Artificial intelligence-based air purification device operation optimization control method and system

Abstract

The application discloses an air purification equipment operation optimization control method and system based on artificial intelligence, and relates to the technical fields of artificial intelligence and energy-saving control, and comprises the following steps: S1, collecting air purification equipment operation parameters and preprocessing; S2, calculating a purification response index and pairing the control amount according to time offset; S3, calculating a purification difference and an energy consumption difference through a double-loop differential strategy network and exchanging reverse channels; S4, performing layer-by-layer disturbance and cross-device coupling reorganization, and screening control combinations; S5, rolling updating an action position weight through multi-period counterfactuals and determining an execution sequence; and S6, correcting the response pairing, the channel exchange position and the action position weight according to the execution result. The application optimizes air purification equipment control parameters and execution sequences through a double-loop differential strategy network, cross-device response coupling and multi-period counterfactuals rolling, improves purification efficiency and control precision, and reduces operation energy consumption.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and energy-saving control technology, and in particular to an artificial intelligence-based method and system for optimizing the operation of air purification equipment. Background Technology

[0002] With the increasing demand for indoor air quality management, air purifiers, air circulators, energy-efficient gas compressors, energy-saving air conditioners, and household ventilation appliances are gradually forming a coordinated air treatment system. Existing air purification equipment typically judges the environmental condition based on sensor data such as particulate matter concentration, gas concentration, temperature and humidity, air volume, and equipment power, and adjusts fan speed, air volume, compressor load, and start / stop status according to preset thresholds. Some control schemes further incorporate artificial intelligence algorithms to predict pollution trends or equipment operating status, in order to reduce pollutant concentrations while taking into account operating energy consumption.

[0003] In actual operation, existing technologies often rely on a direct correlation between pollutant concentration and equipment control parameters for adjustment. They fail to adequately consider the response time lag between pollution reduction, filter resistance, airflow changes, and energy consumption changes after a control action occurs, making it difficult to accurately determine the actual purification effect corresponding to a certain control action. At the same time, different airflows, speeds, loads, and start-stop durations are usually optimized independently, lacking joint analysis of changes in purification benefits and energy consumption increases during the increase or decrease of control parameters. This can easily lead to the problem that as equipment operating intensity continues to increase, pollution reduction benefits gradually decrease.

[0004] Furthermore, when multiple air purifiers operate in tandem, the response intervals of different devices overlap and influence each other. Existing control methods rarely consider the coupling relationship of purification responses between devices, making it difficult to reasonably determine the control combination and execution sequence of different devices. Some cloud-based optimization methods mainly output fixed control strategies based on historical data. Changes in pollution drop, power rebound, and airflow response after execution at the edge side are difficult to be fed back to the control relationship and strategy parameters in a timely manner. This results in insufficient adaptability of the control results to filter aging, equipment load changes, and environmental disturbances, affecting air purification efficiency and energy-saving effects.

[0005] Therefore, how to provide methods and systems for optimizing the operation of air purification equipment based on artificial intelligence is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an artificial intelligence-based method and system for optimizing the operation of air purification equipment. This invention fully utilizes dual-loop differential strategy networks, cross-device response coupling, cloud collaboration, and edge computing technologies to achieve dynamic pairing of air purification equipment control quantities and purification responses, optimization of control combinations, and adaptive adjustment of execution order. It has the advantages of high purification efficiency, low energy consumption, and high collaborative control precision.

[0007] The air purification device operation optimization control method based on artificial intelligence according to an embodiment of the present invention includes the following steps: S1. Collect the operating parameters of the air purification equipment through various sensors and perform preprocessing; S2. Calculate the pollution reduction increment, filter element resistance hysteresis, air volume response time difference and energy consumption slope based on the pre-treated operating parameters, and pair the changes of each control quantity with the corresponding purification response according to the time offset. S3. Input the paired control quantity and purification response into the dual-loop differential strategy network. The inner loop calculates the purification difference corresponding to the increase or decrease of the control quantity, and the outer loop accumulates the energy consumption difference under different response time differences and swaps the feature channels at the difference inversion position. S4. Based on the purification difference and energy consumption difference, perform layer-by-layer positive and negative disturbances on the air volume, speed, load and start-stop duration of each air purification device, calculate the marginal response difference of adjacent disturbance nodes, and reorganize the disturbance nodes according to the cross-device response coupling relationship, and recursively retain the control combination with increasing purification benefits and decreasing energy consumption increase. S5, cloud collaboration performs multi-cycle counterfactual rolling on the retained control combination, writes the pollution drop amount and power rebound amount after the action shift into the action position weight, and edge computing determines the execution order of air purification equipment according to the position weight; S6. Based on the pollution drop, power rebound, and airflow response time difference after execution, correct the pairing relationship between the control quantity and the purification response, and adjust the channel exchange position and action position weights of the dual-loop differential strategy network.

[0008] Optionally, the air purification equipment includes an air purifier, an air circulation fan, an energy-saving gas compressor, an energy-saving air conditioner, and household ventilation appliances; the operating parameters include pollutant concentration, air intake volume, air outlet volume, fan speed, filter pressure difference, gas compressor load, supply air temperature, return air temperature, ventilation volume, operating power, and start / stop status; the preprocessing includes outlier removal, missing value interpolation, timestamp correction, multi-channel sampling alignment, sliding window segmentation, and amplitude normalization.

[0009] Optionally, S2 specifically includes: Based on the pre-processed operating parameters, control quantity channels are established using fan speed, gas compressor load, energy-saving air conditioner air volume, household ventilation appliance ventilation volume and start / stop status. The start and end points of control actions are determined based on continuous unidirectional changes and stable holding positions. A control action index is established by combining equipment identification, control quantity type and change direction. Divide the control action into a pre-action reference window and a post-action response window, centered on the control action start point, and calculate the pollution reduction increment, filter element resistance hysteresis, airflow response time difference, and energy consumption slope, including: The pollutant concentration decrease was calculated based on adjacent sampling points and accumulated by combining the changes in inlet and outlet air volume to obtain the pollution reduction increment. Pair the lift and fall filter elements with the same air intake position, and accumulate the difference to obtain the filter element resistance hysteresis. The airflow response time difference is obtained based on the time interval between continuous unidirectional changes in inlet and outlet airflow until it reaches 80% of its final stable amplitude. By comparing the rate of power change before and after the action, the energy consumption slope can be obtained. Using the air volume response time difference as the offset benchmark, the control action index is shifted backward along a unified time axis. The four types of response change positions are retrieved within the corresponding search interval, and the control quantity and purification response are matched based on the start time difference, duration overlap, change direction consistency, and amplitude change synchronization. The overlapping control actions are divided into independent segments and overlapping segments. The response change ratio corresponding to a single control quantity is determined based on the independent segment. The contribution of the identified control quantity is deducted item by item from the total response of the overlapping segment, and the independent response quantity of the remaining control quantity is separated to complete the time offset pairing of each control action and the purification response.

[0010] Optionally, S3 specifically includes: After completing the time offset pairing, the change amplitude, action direction, action duration, pollution reduction increment, filter resistance hysteresis, airflow response time difference, and energy consumption slope of the control quantity are encoded into action response units according to the control action index. A dual-loop differential strategy network is then constructed based on these action response units. This dual-loop differential strategy network consists of an inner loop differential chain, an outer loop time delay chain, an inter-loop switching layer, and a strategy output layer, wherein: The inner loop differential chain is a closed node chain formed by connecting the corresponding action response units of the same control quantity in the order of the action amplitude from decreasing to increasing; The outer loop time delay chain is a closed node chain formed by connecting action response units with the same direction of action in order of increasing air volume response time difference; The inter-ring switching layer is a connection layer that performs node feature exchange based on the sign reversal of the inner and outer ring differences. The strategy output layer is the output layer that performs joint sorting on the inner loop features and outer loop features after the switch and retains the differential results of the control actions; The inner loop differential chain groups the action response units according to the equipment identification and control quantity type, and arranges the nodes according to the control quantity change range. It performs differential on the pollution reduction increment and filter element resistance hysteresis of adjacent nodes, propagates the pollution reduction difference in the forward direction along the control quantity increase path, and propagates the pollution reduction difference in the reverse direction along the control quantity decrease path, and updates the purification weight recursively based on the resistance hysteresis difference. The outer loop time delay chain arranges the action response units in the same group according to the air volume response time difference, performs differential calculation on the energy consumption slope of adjacent nodes and continuously accumulates along the response time difference direction to obtain the cumulative energy consumption difference value corresponding to different response time differences; The inter-ring switching layer retrieves nodes where the pollution reduction difference and cumulative energy consumption difference reverse direction. It then aligns the pollution reduction difference, resistance hysteresis difference, and purification weight of the inner node with the cumulative energy consumption difference, energy consumption slope, and air volume response time difference of the outer node according to equipment identification, control quantity type, and action direction, and exchanges the corresponding characteristic channels at the difference reversal position. The differential is re-executed on the adjacent nodes after the exchange. When the direction of the purification difference is reversed, the switching position is moved along the direction of the control change and the calculation is recalculated until the direction of the purification difference of the consecutive nodes is consistent. The strategy output layer collects the purification difference, cumulative energy consumption difference, resistance hysteresis difference, air volume response time difference and switching position after the switch according to the control action index, and outputs the purification difference, energy consumption difference, difference reversal position and characteristic channel exchange result corresponding to each control quantity.

[0011] Optionally, the inter-ring switching layer specifically includes: Read the pollution reduction difference, resistance hysteresis difference and purification weight in the inner loop differential chain respectively, and read the cumulative energy consumption difference, energy consumption slope and air volume response time difference in the outer loop time delay chain. Establish the node correspondence relationship according to the equipment identification, control quantity type, action direction and control action index. Along the inner ring differential chain, nodes whose pollution reduction difference changes from positive to negative, from negative to positive, continuously increasing to continuously decreasing, and continuously decreasing to continuously increasing are searched. At the same time, along the outer ring time delay chain, nodes whose cumulative energy consumption difference reverses in the same direction are searched. The two types of reversed nodes are matched adjacently according to the wind volume response time difference from small to large. The pollution reduction difference, resistance hysteresis difference and purification weight corresponding to the matched internal node are recorded as purification characteristic channels, and the cumulative energy consumption difference, energy consumption slope and air volume response time difference corresponding to the external node are recorded as energy consumption characteristic channels. A switching relationship is established between the two types of channels that meet the same control action index. The switching relationship is to exchange the propagation paths of the purification feature channel and the energy consumption feature channel after the corresponding node at the reversal node position, so that the subsequent nodes of the inner ring inherit the energy consumption change state and the subsequent nodes of the outer ring inherit the purification change state. After the switch is completed, the purification difference and energy consumption difference of consecutive nodes before and after the switch position are recalculated. When the direction of the purification difference after the switch is consistent with that before the switch and the increase of the energy consumption difference decreases, the current switch position is retained. When the direction of the purification difference is reversed, one node is moved along the direction of the control quantity change and the switch is repeated until the direction of the purification difference of two consecutive nodes is consistent. The final switch node is recorded as the channel exchange position.

[0012] Optionally, S4 specifically includes: Read the purification difference, energy consumption difference, difference reversal position and characteristic channel exchange results corresponding to each control quantity, and determine the disturbance step size based on the current actual control value and the amplitude interval of adjacent control actions. Apply layer-by-layer positive and negative perturbations to the air purifier fan speed, air circulation fan volume, gas compressor load, energy-saving air conditioner air volume, household ventilation appliance ventilation volume, and equipment start-up and shutdown duration; Record the equipment identifier, control variable type, disturbance direction, disturbance amplitude, purification difference, and energy consumption difference corresponding to each disturbance as a disturbance node; For adjacent disturbance nodes with the same control quantity, calculate the purification marginal response and energy consumption marginal response respectively. Expand the disturbance nodes along the direction of continuous increase of purification marginal response, and delete the nodes where the purification marginal response decreases or the energy consumption marginal response increases continuously and the purification marginal response does not increase synchronously. Based on the pairing relationship between control quantity and purification response time offset, the response overlap interval of different equipment disturbance nodes is retrieved. Cross-equipment connections are established based on the overlapping changes of pollution reduction increment, air volume change, filter resistance hysteresis and energy consumption slope. Cross-equipment response coupling relationship is determined according to the response enhancement direction and response weakening direction. Based on the cross-device response coupling relationship, nodes connected in the same direction are merged according to the response start time, and nodes connected in the opposite direction are moved forward, backward, and deleted and rearranged to construct a joint perturbation branch. The positive and negative disturbance nodes of each control quantity are connected step by step along the joint disturbance branch. The repeated purification contribution caused by response overlap is deducted and the repeated energy consumption is corrected. The joint disturbance branch with increased purification marginal response and decreased energy consumption marginal response is recursively retained. Finally, the control combination is obtained by sorting according to the cumulative purification marginal response and the cumulative energy consumption marginal response.

[0013] Optionally, S5 specifically includes: The cloud-based collaboration establishes an action position sequence based on the control combination, and maps each control action to a continuous control cycle by combining the air volume response time difference and cross-device response coupling relationship; Using the current control combination as a benchmark, the individual control actions are shifted forward, shifted backward, and held in place. After fixing the remaining control quantities, a counterfactual action sequence is formed, and the predicted response state of the current cycle is pushed to the next control cycle to complete multi-cycle counterfactual rolling. Multi-cycle counterfactual rolling involves using the predicted response state generated by each counterfactual action sequence in the current control cycle as the starting state of the next control cycle, and continuously unfolding the action response process for multiple control cycles according to the same action position replacement rule. The purification difference and energy consumption difference are transmitted cycle by cycle along each counterfactual action sequence. The pollution reduction residual and cross-equipment repeated purification contribution are deducted to obtain the pollution reduction amount. At the same time, the power rebound amount is obtained by accumulating the advance power increment, delayed power release amount and power overlap amount. Compare the pollution drop increment and power rebound increment corresponding to different action positions, and perform pairwise position swaps for control actions with cross-device response coupling relationship, retaining the position relationship of increased pollution drop or maintained pollution drop and decreased power rebound. The position weight of each control action is calculated based on the position replacement result, continuous holding period and pair exchange result, and the control actions are rearranged in combination with the cross-device connection direction. Edge computing triggers control actions one by one according to the rearranged action position weights and air volume response time differences, performs staggered control on adjacent actions connected across devices in the opposite direction, and drives each air purification device to run in a determined execution sequence.

[0014] An artificial intelligence-based air purification device operation optimization control system according to an embodiment of the present invention includes: The data acquisition module is used to collect the operating parameters of the air purification equipment through various sensors and to preprocess the operating parameters; The response pairing module is used to calculate the pollution reduction increment, filter element resistance hysteresis, air volume response time difference and energy consumption slope based on the pre-processed operating parameters, and to pair the changes of each control quantity with the corresponding purification response according to the time offset. The differential strategy module is used to input the paired control quantity and purification response into the dual-loop differential strategy network. The inner loop calculates the purification difference corresponding to the increase or decrease of the control quantity, the outer loop accumulates the energy consumption difference corresponding to different response time differences, and the feature channels are exchanged at the difference reversal position. The disturbance calculation module is used to perform layer-by-layer positive and negative disturbances on the air volume, speed, load and start-stop duration of each air purification device based on the purification difference and energy consumption difference. It calculates the marginal response difference of adjacent disturbance nodes, reorganizes the disturbance nodes according to the cross-device response coupling relationship, and recursively retains the control combination with increasing purification benefits and decreasing energy consumption increase. The collaborative control module is used to perform multi-cycle counterfactual rolling on the retained control combination through cloud collaboration. It writes the pollution drop and power rebound after the control action shift into the action position weight, and determines the execution order of the air purification equipment based on the action position weight through edge computing. The feedback correction module is used to correct the pairing relationship between the control quantity and the purification response based on the pollution drop, power rebound, and airflow response time difference after the air purification equipment is executed, and to adjust the channel exchange position and action position weight of the dual-loop differential strategy network.

[0015] The beneficial effects of this invention are: First, this invention collects and preprocesses the operating parameters of air purification equipment in a unified manner, calculates the pollution reduction increment, filter resistance hysteresis, airflow response time difference, and energy consumption slope, and establishes a correspondence between control quantities and purification responses according to time offset. This can reduce control deviations caused by equipment response lag, filter status changes, and asynchronous multiple parameters, making the correlation between equipment control actions and actual purification effects more accurate, and providing a reliable data foundation for subsequent operation optimization.

[0016] Secondly, this invention utilizes a dual-loop differential strategy network to analyze the purification difference corresponding to the increase or decrease of control quantity and the energy consumption difference under different response time differences. It also identifies the turning point relationship between purification benefits and energy consumption changes by exchanging the characteristic channels at the difference reversal position. Based on this, it combines layer-by-layer positive and negative disturbances, marginal response difference calculation, and cross-device response coupling relationship to jointly optimize the air volume, speed, load, and start-stop duration of multiple air purification devices. This reduces repeated purification, energy consumption superposition, and mutual interference between devices caused by independent adjustment of a single device, thereby improving the purification benefits and energy utilization efficiency of the control combination.

[0017] Finally, this invention performs multi-cycle counterfactual rolling on the control combination through cloud collaboration. It adjusts the action position weights based on the pollution drop and power rebound after the control action shifts, and determines the actual execution sequence of each air purification device through edge computing. At the same time, it continuously corrects the pairing relationship between control quantity and purification response, the channel exchange position and action position weight of the dual-loop differential strategy network based on the execution results. This enables the control strategy to be dynamically updated with changes in pollution, equipment load and filter status, thereby improving the stability, energy saving and adaptive control accuracy of the collaborative operation of air purification devices. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of the artificial intelligence-based air purification equipment operation optimization control method proposed in this invention; Figure 2 This is a flowchart illustrating the operation parameter preprocessing and control response time offset pairing process of the AI-based air purification equipment operation optimization control method proposed in this invention. Figure 3 This invention presents a flowchart illustrating the dual-loop differential strategy network structure and feature channel switching process of the artificial intelligence-based air purification equipment operation optimization control method proposed in this invention. Figure 4 This is a flowchart illustrating the multi-device layer-by-layer perturbation and cross-device response coupling optimization of the air purification equipment operation optimization control method based on artificial intelligence proposed in this invention. Figure 5 The flowchart shows the cloud-edge collaborative multi-cycle counterfactual rolling and execution sequence determination process of the air purification equipment operation optimization control method based on artificial intelligence proposed in this invention. Figure 6 This is a flowchart illustrating the control response feedback correction and closed-loop update process of the AI-based air purification equipment operation optimization control method proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figures 1-6 An AI-based method for optimizing and controlling the operation of air purification equipment includes the following steps: S1. Collect the operating parameters of the air purification equipment through various sensors and perform preprocessing; S2. Calculate the pollution reduction increment, filter element resistance hysteresis, air volume response time difference and energy consumption slope based on the pre-treated operating parameters, and pair the changes of each control quantity with the corresponding purification response according to the time offset. S3. Input the paired control quantity and purification response into the dual-loop differential strategy network. The inner loop calculates the purification difference corresponding to the increase or decrease of the control quantity, and the outer loop accumulates the energy consumption difference under different response time differences and swaps the feature channels at the difference inversion position. S4. Based on the purification difference and energy consumption difference, perform layer-by-layer positive and negative disturbances on the air volume, speed, load and start-stop duration of each air purification device, calculate the marginal response difference of adjacent disturbance nodes, and reorganize the disturbance nodes according to the cross-device response coupling relationship, and recursively retain the control combination with increasing purification benefits and decreasing energy consumption increase. S5, cloud collaboration performs multi-cycle counterfactual rolling on the retained control combination, writes the pollution drop amount and power rebound amount after the action shift into the action position weight, and edge computing determines the execution order of air purification equipment according to the position weight; S6. Based on the pollution drop, power rebound, and airflow response time difference after execution, correct the pairing relationship between the control quantity and the purification response, and adjust the channel exchange position and action position weights of the dual-loop differential strategy network.

[0021] In this embodiment, S1 specifically includes: The pollutant concentration, filter element pressure difference, inlet air volume, outlet air volume, supply air temperature, return air temperature, operating power, and fan speed are collected by particulate matter sensor, gas concentration sensor, pressure sensor, air volume sensor, temperature sensor, current sensor, and speed sensor, respectively. The gas compressor load, ventilation volume, and start / stop status are read from the control interface of energy-saving gas compressor, energy-saving air conditioner, and household ventilation appliance. The equipment identifier, parameter identifier, sampling sequence number, and local timestamp are written for each operating parameter according to a unified sampling period. Using the air purifier control cycle as the reference time axis, the time offset between adjacent sampling points and the time difference across channels are continuously calculated for each sensing channel. For the leading channel, the sampling position is moved backward according to the cumulative offset. For the lagging channel, local interpolation is performed based on the slope of the change between adjacent sampling points. The operating parameters of multiple channels are mapped to a unified sampling time. The median deviation is used to detect continuous abrupt changes in the mapped operating parameters. Sampling points that exceed the local change boundary and are not accompanied by synchronous changes in the associated channels are marked as outliers, and replacements are completed based on the local change direction of the valid sampling points before and after the outliers. Extract the rate of change of air volume, power, and pollutant concentration before and after the missing interval, and select a linear recursive value to complete the missing value interpolation according to the consistent direction of the three types of rate of change. The pollutant concentration, filter pressure difference, air volume and operating power at adjacent sampling times are read sequentially along a unified time axis. The corresponding first-order difference is obtained by subtracting the previous sampling value from the current sampling value. At the same time, the start and stop states at adjacent sampling times are compared and the sampling position where the state changes is recorded as the start and stop state transition position. The sampling position where any differential direction is reversed or the start / stop state transitions is set as a candidate boundary point, and short-term repeated boundary points are merged according to the cross-channel change consistency between adjacent candidate boundary points to obtain an adaptive sliding window; Within each adaptive sliding window, amplitude normalization is performed based on the local median and local scale difference of each operating parameter. At the same time, the original direction, magnitude, and duration of change of pollutant concentration, filter pressure difference, air volume, power, and fan speed are preserved to obtain pre-processed operating parameters corresponding to the actual operating stage of the air purification equipment.

[0022] In this embodiment, S2 specifically includes: Based on the preprocessed operating parameters, the fan speed, gas compressor load, energy-saving air conditioner air volume, household ventilation appliance ventilation volume and start / stop status are used as control quantity channels, and the numerical changes and state transitions of adjacent sampling times are compared channel by channel. The starting point for the control action is set at a position where the control quantity maintains the same direction of change for three consecutive sampling points and the cumulative change amplitude reaches twice the local fluctuation amplitude of the corresponding channel. Set the control action endpoint at the position where the control variable change ends and three consecutive sampling points remain stable. A control action index is established based on the start point of the control action, the end point of the control action, the equipment identifier, the type of control quantity, and the direction of change. Taking the starting point of each control action as the center, a complete stable window is captured forward as the reference window before the action, and three equal-length response windows are captured backward as the response windows after the action. The median values ​​of pollutant concentration, inlet air volume, outlet air volume, filter pressure difference, and operating power in the reference window before the action are written into the corresponding control action index. For the pollutant concentration channel, adjacent sampling points in the response window after the action are read in chronological order. The concentration decrease is obtained by subtracting the concentration of the current sampling point from the concentration of the previous sampling point. The concentration decrease is multiplied by the increase in the air volume at the same time and then corrected according to the change in the air volume. Sampling points with rising concentration are recorded as reverse reduction amounts. The forward reduction amounts and reverse reduction amounts are accumulated in chronological order. The cumulative result after the action is subtracted from the cumulative result of the baseline window before the action to obtain the pollution reduction increment corresponding to each control action. For the filter element pressure difference channel, with the air intake volume as the matching benchmark, the air volume values ​​in the response window after the action are sorted from smallest to largest. For each air volume value, the air volume sampling point with the smallest difference is retrieved in the benchmark window before the action. The filter element pressure difference corresponding to the two sampling points is recorded in pairs. Pressure difference pairing sequences are established along the air volume rising stage and the air volume falling stage respectively. The difference between the rise pressure difference and the fall pressure difference at the same air volume position is calculated point by point and continuously accumulated according to the direction of air volume change to obtain the filter element resistance hysteresis corresponding to each control action. To address the airflow response time difference, starting from the first sampling point after the control action start point, the first-order difference between the inlet and outlet airflow is continuously read. When the inlet airflow difference and the outlet airflow difference simultaneously maintain the same direction as the change in the control quantity, and no direction reversal occurs for three consecutive sampling points, the first sampling point that meets the continuous consistency condition is determined as the airflow response start point. Then, the position where the airflow change amplitude first reaches 80% of the final stable amplitude of the response window after the action is searched backward from the airflow response start point. The time interval between the control action start point and this position is determined as the airflow response time difference. For the operating power channel, equal-length power segments are extracted before and after the control action start point. The power change at adjacent sampling times is calculated point by point and divided by the corresponding sampling interval. The median value of the power change rate before the action is taken as the reference slope. The power change rate after the action is weighted from near to far from the control action start point and weighted and accumulated. The difference between the weighted result after the action and the reference slope is obtained. After completing the calculation of the four types of response quantities, the air volume response time difference is used as the initial offset. Each control action index is shifted backward along the unified time axis. The response search intervals of two sampling points before and after the shift position are established. The positions where the pollution reduction increment begins to change continuously, the filter element resistance hysteresis occurs and the direction is maintained, the air volume response enters a stable change, and the energy consumption slope changes continuously are retrieved in sequence. The four types of response positions are traced back to the corresponding control action index. When the same response position corresponds to only one control action, the pairing relationship between the control quantity and the purification response is directly established. When the same response position corresponds to two or more control actions, the start time difference, duration overlap, change direction consistency, and amplitude change synchronization between each control action and the response change are calculated respectively. The control actions are then filtered in order of starting time difference from small to large and duration overlap from large to small. The first control action is retained as the main pairing object, and the remaining control actions are further matched with the unpaired response positions. For control actions that overlap, divide the overlapping segment into independent segment according to the actual start and end times of each control action; For independent sections, the change range of a single control quantity at the start and end positions of the section is paired with the pollution reduction increment, air volume change, filter pressure difference change and power change during the same period, and the response change ratio corresponding to the single control quantity is determined by the proportion of each response change to the total response change of the independent section. By using the response change ratio, the contribution of the identified control quantities is subtracted from the total response of the overlapping section item by item to obtain the independent response quantities corresponding to the remaining control quantities. Finally, for each control action, the device identifier, control quantity type, action direction, action amplitude, action duration, pollution reduction increment, filter resistance hysteresis, air volume response time difference and energy consumption slope are recorded. The time offset of each control quantity change and the corresponding purification response is matched according to the order of the control actions.

[0023] In this embodiment, S3 specifically includes: After completing the time offset pairing, the fan speed, gas compressor load, energy-saving air conditioner air volume, household ventilation appliance ventilation volume, start / stop status, and corresponding pollution reduction increment, filter resistance hysteresis, air volume response time difference, and energy consumption slope are arranged according to the control action index. The control quantity change amplitude, action direction, action duration, pollution reduction increment, filter resistance hysteresis, air volume response time difference, and energy consumption slope corresponding to a single control action are encoded into an action response unit. A dual-loop differential policy network is constructed based on the action response unit. The dual-loop differential policy network consists of an inner loop differential chain, an outer loop time delay chain, an inter-loop switching layer, and a policy output layer, wherein: The inner loop differential chain is a closed node chain formed by connecting the corresponding action response units of the same control quantity in the order of the action amplitude from decreasing to increasing; The outer loop time delay chain is a closed node chain formed by connecting action response units with the same direction of action in order of increasing air volume response time difference; The inter-ring switching layer is a connection layer that performs node feature exchange based on the sign reversal of the inner and outer ring differences. The strategy output layer is the output layer that performs joint sorting on the inner loop features and outer loop features after the switch and retains the differential results of the control actions; When constructing the inner loop differential chain, the action response units are grouped according to the equipment identification and control quantity type. Within each group, the nodes are arranged from small to large according to the change range of the control quantity. Adjacent nodes read the pollution reduction increment and filter element resistance hysteresis respectively. The pollution reduction difference and resistance hysteresis difference are obtained by subtracting the corresponding value of the previous node from the corresponding value of the current node. Nodes with the same action direction and continuous signs of pollution reduction difference are connected sequentially. The position where the action direction changes is connected to the first reverse action node in the same group, forming an inner loop differential chain with the control quantity increase path and the control quantity decrease path connected end to end. The pollution reduction difference is propagated forward along the inner ring differential chain according to the path of increasing control quantity, and in reverse along the path of decreasing control quantity. At each node, the current pollution reduction difference, the pollution reduction difference of the previous node, the direction of change of the pollution reduction difference between two adjacent nodes, and the number of consecutive same-direction changes are synchronously written. The filter element resistance hysteresis difference is used as the node constraint. When the pollution reduction difference continues to increase and the resistance hysteresis difference increases synchronously, the purification weight of the corresponding node is reduced. When the pollution reduction difference increases and the resistance hysteresis difference decreases, the purification weight of the corresponding node is increased. The purification difference corresponding to the increase or decrease of control quantity is obtained by accumulating the purification weights of consecutive nodes. When constructing the outer loop time delay chain, the action response units are recombined according to the equipment identification, action direction and control quantity type, and the nodes are arranged from short to long according to the air volume response time difference. The energy consumption slope of adjacent nodes is calculated point by point, and the short time difference nodes to long time difference nodes are connected in sequence. A loop edge is established between the node with the maximum air volume response time difference and the node with the minimum air volume response time difference to form the outer loop time delay chain. Starting from the shortest air volume response time difference node along the outer ring time delay chain, the difference between the energy consumption slope of the current node and the energy consumption slope of the previous node is written into the corresponding node, and the energy consumption difference is continuously accumulated according to the response time difference interval. When the air volume response time difference crosses the adjacent interval, the cumulative result of the previous time difference interval is retained as the initial value of the next time difference interval. At the same time, the pollution reduction difference is written into the outer ring node at the same position as the cross-ring verification feature to obtain the cumulative energy consumption difference under different response time differences. For each difference reversal position, search for inner and outer nodes with the same equipment identifier, control quantity type and action direction in the inter-ring switching layer. Align the two types of nodes according to the air volume response time difference. Define the pollution reduction difference, resistance hysteresis difference and purification weight in the inner node as purification characteristic channels. Define the cumulative energy consumption difference, energy consumption slope and air volume response time difference in the outer node as energy consumption characteristic channels. Exchange the node affiliation of purification characteristic channels and energy consumption characteristic channels at the reversal position. Make the inner node after the reversal position inherit the energy consumption change state of the corresponding outer node. Make the outer node after the reversal position inherit the purification change state of the corresponding inner node. The difference reversal position is the node position in the inner ring differential chain where the pollution reduction difference changes from positive to negative, from negative to positive, from continuous increase to continuous decrease, or from continuous decrease to continuous increase, and the node position in the outer ring time delay chain where the direction of change of the cumulative energy consumption difference changes accordingly. Differential calculation is performed again on adjacent nodes that have completed the characteristic channel exchange to obtain the purification difference and energy consumption difference after the exchange. When the purification difference after the exchange is in the same direction as the purification difference before the exchange, the current exchange relationship is retained. When the two directions are opposite, the exchange position is moved one node along the direction of control quantity change and recalculated until the purification difference of two consecutive nodes is in the same direction. The strategy output layer collects the control quantity change range, purification difference, cumulative energy consumption difference, resistance hysteresis difference, air volume response time difference and switching position of each node according to the control action index. It is arranged according to the increase path and decrease path of the same control quantity, and outputs the purification difference, energy consumption difference, difference reversal position and characteristic channel exchange result corresponding to each control quantity.

[0024] In this embodiment, the inter-ring switching layer specifically includes: Read the pollution reduction difference, resistance hysteresis difference and purification weight in the inner loop differential chain respectively, and read the cumulative energy consumption difference, energy consumption slope and air volume response time difference in the outer loop time delay chain. Establish the node correspondence relationship according to the equipment identification, control quantity type, action direction and control action index. Along the inner ring differential chain, nodes whose pollution reduction difference changes from positive to negative, from negative to positive, continuously increasing to continuously decreasing, and continuously decreasing to continuously increasing are searched. At the same time, along the outer ring time delay chain, nodes whose cumulative energy consumption difference reverses in the same direction are searched. The two types of reversed nodes are matched adjacently according to the wind volume response time difference from small to large. The pollution reduction difference, resistance hysteresis difference and purification weight corresponding to the matched internal node are recorded as purification characteristic channels, and the cumulative energy consumption difference, energy consumption slope and air volume response time difference corresponding to the external node are recorded as energy consumption characteristic channels. A switching relationship is established between the two types of channels that meet the same control action index. The switching relationship is to exchange the propagation paths of the purification feature channel and the energy consumption feature channel after the corresponding node at the reversal node position, so that the subsequent nodes of the inner ring inherit the energy consumption change state and the subsequent nodes of the outer ring inherit the purification change state. After the switch is completed, the purification difference and energy consumption difference of consecutive nodes before and after the switch position are recalculated. When the direction of the purification difference after the switch is consistent with that before the switch and the increase of the energy consumption difference decreases, the current switch position is retained. When the direction of the purification difference is reversed, one node is moved along the direction of the control quantity change and the switch is repeated until the direction of the purification difference of two consecutive nodes is consistent. The final switch node is recorded as the channel exchange position.

[0025] In this embodiment, S4 specifically includes: Read the purification difference, energy consumption difference, difference reversal position and characteristic channel exchange results corresponding to each control quantity, extract the air purifier fan speed, air circulation fan volume, gas compressor load, energy-saving air conditioner air volume, household ventilation appliance ventilation volume and start-stop duration of each device according to the device identifier, and use the current actual control value as the reference value of each control quantity. The disturbance step size is determined based on the amplitude interval between adjacent control actions of the same control quantity. The reference value is increased by one disturbance step size and recorded as a positive disturbance. The reference value is decreased by one disturbance step size and recorded as a negative disturbance. One disturbance step size is reserved on each side of the difference reversal position as the first layer search boundary. The control quantity is expanded layer by layer along the direction of continuous increase of the purification difference until the current control quantity reaches the difference reversal position or the purification difference between two adjacent layers continuously decreases. Record the equipment identifier, control type, disturbance direction, disturbance amplitude, disturbance level, corresponding purification difference and corresponding energy consumption difference for each single control quantity disturbance as a disturbance node; The disturbance node is a candidate control node that characterizes the purification response and energy consumption response of a control variable under a certain disturbance amplitude; Two adjacent disturbance nodes with the same control quantity are grouped together. The purification marginal response is obtained by subtracting the purification difference of the previous disturbance node from the purification difference of the subsequent disturbance node. The energy consumption marginal response is obtained by subtracting the energy consumption difference of the previous disturbance node from the energy consumption difference of the subsequent disturbance node. The purification marginal response and the energy consumption marginal response are combined according to the corresponding disturbance direction to form the marginal response difference. For each control variable, the marginal response difference is continuously calculated along the positive and negative disturbance directions. The current direction expansion is stopped when the purification marginal response changes from increasing to decreasing. The current disturbance node and the nodes in the same direction after the current disturbance node are deleted when the energy consumption marginal response increases for two consecutive levels and the purification marginal response does not increase synchronously. The disturbance nodes where the purification marginal response increases continuously and the energy consumption marginal response increase decreases continuously are retained. Based on the pairing relationship between control quantity and purification response time offset, the overlapping interval of the response window corresponding to the disturbance node of different equipment is retrieved. For any two equipment, the pollution reduction increment, air volume change, filter resistance hysteresis and energy consumption slope when disturbed alone are compared with the four types of response quantities when the two equipment undergo control changes at the same time. When the response window overlaps in time and the change direction of at least two types of response quantities remains continuous and consistent, a cross-equipment connection is established between the two disturbance nodes, and the difference between the cumulative result of the individual disturbance response quantity and the measured response quantity in the overlapping section is written into the cross-equipment connection. Disturbance nodes with cross-device connections are arranged according to device identifier, response start time and response duration. Connections with the same response direction and mutually reinforcing purification difference are marked as same-direction connections, and connections with opposite response directions or mutually weakening purification difference are marked as opposite-direction connections. Based on the cross-device response coupling relationship, the disturbance nodes are reorganized. The disturbance nodes at both ends of the same-direction connection are merged into a joint disturbance branch from front to back according to the response start time. The disturbance nodes at both ends of the reverse connection are rearranged in three ways: moving one disturbance level forward, moving one disturbance level backward, and deleting one of the disturbance nodes. The purification difference and energy consumption difference corresponding to the rearrangement are read respectively. Cross-device response coupling relationship is the response correlation between control actions of different air purification devices, which is jointly defined by cross-device connection, connection direction, response overlap length, and response difference of overlap segment. Cross-device response coupling relationship is the response correlation between control actions of different air purification devices, which is jointly defined by cross-device connection, connection direction, response overlap length, and response difference of overlap segment. Taking each first-layer disturbance node as the starting node, the positive and negative disturbance nodes of the next control quantity are sequentially connected to the current joint disturbance branch. After each disturbance node is connected, the purification marginal response of each node in the current branch is re-accumulated, and the repeated purification contribution is deducted according to the response difference of the overlapping section in the cross-device connection. At the same time, the energy consumption marginal response of each node is accumulated and the repeated energy consumption is corrected according to the power overlap corresponding to the reverse connection. The revised current layer purification marginal response is compared with the previous layer purification marginal response, and the increase in the current layer energy consumption marginal response is compared with the increase in the previous layer energy consumption marginal response, including: When the marginal response of the current layer purification increases and the increase in the marginal response of energy consumption decreases, the current joint perturbation branch is retained; When the cleanup marginal response decreases, delete the current joint perturbation branch; When the purification marginal response continues to increase but the energy consumption marginal response increases for two consecutive levels, the system reverts to the previous level of disturbance node and switches the disturbance direction of the control quantity. The node access, response correction, branch comparison and direction back-off are repeatedly executed in the order of control quantities such as air purifier fan speed, air circulation fan air volume, gas compressor load, energy-saving air conditioner air volume, household ventilation appliance ventilation volume and equipment start-up and shutdown time, until each control quantity has completed one positive and negative disturbance search. The joint disturbance branches that have completed the search of all control quantities are arranged from largest to smallest according to the cumulative clean marginal response. When the cumulative clean marginal responses are the same, they are arranged from smallest to largest according to the cumulative energy consumption marginal response. The control quantity values, disturbance directions, disturbance levels, execution order, and corresponding cross-device connections of each device are retained to obtain the control combination.

[0026] In this embodiment, S5 specifically includes: The cloud-based collaborative system receives control combinations and establishes an action position sequence according to the execution order of air purifier fan speed, air circulation fan volume, gas compressor load, energy-saving air conditioner air volume, household ventilation appliance ventilation volume, and equipment start-up and shutdown duration in each control combination. The control operation corresponding to a device control quantity at a specific execution position is recorded as a control action. Based on the air volume response time difference and cross-device response coupling relationship, the response duration interval corresponding to the control action is mapped to the continuous control cycle. Taking the current control combination as the reference combination, three types of position replacements are performed for any control action: moving forward one action position, moving backward one action position, and keeping the original action position. The control quantity values ​​and disturbance directions of the other control actions are fixed to obtain the counterfactual action sequence of the current control cycle. Multi-cycle counterfactual rolling involves using the predicted response state generated by each counterfactual action sequence in the current control cycle as the starting state of the next control cycle, and continuously unfolding the action response process for multiple control cycles according to the same action position replacement rule. For each counterfactual action sequence, the corresponding purification difference, energy consumption difference, air volume response time difference and cross-device response coupling relationship are retrieved according to the device identifier of the control action. Starting from the action execution position, the control response is transmitted cycle by cycle along the response time difference direction. The pollution reduction increment of the current cycle is deducted from the pollution reduction residual of the previous cycle, and the repeated purification contribution is deducted according to the response difference of the overlapping section of the cross-device connection in the same cycle, so as to obtain the pollution reduction amount corresponding to each action position. The pollution reduction is the net decrease in pollutant concentration within the corresponding response duration interval after the control action is completed and the location is replaced, after the response time difference is passed, the cross-equipment repeated contribution is deducted, and the residual amount from the previous period is taken over. The energy consumption difference is accumulated cycle by cycle along the position of the action. The power increment caused by the forward movement of the control action, the power release caused by the backward movement of the control action, and the power overlap when the control actions of adjacent devices overlap are recorded respectively. The power increment that continues to the next control cycle after the current cycle ends is passed on to the next cycle. In subsequent cycles, the power drop after the corresponding device exits the high load state is deducted one by one to obtain the power rebound amount corresponding to each action position. Power rebound is the relative baseline power change caused by the combined effects of power advance superposition, delayed release, and cross-cycle load continuation after the control action position changes. For the same control action, the pollution drop amounts of the three counterfactual action sequences—moving forward, moving backward, and maintaining the original position—are accumulated sequentially according to the control cycle, and the corresponding power rebound amounts are accumulated according to the positions within the same cycle. The pollution drop increments and power rebound increments of the three action positions in consecutive control cycles are compared, including: When the pollution drop increment after the action is moved forward continuously increases and the power rebound increment decreases, the control action is moved forward by one action position. When the pollution drop increment remains after the action is moved back and the power rebound increment decreases, the control action is moved back one action position. When both forward and backward movement cause a decrease in the pollution drop, the original action position is retained. For two control actions that have a cross-device response coupling relationship, after completing the single action position replacement, continue to execute the paired position exchange, move the previous control action to the position of the next control action, and move the next control action to the position of the previous control action synchronously. Re-expand the corresponding response interval according to the air volume response time difference, and calculate the pollution drop and power rebound before and after the position exchange respectively. When the amount of pollution drop increases after the position swap, the operating position after the swap is retained; when the amount of pollution drop remains the same and the power rebound decreases, the operating position after the swap is retained; when the amount of pollution drop decreases, the operating position before the swap is restored. According to the number of actions in the control combination, the single action forward movement, single action backward movement, position holding and paired exchange of coupled actions are completed sequentially, and the corresponding changes in pollution drop amount, power rebound amount, action movement direction and number of consecutive holding cycles for each position replacement are written into the action position record. The action position weight is calculated based on the action position record. The action position weight is the action ranking value determined by the pollution drop increment, power rebound increment, number of consecutive control cycles, and cross-equipment position exchange results of the control action relative to the reference position. The first ranking is established according to the increase in pollution drop from large to small, and the second ranking is established according to the increase in power rebound from small to large. The position weight of the control action is increased for the control action that is in the same position in the first and second rankings. The position weight of the control action that is in the first ranking and in the second ranking is maintained. The position weight of the control action that is in the second ranking and in the first ranking is decreased. The position weight of the control action that is in the second ranking and in the first ranking is reduced. The direction weight is added to the control action that maintains the same movement direction for two consecutive control cycles. Cloud-based collaboration rearranges control actions according to their positional weight from highest to lowest, including: When the response duration intervals of two control actions do not overlap, the order of the actions is determined directly according to the position weights. When two control actions are connected across devices in the same direction, the control action with the larger purification marginal response is placed first. When two control actions have a reverse cross-device connection, the control action with the smaller power rebound is placed first, and a control action execution queue is formed according to the rearrangement result. The control action execution queue, the control variable values ​​corresponding to each control action, the action position weight, the execution start position and the execution duration are sent to the edge computing node. The edge computing node reads the control action execution queue item by item according to the local unified timeline. After the previous control action reaches the corresponding air volume response start point, the next control action is triggered. Adjacent control actions connected across devices in the same direction maintain the cloud-based arrangement order, while adjacent control actions connected across devices in opposite directions are executed separately according to the corresponding air volume response time difference. Control commands are sent to the air purifier, air circulation fan, energy-saving gas compressor, energy-saving air conditioner, and household ventilation appliances respectively to determine the actual execution order of the air purification equipment.

[0027] In this embodiment, S6 specifically includes: After each control action is executed, the edge computing node collects pollutant concentration, air intake volume, air output volume, filter pressure difference and operating power according to the action position sequence, and recalculates the pollutant drop, power rebound and air volume response time difference starting from the actual trigger time of the control action. The measured air volume response time difference is compared with the original air volume response time difference of the corresponding control action item by item. The time position of the corresponding purification response is moved according to the time difference offset direction. The initial time difference and duration overlap between the control action and the pollution reduction increment, filter resistance hysteresis and energy consumption slope are recalculated. Pairing relationships with increased overlap after the time position is moved are retained. Pairing relationships with decreased overlap are moved back in the opposite direction by sampling point until the pairing relationship between the control quantity and purification response in the current control cycle is obtained. The measured pollution drop corresponding to the current pairing relationship is compared with the purification difference output by the dual-ring differential strategy network in terms of direction, and the measured power rebound is compared with the corresponding energy consumption difference in terms of direction. The difference reversal position is re-determined at the node where the purification change direction or the energy consumption change direction reverses. Centered on the newly determined difference reversal position, the current channel exchange position of the inter-ring switching layer is moved one node in the direction of increasing control quantity and decreasing control quantity, respectively. The purification difference and energy consumption difference are recalculated for the adjacent nodes after the movement. The position where the purification difference is continuous in the same direction and the energy consumption difference is not continuously increasing is retained as the updated channel exchange position. The updated channel exchange positions are the characteristic channel exchange nodes of the inner loop differential chain and the outer loop time delay chain, which are re-determined based on the measured purification response and energy consumption response. Based on the position of the updated channel exchange, the purification weight, cumulative energy consumption difference and the characteristic channel after the exchange of the corresponding node are updated synchronously, and the update results are written back to the dual-ring differential strategy network. The measured pollution drop at the actual execution position of each control action is subtracted from the pollution drop predicted by the counterfactual action sequence. The measured power rebound is subtracted from the predicted power rebound. The action position weights are then adjusted based on the continuous holding period of the control action and the results of cross-equipment position exchanges. The weighting is increased for the action position where the pollution drop deviation narrows and the power rebound deviation decreases. For action positions where pollution drop deviation increases or power rebound deviation increases continuously, reduce the weight; The control actions are rearranged according to the corrected action position weights, and the rearrangement results, the updated control quantity and purification response pairing relationship, and the updated channel exchange position are written into the next control cycle to complete the closed-loop update of the dual-loop differential strategy network and the execution order of the air purification equipment.

[0028] An AI-based air purification equipment operation optimization and control system includes: The data acquisition module is used to collect the operating parameters of the air purification equipment through various sensors and to preprocess the operating parameters; The response pairing module is used to calculate the pollution reduction increment, filter element resistance hysteresis, air volume response time difference and energy consumption slope based on the pre-processed operating parameters, and to pair the changes of each control quantity with the corresponding purification response according to the time offset. The differential strategy module is used to input the paired control quantity and purification response into the dual-loop differential strategy network. The inner loop calculates the purification difference corresponding to the increase or decrease of the control quantity, the outer loop accumulates the energy consumption difference corresponding to different response time differences, and the feature channels are exchanged at the difference reversal position. The disturbance calculation module is used to perform layer-by-layer positive and negative disturbances on the air volume, speed, load and start-stop duration of each air purification device based on the purification difference and energy consumption difference. It calculates the marginal response difference of adjacent disturbance nodes, reorganizes the disturbance nodes according to the cross-device response coupling relationship, and recursively retains the control combination with increasing purification benefits and decreasing energy consumption increase. The collaborative control module is used to perform multi-cycle counterfactual rolling on the retained control combination through cloud collaboration. It writes the pollution drop and power rebound after the control action shift into the action position weight, and determines the execution order of the air purification equipment based on the action position weight through edge computing. The feedback correction module is used to correct the pairing relationship between the control quantity and the purification response based on the pollution drop, power rebound, and airflow response time difference after the air purification equipment is executed, and to adjust the channel exchange position and action position weight of the dual-loop differential strategy network.

[0029] Example 1: To verify the feasibility of this invention in practice, it was applied to a complex indoor air treatment scenario. This scenario utilizes an air purifier, an air circulation fan, an energy-efficient gas compressor, an energy-efficient air conditioner, and household ventilation appliances to jointly perform particulate matter reduction, air circulation, air supply regulation, and indoor ventilation. The test area, under normal operating conditions, exhibits continuous human activity and intermittent pollutant release. The air purification equipment is not always in a stable operating state, and there are significant response delays in the control actions of different devices. Continuous data collection prior to implementation revealed that when particulate matter concentration increased from approximately 82 μg / m³ to 126 μg / m³, traditional control methods typically involved directly increasing the air purifier fan speed and circulating fan airflow, while simultaneously increasing the energy-saving air conditioner's airflow. This resulted in a short-term increase in total operating power from 1.86 kW to 2.74 kW, but the rate of decrease in pollutant concentration did not keep pace. When the air purifier fan speed was further increased from 1680 r / min to 1920 r / min, the decrease in particulate matter concentration per unit time only increased from 6.9 μg / m³ to 7.4 μg / m³, while the total power increased by approximately 0.29 kW, demonstrating a decrease in the rate of increase in purification benefits while energy consumption continued to rise. Furthermore, the effective airflow response after the air circulating fan activated typically lagged by approximately 12–20 seconds, and the effective response after the energy-saving air conditioner's airflow adjustment lagged by approximately 18–31 seconds. Evaluating the purification effect solely based on the moment the control command was issued could easily lead to incorrect attribution of purification responses from previous equipment to subsequent equipment, thus affecting the operational optimization results.

[0030] In this scenario, particulate matter concentration sensors, gas concentration sensors, airflow sensors, filter differential pressure sensors, temperature sensors, and power detection channels are deployed on the air inlet and outlet sides and main airflow locations of each air purification device. Simultaneously, data such as air purifier fan speed, airflow of the air circulator fan, gas compressor load, airflow of the energy-saving air conditioner, ventilation volume of household ventilation appliances, and the start / stop status of each device are collected. The sampling period is set to 2 seconds, with 1800 sampling points collected in a single continuous run. The collected data is corrected for timestamps, outlier removal, missing value compensation, and multi-channel alignment according to a unified timeline. A baseline window before control action and a response window after action are defined at locations where there are significant changes in pollutant concentration, filter differential pressure, airflow, and power. During a control process where the air purifier fan speed was increased from 1440 r / min to 1680 r / min, the particulate matter concentration before the action was 93.6 μg / m³, and decreased to 61.8 μg / m³ within 60 seconds after the action. After combining the changes in outlet airflow and the ratio of inlet airflow, the pollution reduction increment was 27.4 μg / m³. The cumulative difference between the lift and fall pressure differences of the filter element at the same inlet airflow position was 18.7 Pa, yielding the corresponding filter element resistance hysteresis. The inlet and outlet airflow continued to change in the same direction starting 10 seconds after the action was initiated, reaching 80% of the final stable airflow change amplitude at 18 seconds. Therefore, the airflow response time difference for this control action was determined to be 18 seconds. The median power change rate before the action was 0.006 kW / s, and the weighted power change rate after the action was 0.014 kW / s. Based on this, the corresponding energy consumption slope change result was obtained, and the above response quantities were paired with the fan speed change action according to an 18-second time offset.

[0031] After processing the continuously collected control actions in the same way, the amplitude of the control quantity change, the direction of the action, the duration of the action, the pollution reduction increment, the filter element resistance hysteresis, the air volume response time difference, and the energy consumption slope are encoded into action response units. Taking the fan speed of an air purifier as an example, the action response units corresponding to 1320 r / min, 1440 r / min, 1560 r / min, 1680 r / min, 1800 r / min and 1920 r / min are arranged according to the change range. The change in pollution reduction increment under adjacent control ranges is calculated through the inner loop differential chain of the dual-loop differential strategy network. When the fan speed increases from 1560 r / min to 1680 r / min, the purification difference is 5.8 μg / m³, while when it increases from 1800 r / min to 1920 r / min, the purification difference decreases to 1.1 μg / m³. At the same time, the outer loop time delay chain arranges each action response unit according to the air volume response time difference and accumulates the energy consumption difference. In the section where the response time difference is extended from 16s to 22s, the accumulated energy consumption difference increases from 0.09kW to 0.24kW. After the dual-ring differential strategy network detects that the purification difference is continuously decreasing while the cumulative energy consumption difference is continuously increasing, it determines the corresponding node as the difference reversal position and exchanges the corresponding purification characteristic channel and energy consumption characteristic channel in the inter-ring switching layer. After recalculation, the optimal search range of the air purifier fan speed is narrowed to 1560~1800r / min to avoid further ineffective expansion to the high speed range.

[0032] During the multi-device joint operation phase, based on the current actual control values, layer-by-layer positive and negative disturbances are applied to the air purifier fan speed, airflow of the air circulator fan, gas compressor load, airflow of the energy-saving air conditioner, ventilation volume of household ventilation appliances, and start / stop duration. Taking the air circulator fan as an example, after increasing the airflow by 40 m³ / h, 80 m³ / h, and 120 m³ / h, the corresponding purification marginal response increases by 4.6%, 3.8%, and 1.7%, respectively, while the energy consumption marginal response increases by 2.3%, 4.9%, and 8.2%, respectively. Therefore, further increasing the airflow is stopped at the third disturbance level. Further overlap analysis of the response windows of the air purifier and the energy-saving air conditioner reveals that when both increase their control values ​​simultaneously, the measured pollution reduction is approximately 7.6% lower than the direct sum of the responses of the two devices operating independently, indicating that the two devices contribute to purification redundantly during this period. Based on this, the present invention establishes a cross-device response coupling relationship, deducts redundant contributions in the joint disturbance branch, and shifts, postpones, and rearranges mutually weakening control actions. After recursive screening, the control combination that was retained was changed from the original simultaneous upgrade of multiple devices to "the air circulation fan first increases the air volume, the air purifier delays the start of the high air volume segment, the energy-saving air conditioner maintains medium air supply, and the household ventilation appliances enter the low ventilation segment after the pollution decreases", which significantly reduces the instantaneous power peak under the same purification target.

[0033] Cloud-based collaboration further implements multi-cycle counterfactual rolling on the retained control combinations. In a typical control combination, the high-speed operation of the air purifier is compared by maintaining its original position, moving it forward one position, and moving it backward one position. The pollution reduction amounts obtained for three consecutive control cycles are 29.8 μg / m³, 32.6 μg / m³, and 28.9 μg / m³, respectively, with corresponding power rebounds of 0.23 kW, 0.17 kW, and 0.19 kW. Therefore, the weight of the forward operation is increased. For cases where the air circulator and energy-saving air conditioner are connected across devices in the same direction, the air circulator, which has a larger purification marginal response, is placed in the first position. For cases where the air purifier and household ventilation appliances have overlapping reverse responses, staggered execution is implemented according to the time difference of the airflow response. Edge computing nodes trigger device control item by item according to the rearranged action position weights, so that the control actions are no longer simply executed simultaneously, but rather coordinated operation is completed based on the actual sequential relationship formed by the purification response.

[0034] After verification under multiple pollution levels and equipment load conditions, the average particulate matter concentration of the air purification equipment during operation decreased from an initial 96.8 μg / m³ to 35.7 μg / m³. The average time required to reach a stable air quality range was shortened from 13.6 min in the traditional control method to 9.4 min, a reduction of approximately 30.9%. The average operating power decreased from 2.31 kW to 1.91 kW, a reduction of approximately 17.3%. The average duration of high load overlap during a single control process decreased from 96 s to 41 s, a reduction of approximately 57.3%. The number of times the pollutant concentration rebounded significantly after entering the target range decreased from an average of 2.8 times per 10 control processes to 0.9 times. Further statistical analysis of the pairing results between control quantities and purification responses showed that the response matching accuracy was approximately 78.6% when using direct time correspondence of control commands. After introducing airflow response time difference and feedback correction, this accuracy increased to 94.1%, indicating that the present invention can effectively reduce erroneous attribution caused by response lag.

[0035] Table 1 Comparison of Air Purification Optimization Control Effects under Different Operating Conditions As shown in Table 1, under different initial pollution levels, the present invention maintained a short pollution stabilization time and low equipment operating power. When the initial particulate matter concentration was around 95 μg / m³, the present invention shortened the stabilization time from 13.1 min to 9.1 min, while reducing the average power from 2.26 kW to 1.88 kW. Under high pollution conditions with an initial particulate matter concentration exceeding 120 μg / m³, the stabilization time was still shortened from 17.6 min to 13.1 min, and the average power was reduced from 2.68 kW to 2.35 kW. Based on multiple sets of operating data, the present invention improved the purification speed while reducing the average operating power and peak power. Furthermore, through time offset pairing of control quantity and purification response, dual-loop differential strategy network, cross-device response coupling and reorganization, multi-cycle counterfactual rolling, and execution result feedback correction, the control actions between different air purification devices were made more coordinated, thus verifying the feasibility and energy-saving effect of the present invention in the optimized control of air purification equipment operation.

[0036] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based method for optimizing and controlling the operation of air purification equipment, characterized in that, Includes the following steps: S1. Collect the operating parameters of the air purification equipment through various sensors and perform preprocessing; S2. Calculate the pollution reduction increment, filter element resistance hysteresis, air volume response time difference and energy consumption slope based on the pre-treated operating parameters, and pair the changes of each control quantity with the corresponding purification response according to the time offset. S3. Input the paired control quantity and purification response into the dual-loop differential strategy network. The inner loop calculates the purification difference corresponding to the increase or decrease of the control quantity, and the outer loop accumulates the energy consumption difference under different response time differences and swaps the feature channels at the difference inversion position. S4. Based on the purification difference and energy consumption difference, perform layer-by-layer positive and negative disturbances on the air volume, speed, load and start-stop duration of each air purification device, calculate the marginal response difference of adjacent disturbance nodes, and reorganize the disturbance nodes according to the cross-device response coupling relationship, and recursively retain the control combination with increasing purification benefits and decreasing energy consumption increase. S5, cloud collaboration performs multi-cycle counterfactual rolling on the retained control combination, writes the pollution drop amount and power rebound amount after the action shift into the action position weight, and edge computing determines the execution order of air purification equipment according to the position weight; S6. Based on the pollution drop, power rebound, and airflow response time difference after execution, correct the pairing relationship between the control quantity and the purification response, and adjust the channel exchange position and action position weights of the dual-loop differential strategy network.

2. The air purification equipment operation optimization control method based on artificial intelligence according to claim 1, characterized in that, The air purification equipment includes an air purifier, an air circulation fan, an energy-saving gas compressor, an energy-saving air conditioner, and household ventilation appliances; the operating parameters include pollutant concentration, air intake volume, air output volume, fan speed, filter pressure difference, gas compressor load, supply air temperature, return air temperature, ventilation volume, operating power, and start / stop status; the preprocessing includes outlier removal, missing value interpolation, timestamp correction, multi-channel sampling alignment, sliding window segmentation, and amplitude normalization.

3. The air purification equipment operation optimization control method based on artificial intelligence according to claim 1, characterized in that, S2 specifically includes: Based on the pre-processed operating parameters, control quantity channels are established using fan speed, gas compressor load, energy-saving air conditioner air volume, household ventilation appliance ventilation volume and start / stop status. The start and end points of control actions are determined based on continuous unidirectional changes and stable holding positions. A control action index is established by combining equipment identification, control quantity type and change direction. Divide the control action into a pre-action reference window and a post-action response window, centered on the control action start point, and calculate the pollution reduction increment, filter element resistance hysteresis, airflow response time difference, and energy consumption slope, including: The pollutant concentration decrease was calculated based on adjacent sampling points and accumulated by combining the changes in inlet and outlet air volume to obtain the pollution reduction increment. Pair the lift and fall filter elements with the same air intake position, and accumulate the difference to obtain the filter element resistance hysteresis. The airflow response time difference is obtained based on the time interval between continuous unidirectional changes in inlet and outlet airflow until it reaches 80% of its final stable amplitude. By comparing the rate of power change before and after the action, the energy consumption slope can be obtained. Using the air volume response time difference as the offset benchmark, the control action index is shifted backward along a unified time axis. The four types of response change positions are retrieved within the corresponding search interval, and the control quantity and purification response are matched based on the start time difference, duration overlap, change direction consistency, and amplitude change synchronization. The overlapping control actions are divided into independent segments and overlapping segments. The response change ratio corresponding to a single control quantity is determined based on the independent segment. The contribution of the identified control quantity is deducted item by item from the total response of the overlapping segment, and the independent response quantity of the remaining control quantity is separated to complete the time offset pairing of each control action and the purification response.

4. The air purification equipment operation optimization control method based on artificial intelligence according to claim 1, characterized in that, S3 specifically includes: After completing the time offset pairing, the change amplitude, action direction, action duration, pollution reduction increment, filter resistance hysteresis, airflow response time difference, and energy consumption slope of the control quantity are encoded into action response units according to the control action index. A dual-loop differential strategy network is then constructed based on these action response units. This dual-loop differential strategy network consists of an inner loop differential chain, an outer loop time delay chain, an inter-loop switching layer, and a strategy output layer. The inner loop differential chain is a closed node chain formed by connecting the corresponding action response units of the same control quantity in the order of the action amplitude from decreasing to increasing; The outer loop time delay chain is a closed node chain formed by connecting action response units with the same direction of action in order of increasing air volume response time difference; The inter-ring switching layer is a connection layer that performs node feature exchange based on the sign reversal of the inner and outer ring differences. The strategy output layer is the output layer that performs joint sorting on the inner loop features and outer loop features after the switch and retains the differential results of the control actions; The inner loop differential chain groups the action response units according to the equipment identification and control quantity type, and arranges the nodes according to the control quantity change range. It performs differential on the pollution reduction increment and filter element resistance hysteresis of adjacent nodes, propagates the pollution reduction difference in the forward direction along the control quantity increase path, and propagates the pollution reduction difference in the reverse direction along the control quantity decrease path, and updates the purification weight recursively based on the resistance hysteresis difference. The outer loop time delay chain arranges the action response units in the same group according to the air volume response time difference, performs differential calculation on the energy consumption slope of adjacent nodes and continuously accumulates along the response time difference direction to obtain the cumulative energy consumption difference value corresponding to different response time differences; The inter-ring switching layer retrieves nodes where the pollution reduction difference and cumulative energy consumption difference reverse direction. It then aligns the pollution reduction difference, resistance hysteresis difference, and purification weight of the inner node with the cumulative energy consumption difference, energy consumption slope, and air volume response time difference of the outer node according to equipment identification, control quantity type, and action direction, and exchanges the corresponding characteristic channels at the difference reversal position. The differential is re-executed on the adjacent nodes after the exchange. When the direction of the purification difference is reversed, the switching position is moved along the direction of the control change and the calculation is recalculated until the direction of the purification difference of the consecutive nodes is consistent. The strategy output layer collects the purification difference, cumulative energy consumption difference, resistance hysteresis difference, air volume response time difference and switching position after the switch according to the control action index, and outputs the purification difference, energy consumption difference, difference reversal position and characteristic channel exchange result corresponding to each control quantity.

5. The air purification equipment operation optimization control method based on artificial intelligence according to claim 4, characterized in that, The inter-ring switching layer specifically includes: Read the pollution reduction difference, resistance hysteresis difference and purification weight in the inner loop differential chain respectively, and read the cumulative energy consumption difference, energy consumption slope and air volume response time difference in the outer loop time delay chain. Establish the node correspondence relationship according to the equipment identification, control quantity type, action direction and control action index. Along the inner ring differential chain, nodes whose pollution reduction difference changes from positive to negative, from negative to positive, continuously increasing to continuously decreasing, and continuously decreasing to continuously increasing are searched. At the same time, along the outer ring time delay chain, nodes whose cumulative energy consumption difference reverses in the same direction are searched. The two types of reversed nodes are matched adjacently according to the wind volume response time difference from small to large. The pollution reduction difference, resistance hysteresis difference and purification weight corresponding to the matched internal node are recorded as purification characteristic channels, and the cumulative energy consumption difference, energy consumption slope and air volume response time difference corresponding to the external node are recorded as energy consumption characteristic channels. A switching relationship is established between the two types of channels that meet the same control action index. The switching relationship is to exchange the propagation paths of the purification feature channel and the energy consumption feature channel after the corresponding node at the reversal node position, so that the subsequent nodes of the inner ring inherit the energy consumption change state and the subsequent nodes of the outer ring inherit the purification change state. After the switch is completed, the purification difference and energy consumption difference of consecutive nodes before and after the switch position are recalculated. When the direction of the purification difference after the switch is consistent with that before the switch and the increase of the energy consumption difference decreases, the current switch position is retained. When the direction of the purification difference is reversed, one node is moved along the direction of the control quantity change and the switch is repeated until the direction of the purification difference of two consecutive nodes is consistent. The final switch node is recorded as the channel exchange position.

6. The air purification equipment operation optimization control method based on artificial intelligence according to claim 1, characterized in that, S4 specifically includes: Read the purification difference, energy consumption difference, difference reversal position and characteristic channel exchange results corresponding to each control quantity, and determine the disturbance step size based on the current actual control value and the amplitude interval of adjacent control actions. Apply layer-by-layer positive and negative perturbations to the air purifier fan speed, air circulation fan volume, gas compressor load, energy-saving air conditioner air volume, household ventilation appliance ventilation volume, and equipment start-up and shutdown duration; Record the equipment identifier, control variable type, disturbance direction, disturbance amplitude, purification difference, and energy consumption difference corresponding to each disturbance as a disturbance node; For adjacent disturbance nodes with the same control quantity, calculate the purification marginal response and energy consumption marginal response respectively. Expand the disturbance nodes along the direction of continuous increase of purification marginal response, and delete the nodes where the purification marginal response decreases or the energy consumption marginal response increases continuously and the purification marginal response does not increase synchronously. Based on the pairing relationship between control quantity and purification response time offset, the response overlap interval of different equipment disturbance nodes is retrieved. Cross-equipment connections are established based on the overlapping changes of pollution reduction increment, air volume change, filter resistance hysteresis and energy consumption slope. Cross-equipment response coupling relationship is determined according to the response enhancement direction and response weakening direction. Based on the cross-device response coupling relationship, nodes connected in the same direction are merged according to the response start time, and nodes connected in the opposite direction are moved forward, backward, and deleted and rearranged to construct a joint perturbation branch. The positive and negative disturbance nodes of each control quantity are connected step by step along the joint disturbance branch. The repeated purification contribution caused by response overlap is deducted and the repeated energy consumption is corrected. The joint disturbance branch with increased purification marginal response and decreased energy consumption marginal response is recursively retained. Finally, the control combination is obtained by sorting according to the cumulative purification marginal response and the cumulative energy consumption marginal response.

7. The air purification equipment operation optimization control method based on artificial intelligence according to claim 1, characterized in that, S5 specifically includes: The cloud-based collaboration establishes an action position sequence based on the control combination, and maps each control action to a continuous control cycle by combining the air volume response time difference and cross-device response coupling relationship; Using the current control combination as a benchmark, the individual control actions are shifted forward, shifted backward, and held in place. After fixing the remaining control quantities, a counterfactual action sequence is formed, and the predicted response state of the current cycle is pushed to the next control cycle to complete multi-cycle counterfactual rolling. Multi-cycle counterfactual rolling involves using the predicted response state generated by each counterfactual action sequence in the current control cycle as the starting state of the next control cycle, and continuously unfolding the action response process for multiple control cycles according to the same action position replacement rule. The purification difference and energy consumption difference are transmitted cycle by cycle along each counterfactual action sequence. The pollution reduction residual and cross-equipment repeated purification contribution are deducted to obtain the pollution reduction amount. At the same time, the power rebound amount is obtained by accumulating the advance power increment, delayed power release amount and power overlap amount. Compare the pollution drop increment and power rebound increment corresponding to different action positions, and perform pairwise position swaps for control actions with cross-device response coupling relationship, retaining the position relationship of increased pollution drop or maintained pollution drop and decreased power rebound. The position weight of each control action is calculated based on the position replacement result, continuous holding period and pair exchange result, and the control actions are rearranged in combination with the cross-device connection direction. Edge computing triggers control actions one by one according to the rearranged action position weights and air volume response time differences, performs staggered control on adjacent actions connected across devices in the opposite direction, and drives each air purification device to run in a determined execution sequence.

8. An artificial intelligence-based air purification equipment operation optimization control system, executing the artificial intelligence-based air purification equipment operation optimization control method according to any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to collect the operating parameters of the air purification equipment through various sensors and to preprocess the operating parameters; The response pairing module is used to calculate the pollution reduction increment, filter element resistance hysteresis, air volume response time difference and energy consumption slope based on the pre-processed operating parameters, and to pair the changes of each control quantity with the corresponding purification response according to the time offset. The differential strategy module is used to input the paired control quantity and purification response into the dual-loop differential strategy network. The inner loop calculates the purification difference corresponding to the increase or decrease of the control quantity, the outer loop accumulates the energy consumption difference corresponding to different response time differences, and the feature channels are exchanged at the difference reversal position. The disturbance calculation module is used to perform layer-by-layer positive and negative disturbances on the air volume, speed, load and start-stop duration of each air purification device based on the purification difference and energy consumption difference. It calculates the marginal response difference of adjacent disturbance nodes, reorganizes the disturbance nodes according to the cross-device response coupling relationship, and recursively retains the control combination with increasing purification benefits and decreasing energy consumption increase. The collaborative control module is used to perform multi-cycle counterfactual rolling on the retained control combination through cloud collaboration. It writes the pollution drop and power rebound after the control action shift into the action position weight, and determines the execution order of the air purification equipment based on the action position weight through edge computing. The feedback correction module is used to correct the pairing relationship between the control quantity and the purification response based on the pollution drop, power rebound, and airflow response time difference after the air purification equipment is executed, and to adjust the channel exchange position and action position weight of the dual-loop differential strategy network.

Images