging resolution switching, as well as the output of dynamic gradient coefficient k, k∈[0.8,1.2]. The dynamic gradient coefficient is used to correct the basic adjustment range of the gradient control model (small gradient adjustment range = [0.3℃, 0.5℃]×k, large gradient adjustment range = [1.2℃, 1.5℃]×k). Its reward function is: In the formula, To reduce energy consumption, To achieve the maximum energy consumption reduction, Rate user comfort. The user comfort score is the maximum. For equipment operating costs, To maximize equipment operating costs, All are weighting coefficients. A quadratic term for energy consumption cost is added to penalize extreme cases of excessively high energy consumption, improving the model's ability to capture nonlinear relationships. These weighting coefficients are dynamically generated by the cloud platform layer data analysis model based on real-time scene characteristics: when the gradient control model detects a person's attention span > 70% based on the percentage of time their head remains stationary in the thermal image, Increased to 0.4 (baseline value 0.3). Reduce by 10%; when the energy consumption analysis model detects that the equipment efficiency degradation rate is >15%, Increased to 0.5 (baseline value 0.3); Q network update mechanism: The historical state-action-reward triplet is stored in the experience replay pool. Batch data is extracted every 5 minutes to update the Q network parameters. The target network synchronizes the weight of the main network once every 1 hour. In addition, node collaboration features are introduced during the update process. The node collaboration features include cross-node temperature uniformity deviation and correlation between device start-up and shutdown frequency.

[0031] To further explain, the dynamic policy adjustment module optimizes the control policy in real time through a deep Q-network in the following ways: A1. The data analysis model generates an initial gradient control strategy, which includes small or large gradient control signals, and simultaneously outputs the control effect prediction features to the deep Q network. A2. The deep Q network takes the environmental state, historical data of gradient control strategy, and prediction characteristics of control effect as input, and outputs optimized control actions and cross-node resource scheduling instructions. The cross-node resource scheduling instructions are used to coordinate the startup sequence of multi-node devices. The optimized control actions include dynamic gradient coefficients. A3. Feed the optimized actions back to the data analysis model: For the gradient control model, the basic adjustment amplitude is reconstructed by the dynamic gradient coefficient k; For energy consumption analysis models, nonlinear loss characteristics of equipment are injected, such as the efficiency decay rate of air conditioners under high load operation, to correct energy consumption prediction. A4: When the temperature deviation exceeds the preset temperature deviation threshold within a unit of time, and the rate of change of the deviation exceeds the preset rate of change threshold, it is determined that a sudden environmental change has occurred. At this time, the Q network parameter update is frozen, and the temperature adjustment amplitude of the current gradient is increased according to a preset ratio. After the equipment has been running for a set period of time, if the rate of increase in energy consumption exceeds the preset energy consumption rate threshold within a unit of time, it is determined that a sudden environmental change has occurred. At this time, the Q network parameter update is frozen, and the temperature adjustment range of the current gradient is reduced by a preset ratio. A5: If a sudden environmental change is detected, the deep Q network activates its fault-tolerant action space: while maintaining the core control objectives, a temporary "equipment load buffer coefficient" is added (e.g., the upper limit of the air conditioner compressor speed is reduced by 15%). Normal optimization is resumed after the environment stabilizes. The method for determining environmental stability is: the temperature deviation lasts for 10 minutes < 1℃ and the energy consumption increase rate < 2%·min. -1 ; A6: In multi-node scenarios, the global deep Q network integrates the optimization results of the local deep Q networks of each node, and corrects the total energy consumption calculation logic of the cloud platform data analysis model through the cross-node energy consumption transfer coefficient, thereby realizing complementary energy scheduling between nodes.

[0032] For example, if the temperature deviation exceeds the preset temperature deviation threshold within a unit of time, and the rate of change of the deviation exceeds the preset rate of change threshold, it is determined that a sudden environmental change has occurred. At this time, the Q-network parameter update is frozen, and the temperature adjustment amplitude of the current gradient is increased by a preset ratio. For example, if the temperature deviation is >2℃ and the rate of change of the deviation is greater than 0.5℃ / min, the temperature adjustment amplitude is increased to 120% of the current gradient. After the equipment has been running for a set period of time, if the rate of energy consumption increase exceeds the preset energy consumption increase rate threshold within a unit of time, a sudden environmental change is identified. At this time, the Q-network parameter update is frozen, and the temperature adjustment amplitude of the current gradient is reduced by a preset ratio. If the energy consumption increase rate is greater than 5%·min -1 If so, the temperature adjustment range is limited to 80% of the current gradient.

[0033] To further explain, the input to the deep Q network can also be extended to the following dimensions: ambient temperature and humidity, equipment energy consumption, thermal image features of personnel, electricity price time period identifiers, and the historical execution effect of the gradient control strategy. The historical execution effect of the gradient control strategy includes, but is not limited to, the rate of change of temperature adjustment amplitude and the energy consumption deviation rate.

[0034] As a preferred option, thermal imaging resolution switching methods include: The structure reparameterization technique is used to perform real-time super-resolution reconstruction of low-resolution thermal images. The reconstructed thermal images meet the following requirements: peak signal-to-noise ratio > 30dB and structural similarity > 0.85. In the emergency control mode, the system switches to low-resolution mode to reduce computing power consumption. The resolution switching command is dynamically generated by the deep Q network based on the edge layer computing power occupancy rate and the node data processing load.

[0035] As a detailed explanation, the method for determining the location of a person at the previous moment is as follows: obtain the previous time period closest to the time point when the thermal image of the current person is acquired, and determine the location of the person at the previous moment based on the person's identity information. Conversely, if the previous time period does not exist in any of the nodes, then it is determined that the person is in an outdoor environment.

[0036] Once the location of a person at the previous moment is determined, the ambient temperature at that location can be determined. Combined with the ambient temperature at the current location, if it is determined that the person needs to raise, lower, or maintain the current temperature of the environment, the ambient temperature can be adjusted accordingly.

[0037] As a preferred solution, in order to ensure indoor temperature control when multiple people are in the same node, when there is a single high-priority user in the same node, the target is the ambient temperature requirement of the high-priority user. When several personnel at the current node share the same high-priority access, the environmental temperature requirements of all high-priority personnel at the current node are represented in a ternary form. Then, the environmental temperature requirement control target is determined. The determination method includes: In the formula, To quantify the indicator value, This represents the sum of the ternary representations of the environmental temperature requirements for all high-priority users at the current node. This represents the total number of all high-priority users on the current node. when At that time, it was determined that the ambient temperature requirement was to increase the temperature. when At that time, the ambient temperature requirement is determined to be to maintain the current temperature; when At that time, it was determined that the ambient temperature required cooling.

[0038] Simulation Experiment Outdoor temperature and humidity, current indoor temperature, and thermal image data of three people entering the room from outdoors were collected. Based on the judgment method provided in this invention, it was determined whether the temperature needed to be raised, lowered, or maintained, and the indoor temperature was controlled accordingly. The results were compared with the actual feelings of the people. The experimental results are shown in Table 2. Table 2 As can be seen from Table 2, the present invention can determine whether a person's temperature is rising, falling, or maintaining the current temperature based on thermal images of the person.

[0039] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An optimization control system for construction equipment based on multi-agent, characterized by: The application relates to a multi-layered energy-saving and comfort control system for a building. The device layer communicates with the edge layer through a communication protocol, and is used for collecting environmental data, personnel thermal images and related data of power equipment, wherein the environmental data comprises outdoor temperature and humidity and indoor temperature and humidity. The edge layer communicates with the cloud platform layer through a communication protocol, and is internally provided with an edge control strategy, which is used for node division and control of corresponding node power equipment. The cloud platform layer is internally provided with a data analysis model and a dynamic strategy adjustment module, wherein the data analysis model is used for generating device layer control logic according to preprocessed analysis data, so as to realize gradient adjustment of node temperature, and the dynamic strategy adjustment module realizes real-time optimization of the control strategy through a deep Q network.

2. The multi-agent based optimized control system for construction equipment according to claim 1, characterized by: The edge control strategy is used for node division and acquisition of a plurality of functional areas, wherein the node division mode comprises the following steps: The indoor area is divided into functional areas, a single functional area is taken as a node, and a single functional area is taken as a device layer under the jurisdiction of the node.

3. The multi-agent based optimized control system for construction equipment according to claim 2, characterized by: The data analysis model comprises an energy consumption analysis model and a gradient regulation model. The energy consumption analysis model is used for energy consumption analysis when the indoor temperature and humidity reach a target temperature and humidity under different outdoor temperature and humidity environments. The gradient regulation model is used for personnel environmental comfort temperature judgment according to personnel thermal images, and small gradient or large gradient control of power equipment is realized in combination with the energy consumption analysis model, wherein the small gradient control is realized through 0.3-0.5 DEG C adjustment, the large gradient control is realized through 1.2-1.5 DEG C adjustment, and the specific control mode comprises the following steps: In a high electricity price period, small gradient control of power equipment is realized for the purpose of energy saving; In a high electricity price period, large gradient control of power equipment is realized for the purpose of comfort; In a low electricity price period, large gradient control of power equipment is realized for the purpose of comfort; In a low electricity price period, small gradient control of power equipment is realized for the purpose of energy saving.

4. The multi-agent based optimized control system for construction equipment according to claim 3, characterized by: The deep Q network in the dynamic strategy adjustment module comprises: A state space comprises environmental temperature and humidity, device energy consumption, personnel thermal image features, current control strategy parameters, electricity price period identification and regulation effect prediction features output by the data analysis model, and the regulation effect prediction features comprise energy consumption prediction deviation rate and personnel comfort dynamic sensitivity. An action space comprises temperature gradient adjustment, device start-stop, wind speed adjustment, fresh air system linkage and thermal imaging resolution switching, and dynamic gradient coefficient output, wherein the dynamic gradient coefficient is used for correcting the basic adjustment amplitude of the gradient regulation model. The reward function is: In the formula, is the energy consumption reduction amount, is the maximum energy consumption reduction amount, is the user comfort score, is the user comfort full score, is the equipment operation cost, is the maximum equipment operation cost, are all weight coefficients; A Q network updating mechanism adopts an experience replay pool to store historical state-action-reward triplets, a batch of data is extracted every 5 minutes to update Q network parameters, a target network synchronizes the weight of a main network once every hour, and a node cooperation feature is introduced in the updating process, wherein the node cooperation feature comprises cross-node temperature uniformity deviation and device start-stop frequency correlation.

5. The multi-agent based optimized control system for construction equipment according to claim 4, characterized by: The dynamic strategy adjustment module realizes real-time optimization of the control strategy through a deep Q network. A1, the data analysis model generates an initial gradient control strategy, including a small gradient or a large gradient control signal, and synchronously outputs a control effect prediction feature to a deep Q network; A2, the deep Q network takes an environment state, historical data of the gradient control strategy, and the control effect prediction feature as inputs, and outputs an optimized control action and a cross-node resource scheduling instruction, which is used to coordinate a multi-node device start timing; A3, the optimized action is fed back to the data analysis model; For the gradient control model, the basic adjustment amplitude is reconstructed by a dynamic gradient coefficient; For the energy consumption analysis model, a device nonlinear loss feature is injected to correct energy consumption prediction; A4: When the temperature deviation exceeds the preset temperature deviation threshold and the deviation rate exceeds the preset rate threshold in a unit time, it is judged that an environmental mutation occurs, at which time the Q network parameter update is frozen, and the temperature adjustment amplitude of the current gradient is increased by a preset proportion; After the device runs for a set length of time, when the energy consumption rate exceeds the preset energy consumption rate threshold in a unit time, it is judged that an environmental mutation occurs, at which time the Q network parameter update is frozen, and the temperature adjustment amplitude of the current gradient is reduced by a preset proportion; A5: If an environmental mutation is detected, the deep Q network activates a fault-tolerant action space: under the premise of maintaining the core control target, a temporary device load buffer coefficient is increased, and the normal optimization is restored after the environment stabilizes; A6: In a multi-node scenario, the global deep Q network integrates the optimization results of the local deep Q networks of each node, corrects the total energy consumption calculation logic of the cloud platform data analysis model through a cross-node energy consumption transfer coefficient, and realizes complementary scheduling of energy between nodes.

6. The multi-agent based optimized control system for construction equipment according to claim 5, characterized by: The thermal imaging resolution switching mode comprises: The low-resolution thermal image is reconstructed in real time by using a structure reparameterization technology, and is switched to a low-resolution mode in an emergency control mode.

7. The multi-agent based optimized control system for construction equipment according to Claim 6, characterized by: The personnel environmental comfort temperature judgment mode according to the personnel thermal image comprises: An infrared thermal imager is arranged in the node to collect personnel thermal images, and a mapping relationship between the personnel thermal images and the personnel identities is established; According to the personnel thermal images, the regional temperature is compared, the environmental temperature demand of the personnel needing cooling or heating is judged, the control logic of the air conditioning system and the fresh air system is generated according to the judgment result combined with the last time position, the current position, the current state of the air conditioning system and the fresh air system of the personnel, and is sent to the corresponding node to control the device layer.

8. The multi-agent based optimized control system for construction equipment according to claim 6, characterized by: The mapping relationship between the personnel thermal images and the personnel identities is established by using a Canny edge detection algorithm to extract the outer contour of the human body in the thermal image, calculating the height-width ratio and the trunk proportion as the personnel exclusive thermal image feature parameters. The personnel environmental temperature demand needing cooling or heating is judged by:

9. The multi-agent based optimized control system for construction equipment according to Claim 6, characterized by: After the mapping relationship between the personnel thermal images and the personnel identities is established, different priority permissions are prepared for the personnel; When the person is judged to need warming up; While the current temperature is maintained; wherein, is a temperature difference ratio of the limbs and the torso, is an average temperature of the hands, is an average temperature of the feet, is a chest surface temperature, is an abdominal surface temperature, is a temperature increase determination threshold value; When the person is judged to need cooling; When the current temperature is maintained; In the formula, is a face-neck temperature load index, is a function of forehead surface temperature change over time, is an air conditioning system set temperature, is a neck surface temperature change rate, is a neck temperature change rate weight coefficient, is an integral time interval, is a cooling judgment threshold.

10. The multi-agent based optimized control system for construction equipment according to Claim 7, characterized by: When there is a single high-priority personnel in the same node, the environmental temperature demand corresponding to the high-priority personnel is taken as the target; ​ When several personnel at the current node are the same high-priority permission personnel, the ambient temperature requirements of all high-priority permission personnel at the current node are represented by three values, and then the judgment of the ambient temperature requirement control target is performed, and the judgment methods include: In the formula, is the quantization index value, is the sum of the three-valued representation results of the environmental temperature requirements of all high-priority personnel at the current node, is the total number of all high-priority personnel at the current node; When the ambient temperature requirement is determined to be a need for temperature increase; When the ambient temperature requirement is determined to be to maintain the current temperature; When the ambient temperature requirement is determined to be a need for cooling.

Images