Medical purification system dynamic regulation and control method based on swarm intelligence

Abstract

The invention discloses a medical purification system dynamic regulation and control method based on swarm intelligence. The method comprises the following steps that environment state data and equipment state data of a medical purification area are collected and preprocessed; carrying out fusion processing, generating a region state vector, and forming a region state set; establishing a local observation vector, a neighborhood communication set and a control action set for each agent; decomposing the global control target set into a local control target set; performing swarm intelligent collaborative optimization iterative search on the control action set by adopting an improved SCSO algorithm to obtain a candidate control action set; performing consistency verification and constraint verification to obtain a cooperative control action set; forming control execution feedback data; the local control target set is updated, and collaborative optimization control of the medical purification system under multi-target constraints such as cleaning risk, energy consumption, airflow organization and differential pressure safety is achieved.

Description

A dynamic control method for medical purification systems based on swarm intelligence Technical Field

[0001] This invention relates to the field of medical environment purification and control, and in particular to a dynamic control method for medical purification systems based on swarm intelligence. Background Technology

[0002] Medical purification systems are crucial infrastructure for ensuring a clean environment and infection control in hospitals, widely used in operating rooms, intensive care units, isolation wards, laboratories, and pharmaceutical and sterilization supply areas. Existing medical purification systems typically consist of purification units, air supply units, return air units, fan units, valve units, and disinfection units. They control airborne particulate matter, pathogenic aerosols, volatile organic compounds, and other pollutants through filtration, ventilation, differential pressure control, and disinfection. As hospitals increasingly demand higher cleanliness levels, energy consumption control, and emergency response capabilities, the control methods for medical purification systems are gradually shifting from manual settings or fixed operating conditions towards automation and intelligence.

[0003] Currently, control methods for medical purification systems can be mainly categorized into two types: centralized control and zone control. Centralized control typically involves setting fixed thresholds or empirical rules in a central controller or host computer. Adjustments to purification power, fan speed, valve opening, or supply and return air volume are triggered by detecting environmental parameters at one or a few monitoring points. Zone control divides the medical purification area into several zones, each with its own local controller that independently adjusts relevant equipment to meet the corresponding cleanliness or pressure differential requirements. While these control methods can achieve basic purification functions, they still have significant limitations under complex and dynamic medical conditions. Summary of the Invention

[0004] One objective of this invention is to propose a dynamic control method for a medical purification system based on swarm intelligence. This invention fully utilizes multi-source environmental and equipment status perception, regional status fusion modeling, multi-agent collaborative control, and an improved SCSO swarm intelligence optimization algorithm to achieve distributed collaborative decision-making and closed-loop dynamic adjustment of purification units, air supply units, return air units, fan units, valve units, and disinfection units within the medical purification area. It has the advantages of fast response speed, strong global coordination, stable cleanliness maintenance, low energy consumption, and robust fault tolerance to equipment failures / communication anomalies.

[0005] A dynamic control method for a medical purification system based on swarm intelligence according to an embodiment of the present invention includes the following steps:

[0006] Collect and preprocess environmental and equipment status data from the medical purification area;

[0007] The preprocessed environmental status data and equipment status data are fused to generate regional status vectors, forming a regional status set.

[0008] Each purification unit, air supply unit, return air unit, fan unit, valve unit, and disinfection unit in the medical purification system is defined as an intelligent agent, and a local observation vector, neighborhood communication set, and control action set are established for each intelligent agent.

[0009] Based on the set of regional states and the local observation vectors of each agent, a set of global control objectives and a set of operational constraints are obtained, and the set of global control objectives is decomposed into a set of local control objectives.

[0010] Based on the local control objective set, the operational constraint set, and the neighborhood communication set, an improved SCSO algorithm is used to perform a swarm intelligent collaborative optimization iterative search on the control action set to obtain a candidate control action set.

[0011] Consistency and constraint checks are performed on the candidate control actions of the candidate control action set to obtain the collaborative control action set;

[0012] Control is executed on the corresponding units of each intelligent agent according to the set of collaborative control actions, and control execution feedback data is generated.

[0013] Failure determination is performed based on control execution feedback data, failure determination results are generated, the local control target set corresponding to the failed agent is reallocated, and the local control target set is updated.

[0014] Optionally, the environmental status data includes particulate matter concentration, pathogenic aerosol concentration, volatile organic compound concentration, carbon dioxide concentration, temperature, humidity, pressure difference, supply air volume, return air volume, and personnel activity density data. The equipment status data includes the operating status data of the purification unit, supply air unit, return air unit, fan unit, valve unit, and disinfection unit. The preprocessing includes time synchronization, anomaly removal, missing data completion, and noise filtering.

[0015] Optionally, the formation of the region state set specifically includes:

[0016] Based on the spatial location of environmental monitoring nodes and equipment monitoring nodes, the preprocessed environmental status data and equipment status data are matched according to the medical purification area to obtain the environmental status subset and equipment status subset corresponding to each medical purification area.

[0017] Align and stitch together the environmental state subset and equipment state subset of each medical purification area to generate a region fusion vector corresponding to the medical purification area.

[0018] The evidence reasoning multi-source fusion processing is performed on the regional fusion vector. The evidence reasoning multi-source fusion processing is to construct each data in the regional fusion vector into a corresponding evidence item and generate a basic probability allocation. The evidence items are then synthesized according to a predetermined evidence synthesis rule to output the regional state vector corresponding to the medical purification area.

[0019] The region state vectors of all medical purification areas are aggregated according to the medical purification area number to form a region state set.

[0020] Optionally, the establishment of the local observation vector, the neighborhood communication set, and the control action set specifically includes:

[0021] Each purification unit, air supply unit, return air unit, fan unit, valve unit, and disinfection unit within the medical purification system is uniquely identified, and each uniquely identified unit is defined as an intelligent agent, forming a set of intelligent agents.

[0022] Within each control cycle, the device status data of the corresponding unit and the environmental status data of the medical purification area where the agent is located are called for each agent in the agent set, and the local observation vector of the agent is generated by combining them in a predetermined order.

[0023] Based on the spatial positional relationship between each agent and the predetermined neighborhood communication criteria, for each agent in the agent set, agents that meet the neighborhood communication conditions are obtained, forming a neighborhood communication set of agents. The neighborhood communication conditions refer to the spatial distance between agents being less than the predetermined neighborhood communication distance threshold and the communication link being in a connected state.

[0024] For each agent in the agent set, a set of control actions for the agent is constructed based on the type of executable control quantity of the corresponding unit. The set of control actions includes control actions corresponding to purification power, fan speed, valve opening, supply air volume, return air volume, airflow path, and differential pressure mode.

[0025] Optionally, the generation of the local control target set specifically includes:

[0026] A global control target set is constructed based on the regional state vector and the local observation vector. The construction process involves extracting key state quantities such as cleanliness risk, energy consumption, airflow organization, and pressure difference from the regional state vector of the medical purification area and the local observation vector of each agent according to the pre-set target items, generating corresponding cleanliness risk targets, energy consumption targets, airflow organization targets, and pressure difference targets, thus forming a global control target set.

[0027] The set of operational constraints is constructed based on the regional state vector and the local observation vector. The construction process reads the regional state vector and the local observation vector in each control cycle, locates the state quantities corresponding to the control action, pressure difference, supply air volume, return air volume and equipment operating status, classifies the state quantities according to the preset constraint items and threshold ranges and assigns upper and lower limit requirements, and obtains the control action boundary constraints, regional pressure difference constraints, supply air volume constraints, return air volume constraints and equipment operating status constraints, forming the set of operational constraints.

[0028] Based on the medical purification area to which each intelligent agent belongs and the control action set of the corresponding unit of each intelligent agent, a one-to-one correspondence between the control action set and the global control target set is established.

[0029] The global control objective set is decomposed into a set of local control objectives corresponding to each agent according to a one-to-one correspondence.

[0030] Optionally, obtaining the candidate control action set specifically includes:

[0031] Within each control cycle, a search population for the improved SCSO algorithm is established for each agent in the agent set. A predetermined number of search individuals are randomly generated within the control action boundary constraints corresponding to the agent's control action set, forming the initial search individual set of the agent. The improved SCSO algorithm includes an adaptive search scheduling unit, a neighborhood cooperative interaction unit, and a constraint robust processing unit. The adaptive search scheduling unit refers to using the initial search individual set as the iteration object to obtain the step size adjustment factor. The neighborhood cooperative interaction unit refers to introducing a chaotic mapping cooperative perturbation mechanism to form a cooperative update step size and obtain the updated search individuals. The constraint robust processing unit refers to performing operation constraint set verification and control action boundary constraint verification on each control action corresponding to the search individual to obtain the corrected search individual and recalculate the fitness value.

[0032] In the adaptive search scheduling unit, the initial set of search individuals is used as the iteration object. Within each iteration step, the search intensity coefficient is calculated based on the correspondence between the current iteration step and the preset maximum iteration step, and the step size adjustment factor of each search individual is obtained based on the search intensity coefficient.

[0033] For each search individual of each agent, a fitness value is calculated based on the local control target set and the operational constraint set. The fitness value is obtained by weighting and summing the local target values ​​corresponding to the cleanliness risk target, the energy consumption target, the airflow organization target, and the pressure difference target according to preset weights.

[0034] In the neighborhood cooperative interaction unit, for each agent, the neighborhood optimal search individual with the best fitness value in the neighborhood communication set is obtained in each iteration step, and the global optimal search individual with the best fitness value is obtained among all search individuals of all agents. A chaotic mapping cooperative perturbation mechanism is introduced to generate a perturbation factor, and the step size adjustment factor is coupled with the perturbation factor to form a cooperative update step size, so as to obtain the updated search individual.

[0035] In the constraint robust processing unit, the updated search individual is subjected to feasibility constraint processing. The feasibility constraint processing refers to performing operation constraint set verification and control action boundary constraint verification on each control action corresponding to the search individual, correcting the control action components that do not meet the operation constraint set or control action boundary constraints according to the corresponding constraint threshold, and recalculating the fitness value of the corrected search individual.

[0036] The search population is iteratively updated. When the number of iterations reaches a preset maximum value, the control action corresponding to the search individual with the best fitness value is selected as the candidate control action of the agent. The candidate control actions of each agent are then aggregated to form a candidate control action set.

[0037] Optionally, obtaining the set of coordinated control actions specifically includes:

[0038] The control variables corresponding to the control action set in the candidate control action set are aligned and compared item by item to generate a consistency evaluation value of the agent's candidate control actions.

[0039] For each candidate control action of the agent, constraint verification is performed item by item according to the set of operational constraints and the control action boundary constraints corresponding to the control action set of the agent, and constraint verification results of the candidate control actions of the agent are generated.

[0040] For each agent, the candidate control action with the best consistency evaluation value is selected from the candidate control actions whose constraint verification results are passed, and the agent's cooperative control action is obtained.

[0041] All the collaborative control actions of the agents are aggregated into a collaborative control action set according to the agent number.

[0042] Optionally, the formation of the control execution feedback data specifically includes:

[0043] Based on the one-to-one correspondence between intelligent agents and corresponding units, the collaborative control actions in the collaborative control action set are assigned to the corresponding intelligent agents;

[0044] The collaborative control actions of each intelligent agent are unpacked to obtain the control quantity setting values ​​corresponding to each item of the intelligent agent's control action set. The control quantity setting values ​​include purification power setting value, fan speed setting value, valve opening setting value, supply air volume setting value, return air volume setting value, airflow path switching setting value, and differential pressure mode setting value.

[0045] The control setpoint is sent to the execution interface of the corresponding unit of the intelligent agent to perform purification power adjustment, fan speed adjustment, valve opening adjustment, supply air volume adjustment, return air volume adjustment, airflow path switching, and differential pressure mode switching.

[0046] After the corresponding unit completes its execution, the actual execution result data of the corresponding unit is collected. The actual execution result data includes actual purification power data, actual fan speed data, actual valve opening data, actual supply air volume data, actual return air volume data, actual airflow path status data, and actual differential pressure mode status data.

[0047] The actual execution result data of each intelligent agent's corresponding unit are combined to form control execution feedback data.

[0048] Optionally, the update of the local control target set specifically includes:

[0049] The control execution feedback data is collected according to the agent number to obtain the feedback data subset of each agent's corresponding unit;

[0050] The feedback data subset is compared with the corresponding collaborative control action of the intelligent agent to generate control deviation data of the intelligent agent. Based on the control deviation data and the device operation status indication information in the feedback data subset, a failure judgment is made for each intelligent agent and a failure judgment result is generated.

[0051] When the failure determination result indicates that the agent has failed, the agent is marked as a failed agent and its candidate control actions are frozen in the current control cycle.

[0052] Based on the neighborhood communication set of the failed agent, the agents within the neighborhood communication set reallocate the local control target set corresponding to the failed agent according to a predetermined reallocation rule, and update the local control target set.

[0053] The beneficial effects of this invention are:

[0054] This invention, by simultaneously collecting and preprocessing environmental and equipment status data within a medical purification area, further integrates and generates regional status vectors and sets. This enables the system to replace traditional coarse-grained monitoring methods that rely on single points or a limited number of indicators with multi-source, global, and continuous status representation, thus providing a consistent, comparable, and real-time updated global input foundation for subsequent control. Furthermore, the purification unit, air supply unit, return air unit, fan unit, valve unit, and disinfection unit are intelligently integrated, and local observation vectors and neighborhood communication sets are established. This empowers each execution unit with autonomous decision-making capabilities based on local information and neighborhood collaborative information, avoiding the response lag and single-point failure risks of existing centralized control systems under multi-regional dynamic disturbances, and improving the system's real-time performance and stability in complex medical scenarios.

[0055] This invention, within a unified framework of global control objectives and operational constraints, decomposes multiple objectives such as cleanliness risk, energy consumption, airflow organization, and differential pressure safety into local control objective sets corresponding to each agent. It then employs an improved SCSO algorithm, incorporating adaptive search scheduling, neighborhood collaborative interaction, and constraint robustness processing, to perform swarm intelligent collaborative optimization of the control action set. This enables the control actions to obtain the optimal candidate control actions online, oriented towards multiple objectives, while satisfying rigid constraints such as control action boundaries, differential pressure thresholds, supply and return air volume thresholds, and equipment operating status thresholds. This significantly reduces the excessive energy consumption problem caused by "long-term full-load operation to meet standards" in traditional rule-based control. Furthermore, it allows for rapid adaptive reallocation of purification resources under dynamic conditions such as personnel movement, pollution diffusion, and cleanliness level switching, stably maintaining the target cleanliness level and improving response efficiency to sudden pollution or pathogenic aerosol risks.

[0056] This invention generates a set of collaborative control actions by performing consistency and constraint checks on candidate control actions. After execution, it generates control execution feedback data for failure determination and local control target reallocation, enabling the system to possess closed-loop self-correction and collaborative fault tolerance capabilities. When local device failure or communication anomalies occur, the system can freeze the candidate control actions of the failed agent and allow neighboring agents to take over its local control target, thereby ensuring uninterrupted overall purification control and minimal degradation of global performance. This distributed collaborative and robust fault-tolerant mechanism effectively overcomes the dependence of existing technologies on a central controller or a single zone controller, enhancing the adaptability and safety of medical purification systems under high reliability and high continuous operation requirements. Attached Figure Description

[0057] 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:

[0058] Figure 1 is an overall flowchart of a dynamic control method for a medical purification system based on swarm intelligence proposed in this invention;

[0059] Figure 2 is a schematic diagram of the construction of the local observation vector, neighborhood communication set, and control action set of a dynamic control method for a medical purification system based on swarm intelligence proposed in this invention.

[0060] Figure 3 is a schematic diagram of the improved SCSO algorithm of a dynamic control method for a medical purification system based on swarm intelligence proposed in this invention. Detailed Implementation

[0061] 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.

[0062] Referring to Figures 1-3, a dynamic control method for a medical purification system based on swarm intelligence includes the following steps:

[0063] Collect and preprocess environmental and equipment status data from the medical purification area;

[0064] The preprocessed environmental status data and equipment status data are fused to generate regional status vectors, forming a regional status set.

[0065] Each purification unit, air supply unit, return air unit, fan unit, valve unit, and disinfection unit in the medical purification system is defined as an intelligent agent, and a local observation vector, neighborhood communication set, and control action set are established for each intelligent agent.

[0066] Based on the set of regional states and the local observation vectors of each agent, a set of global control objectives and a set of operational constraints are obtained, and the set of global control objectives is decomposed into a set of local control objectives.

[0067] Based on the local control objective set, the operational constraint set, and the neighborhood communication set, an improved SCSO algorithm is used to perform a swarm intelligent collaborative optimization iterative search on the control action set to obtain a candidate control action set.

[0068] Consistency and constraint checks are performed on the candidate control actions of the candidate control action set to obtain the collaborative control action set;

[0069] Control is executed on the corresponding units of each intelligent agent according to the set of collaborative control actions, and control execution feedback data is generated.

[0070] Failure determination is performed based on control execution feedback data, failure determination results are generated, the local control target set corresponding to the failed agent is reallocated, and the local control target set is updated.

[0071] In this embodiment, the environmental status data includes particulate matter concentration, pathogenic aerosol concentration, volatile organic compound concentration, carbon dioxide concentration, temperature, humidity, pressure difference, supply air volume, return air volume, and personnel activity density data. The equipment status data includes the operating status data of the purification unit, supply air unit, return air unit, fan unit, valve unit, and disinfection unit. The preprocessing includes time synchronization, anomaly removal, missing data completion, and noise filtering.

[0072] In this embodiment, the formation of the region state set specifically includes:

[0073] Based on the spatial location of environmental monitoring nodes and equipment monitoring nodes, the preprocessed environmental status data and equipment status data are matched according to the medical purification area to obtain the environmental status subset and equipment status subset corresponding to each medical purification area.

[0074] Align and stitch together the environmental state subset and equipment state subset of each medical purification area to generate a region fusion vector corresponding to the medical purification area.

[0075] The evidence reasoning multi-source fusion processing is performed on the regional fusion vector. The evidence reasoning multi-source fusion processing is to construct each data in the regional fusion vector into a corresponding evidence item and generate a basic probability allocation. The evidence items are then synthesized according to a predetermined evidence synthesis rule to output the regional state vector corresponding to the medical purification area.

[0076] The region state vectors of all medical purification areas are aggregated according to the medical purification area number to form a region state set.

[0077] In this embodiment, the establishment of the local observation vector, the neighborhood communication set, and the control action set specifically includes:

[0078] Each purification unit, air supply unit, return air unit, fan unit, valve unit, and disinfection unit within the medical purification system is uniquely identified, and each uniquely identified unit is defined as an intelligent agent, forming a set of intelligent agents.

[0079] Within each control cycle, the device status data of the corresponding unit and the environmental status data of the medical purification area where the agent is located are called for each agent in the agent set, and the local observation vector of the agent is generated by combining them in a predetermined order.

[0080] Based on the spatial positional relationship between each agent and the predetermined neighborhood communication criteria, for each agent in the agent set, agents that meet the neighborhood communication conditions are obtained, forming a neighborhood communication set of agents. The neighborhood communication conditions refer to the spatial distance between agents being less than the predetermined neighborhood communication distance threshold and the communication link being in a connected state.

[0081] For each agent in the agent set, a set of control actions for the agent is constructed based on the type of executable control quantity of the corresponding unit. The set of control actions includes control actions corresponding to purification power, fan speed, valve opening, supply air volume, return air volume, airflow path, and differential pressure mode.

[0082] In this embodiment, the generation of the local control target set specifically includes:

[0083] A global control target set is constructed based on the regional state vector and the local observation vector. The construction process involves extracting key state quantities such as cleanliness risk, energy consumption, airflow organization, and pressure difference from the regional state vector of the medical purification area and the local observation vector of each agent according to the pre-set target items, generating corresponding cleanliness risk targets, energy consumption targets, airflow organization targets, and pressure difference targets, thus forming a global control target set.

[0084] The set of operational constraints is constructed based on the regional state vector and the local observation vector. The construction process reads the regional state vector and the local observation vector in each control cycle, locates the state quantities corresponding to the control action, pressure difference, supply air volume, return air volume and equipment operating status, classifies the state quantities according to the preset constraint items and threshold ranges and assigns upper and lower limit requirements, and obtains the control action boundary constraints, regional pressure difference constraints, supply air volume constraints, return air volume constraints and equipment operating status constraints, forming the set of operational constraints.

[0085] Based on the medical purification area to which each intelligent agent belongs and the control action set of the corresponding unit of each intelligent agent, a one-to-one correspondence between the control action set and the global control target set is established.

[0086] The global control objective set is decomposed into a set of local control objectives corresponding to each agent according to a one-to-one correspondence.

[0087] In this embodiment, obtaining the candidate control action set specifically includes:

[0088] Within each control cycle, a search population for the improved SCSO algorithm is established for each agent in the agent set. A predetermined number of search individuals are randomly generated within the control action boundary constraints corresponding to the agent's control action set, forming the initial search individual set of the agent. The improved SCSO algorithm includes an adaptive search scheduling unit, a neighborhood cooperative interaction unit, and a constraint robust processing unit. The adaptive search scheduling unit refers to using the initial search individual set as the iteration object to obtain the step size adjustment factor. The neighborhood cooperative interaction unit refers to introducing a chaotic mapping cooperative perturbation mechanism to form a cooperative update step size and obtain the updated search individuals. The constraint robust processing unit refers to performing operation constraint set verification and control action boundary constraint verification on each control action corresponding to the search individual to obtain the corrected search individual and recalculate the fitness value.

[0089] In the adaptive search scheduling unit, the initial set of search individuals is used as the iteration object. Within each iteration step, the search intensity coefficient is calculated based on the correspondence between the current iteration step and the preset maximum iteration step, and the step size adjustment factor of each search individual is obtained based on the search intensity coefficient.

[0090] For each search individual of each agent, a fitness value is calculated based on the local control target set and the operational constraint set. The fitness value is obtained by weighting and summing the local target values ​​corresponding to the cleanliness risk target, the energy consumption target, the airflow organization target, and the pressure difference target according to preset weights.

[0091] In the neighborhood cooperative interaction unit, for each agent, the neighborhood optimal search individual with the best fitness value in the neighborhood communication set is obtained in each iteration step, and the global optimal search individual with the best fitness value is obtained among all search individuals of all agents. A chaotic mapping cooperative perturbation mechanism is introduced to generate a perturbation factor, and the step size adjustment factor is coupled with the perturbation factor to form a cooperative update step size, so as to obtain the updated search individual.

[0092] In the constraint robust processing unit, the updated search individual is subjected to feasibility constraint processing. The feasibility constraint processing refers to performing operation constraint set verification and control action boundary constraint verification on each control action corresponding to the search individual, correcting the control action components that do not meet the operation constraint set or control action boundary constraints according to the corresponding constraint threshold, and recalculating the fitness value of the corrected search individual.

[0093] The search population is iteratively updated. When the number of iterations reaches a preset maximum value, the control action corresponding to the search individual with the best fitness value is selected as the candidate control action of the agent. The candidate control actions of each agent are then aggregated to form a candidate control action set.

[0094] In this embodiment, obtaining the set of coordinated control actions specifically includes:

[0095] The control variables corresponding to the control action set in the candidate control action set are aligned and compared item by item to generate a consistency evaluation value of the agent's candidate control actions.

[0096] For each candidate control action of the agent, constraint verification is performed item by item according to the set of operational constraints and the control action boundary constraints corresponding to the control action set of the agent, and constraint verification results of the candidate control actions of the agent are generated.

[0097] For each agent, the candidate control action with the best consistency evaluation value is selected from the candidate control actions whose constraint verification results are passed, and the agent's cooperative control action is obtained.

[0098] All the collaborative control actions of the agents are aggregated into a collaborative control action set according to the agent number.

[0099] In this embodiment, the formation of the control execution feedback data specifically includes:

[0100] Based on the one-to-one correspondence between intelligent agents and corresponding units, the collaborative control actions in the collaborative control action set are assigned to the corresponding intelligent agents;

[0101] The collaborative control actions of each intelligent agent are unpacked to obtain the control quantity setting values ​​corresponding to each item of the intelligent agent's control action set. The control quantity setting values ​​include purification power setting value, fan speed setting value, valve opening setting value, supply air volume setting value, return air volume setting value, airflow path switching setting value, and differential pressure mode setting value.

[0102] The control setpoint is sent to the execution interface of the corresponding unit of the intelligent agent to perform purification power adjustment, fan speed adjustment, valve opening adjustment, supply air volume adjustment, return air volume adjustment, airflow path switching, and differential pressure mode switching.

[0103] After the corresponding unit completes its execution, the actual execution result data of the corresponding unit is collected. The actual execution result data includes actual purification power data, actual fan speed data, actual valve opening data, actual supply air volume data, actual return air volume data, actual airflow path status data, and actual differential pressure mode status data.

[0104] The actual execution result data of each intelligent agent's corresponding unit are combined to form control execution feedback data.

[0105] In this embodiment, the updating of the local control target set specifically includes:

[0106] The control execution feedback data is collected according to the agent number to obtain the feedback data subset of each agent's corresponding unit;

[0107] The feedback data subset is compared with the corresponding collaborative control action of the intelligent agent to generate control deviation data of the intelligent agent. Based on the control deviation data and the device operation status indication information in the feedback data subset, a failure judgment is made for each intelligent agent and a failure judgment result is generated.

[0108] When the failure determination result indicates that the agent has failed, the agent is marked as a failed agent and its candidate control actions are frozen in the current control cycle.

[0109] Based on the neighborhood communication set of the failed agent, the agents within the neighborhood communication set reallocate the local control target set corresponding to the failed agent according to a predetermined reallocation rule, and update the local control target set.

[0110] Example 1:

[0111] This embodiment uses a newly built intensive care unit (ICU) and adjacent isolation ward in a tertiary hospital as an application scenario for a medical purification area. This area consists of three functional zones: the ICU zone, the isolation ward zone, and the medical staff buffer corridor zone. Airflow between these three zones is organized via supply and return air ducts and interlocking valves, simultaneously meeting both surgical-grade cleanliness and isolation negative pressure requirements. The ICU zone requires maintaining high cleanliness and stable temperature and humidity, while the isolation ward zone requires continuous negative pressure to prevent the escape of pathogenic aerosols. The corridor zone, as a transition area, needs to ensure that airflow direction and cleanliness are not disturbed in the opposite direction during periods of high personnel density and bed movement. The area is equipped with two independent purification units, three supply air units, three return air units, six variable frequency fans, nine adjustable valves, and two ultraviolet disinfection units. Each unit can independently adjust purification power, fan speed, valve opening, supply and return air volume, airflow path, and pressure differential mode. Environmental monitoring nodes are deployed in different areas to collect data in real time, including particulate matter concentration, equivalent concentration of pathogenic aerosols, concentration of volatile organic compounds, carbon dioxide concentration, temperature and humidity, regional pressure difference, supply and return air volume, and personnel activity density. Equipment monitoring nodes collect equipment status data in real time, including the operating load of the purification unit, fan speed, valve opening, and disinfection module power.

[0112] In this scenario, the hospital's original control method was a typical zone threshold linkage control: when particulate matter or carbon dioxide levels exceeded the standard in a certain area, only the purification power and air volume of that area were increased. The negative pressure control of the isolation wards was maintained by an independent differential pressure controller, the air volume in the corridor was fixed, and the linkage between zones was only triggered by valve fine-tuning when the pressure difference deviated. This method was basically usable under normal low-disturbance conditions, but when strong disturbances occurred, such as concentrated entry and exit of personnel, bed changes, suctioning / nebulization, etc., the monitored pollution peaks showed a significant lag. Often, purification was increased only after the levels exceeded the standard, and the local increase in air supply caused the pressure difference between the corridor and the isolation ward to reverse briefly, with the negative pressure once approaching the lower limit. Medical staff reported fluctuations in air velocity and increased noise, and energy consumption remained at a high level for a long time.

[0113] After deployment in the region, the method of this invention operates according to the process described in the claims. The system first simultaneously collects environmental and equipment status data, and performs preprocessing such as time synchronization, anomaly removal, missing data completion, filtering, and smoothing. Then, the preprocessed multi-source data is fused according to region affiliation to generate regional state vectors for three functional areas, forming a regional state set. The purification unit, air supply unit, return air unit, fan unit, valve unit, and disinfection unit are defined as intelligent agents. Each intelligent agent generates a local observation vector within each control cycle and shares neighboring region states based on a neighborhood communication set. The system constructs a global control target set of cleanliness risk, energy consumption, airflow organization, and differential pressure safety based on the regional state set and local observation vectors. Simultaneously, it forms an operational constraint set consisting of control action boundaries, differential pressure thresholds, supply and return air volume thresholds, and equipment operating status thresholds, and decomposes the global target into the local control target sets of each intelligent agent. The improved SCSO algorithm uses the control action sets of each agent as the search object. An adaptive search scheduling unit maintains a high search intensity in the early stages of disturbance and gradually converges during the stable period. A neighborhood cooperative interaction unit combined with a chaotic mapping disturbance mechanism enables rapid cross-regional cooperative updates. A constraint robust processing unit ensures that candidate control actions always fall within the set of operating constraints. The candidate control action sets obtained by all agents undergo consistency and constraint checks. After screening, a cooperative control action set is formed and issued for execution. The execution results form control execution feedback data, used for closed-loop updates and failure determination in the next cycle.

[0114] To verify the beneficial effects, the hospital conducted a two-week comparative test under the same season and load level: the first week used the original threshold linkage control as the baseline, and the second week used the method of this invention. Both weeks included similar daily average patient flow, shift change frequency, and medical operation events, and the external fresh air conditions were consistent. Statistical results show that the invention exhibits significant advantages under strong disturbance scenarios. For example, during the morning and evening shift changes and peak bed-pushing periods, the peak value of the equivalent concentration of pathogenic aerosols in the isolation ward and corridors was significantly reduced, and the peak duration was shortened; the negative pressure was maintained more stably, and there was no situation where the pressure difference approached the lower limit; the purification power and air volume were adaptively allocated according to risk, and the load was automatically reduced during non-high-risk periods, resulting in a significant decrease in overall energy consumption. In the subjective feedback from medical staff, wind speed fluctuations and noise complaints decreased, indicating improved comfort. More importantly, in a simulated local fan failure in the isolation ward, the system identified the failure trend through control execution feedback data and triggered a neighboring intelligent agent to take over the local control target and reallocate purification resources. The isolation negative pressure was not interrupted, and the cleanliness index did not exceed the standard, demonstrating robust fault tolerance.

[0115] Table 1. Statistical Table of Operational Data for Medical Purification Areas in ICU-Isolation Wards

[0116] The following table shows the improvement rates of different indicators: Intensive Care Unit (ICU) 0.3µm Particulate Matter Average Concentration (particles / L): 3120 2540 ↓ 18.6%; Intensive Care Unit (ICU) 0.3µm Particulate Matter Peak Concentration (particles / L): 6480 4920 ↓ 24.1%; Isolation Ward Equivalent Average Concentration of Pathogen Aerosols (AU): 1.42 1.12 ↓ 21.1%; Isolation Ward Equivalent Peak Concentration of Pathogen Aerosols (AU): 3.10 2.05 ↓ 33.9%; Corridor Area CO2 Average Concentration (ppm): 820 760 ↓ 7.3%; Intensive Care Unit Temperature Fluctuation Range (°C): ±0.8 ±0.4 ↓ 50.0%; Isolation Ward Area Temperature Fluctuation Range (°C): ±0.8 ±0.4 ↓ 50.0% Pressure difference compliance rate of ward area (percentage of time negative pressure maintained within the threshold) % 93.2 99.1 ↑ 5.9 pct Lowest pressure difference Pa -7.1 -9.3 Safer Typical disturbance event response time (from risk escalation to control action taking effect) s 120 45 ↓ 62.5% Average daily total regional air supply m³ / h 15800 14200 ↓ 10.1% Average daily total regional energy consumption (purification + fan + valve + disinfection) kWh / d 1120 910 ↓ 18.8% Number of complaints about medical staff comfort (related to wind speed / noise) times / week 145 ↓ 64.3% Number of times negative pressure is interrupted due to equipment failure times / week 20 Absolute improvement surface

[0117] From the cleanliness risk indicators in Table 1, the average concentration of 0.3µm particulate matter in the intensive care unit decreased from 3120 particles / L to 2540 particles / L, and the peak concentration decreased from 6480 particles / L to 4920 particles / L. This indicates that the present invention not only reduced the daily pollution baseline but also had a stronger ability to suppress disturbance peaks. This directly corresponds to the present invention's "cross-regional coordinated allocation of purification resources" achieved through regional state sets and multi-agent collaborative optimization. Traditional control only increases purification locally after exceeding the standard, while the present invention can coordinate and respond in advance during the risk escalation stage, thereby lowering the peak value and shortening the duration of high risk. The equivalent peak value of pathogen aerosols in the isolation ward area decreased by 33.9%, which is the most significant improvement in cleanliness risk control, reflecting that the improved SCSO algorithm can still stably provide feasible and near-optimal collaborative control actions under strong disturbances.

[0118] From the perspective of pressure difference and airflow safety, the pressure difference compliance rate of this invention reaches 99.1%, and the minimum negative pressure also has a greater "margin." This indicates that the swarm intelligent dynamic control can simultaneously consider the pressure difference target when adjusting the purification power and air volume, avoiding the problem of "a sudden increase in local air supply causing a short-term reversal in pressure difference" in traditional threshold linkage. The response time has been shortened from 120 seconds to 45 seconds, demonstrating that adaptive search scheduling and neighborhood collaborative interaction can quickly converge to effective control actions when disturbances occur, transforming from passive over-limit triggering to active and rapid suppression.

[0119] From the perspective of energy consumption and comfort, the total daily air supply volume of the area decreased by approximately 10%, but the cleanliness index was actually better. This indicates that the present invention achieves "precise supply based on risk," concentrating purification and ventilation resources in high-risk areas and high-risk periods, while automatically reducing load during non-high-risk periods, resulting in a 18.8% decrease in daily energy consumption. The number of complaints about medical staff comfort decreased by 64.3%, which is corroborated by the halving of the temperature fluctuation range, indicating that under the multi-objective optimization framework, comfort and energy consumption objectives were not sacrificed, but rather dynamically balanced together with the cleanliness risk objective. Finally, the present invention achieved zero negative pressure interruptions under simulated fault conditions, demonstrating that through closed-loop judgment of control execution feedback data and a neighborhood target redistribution mechanism, the system possesses the robust fault tolerance required in actual medical scenarios and can continuously and stably maintain the purification control effect.

[0120] Overall, this embodiment fully demonstrates the systematic improvement brought about by the present invention to address the problems of "reaction lag, inconsistency between local control and global objectives, high energy consumption and unstable differential pressure, and lack of collaborative fault tolerance in the event of failure" in the prior art, and verifies the beneficial effects with two consecutive weeks of real working condition comparison data.

Claims

1. A dynamic control method for a medical purification system based on swarm intelligence, characterized in that, The process includes the following steps: collecting and preprocessing environmental and equipment status data from the medical purification area; fusing the preprocessed environmental and equipment status data to generate a regional status vector, forming a regional status set; defining each purification unit, air supply unit, return air unit, fan unit, valve unit, and disinfection unit in the medical purification system as an agent, and establishing a local observation vector, neighborhood communication set, and control action set for each agent; obtaining a global control target set and a set of operational constraints based on the regional status set and the local observation vectors of each agent, and decomposing the global control target set into a set of local control targets; using an improved SCSO algorithm to perform swarm intelligent collaborative optimization iterative search on the control action set based on the set of local control targets, the set of operational constraints, and the set of neighborhood communication, obtaining a set of candidate control actions; performing consistency and constraint checks on the candidate control actions in the set of candidate control actions to obtain a set of collaborative control actions; executing control on the corresponding units of each agent according to the set of collaborative control actions, forming control execution feedback data; performing failure determination based on the control execution feedback data, generating failure determination results, reallocating the local control target set corresponding to the failed agent, and updating the local control target set.

2. The dynamic control method for a medical purification system based on swarm intelligence according to claim 1, characterized in that, The environmental status data includes particulate matter concentration, pathogenic aerosol concentration, volatile organic compound concentration, carbon dioxide concentration, temperature, humidity, pressure difference, supply air volume, return air volume, and personnel activity density data. The equipment status data includes the operating status data of the purification unit, supply air unit, return air unit, fan unit, valve unit, and disinfection unit. The preprocessing includes time synchronization, anomaly removal, missing data completion, and noise filtering.

3. The dynamic control method for a medical purification system based on swarm intelligence according to claim 1, characterized in that, The formation of the regional state set specifically includes: matching the preprocessed environmental state data and equipment state data according to the spatial location of environmental monitoring nodes and equipment monitoring nodes, based on the medical purification area, to obtain the environmental state subset and equipment state subset corresponding to each medical purification area; aligning and splicing the environmental state subset and equipment state subset of each medical purification area to generate the regional fusion vector corresponding to the medical purification area; performing multi-source fusion processing of evidence reasoning on the regional fusion vector, wherein the multi-source fusion processing of evidence reasoning constructs each data item in the regional fusion vector into a corresponding evidence item and generates a basic probability allocation, synthesizes each evidence item according to a predetermined evidence synthesis rule, and outputs the regional state vector corresponding to the medical purification area; and collecting the regional state vectors of all medical purification areas according to the medical purification area sequence number to form the regional state set.

4. The dynamic control method for a medical purification system based on swarm intelligence according to claim 1, characterized in that, The establishment of the local observation vector, neighborhood communication set, and control action set specifically includes: uniquely identifying each purification unit, air supply unit, return air unit, fan unit, valve unit, and disinfection unit within the medical purification system, and defining each uniquely identified unit as an intelligent agent, forming an intelligent agent set; within each control cycle, calling the corresponding unit's equipment status data and the environmental status data of the medical purification area where the intelligent agent is located for each intelligent agent in the intelligent agent set, and combining them in a predetermined order to generate the intelligent agent's local observation vector; based on the spatial positional relationship between each intelligent agent and the predetermined neighborhood communication criteria, obtaining intelligent agents that meet the neighborhood communication conditions for each intelligent agent in the intelligent agent set, forming an intelligent agent's neighborhood communication set, wherein the neighborhood communication conditions refer to the spatial distance between intelligent agents being less than a predetermined neighborhood communication distance threshold and the communication link being in a connected state; for each intelligent agent in the intelligent agent set, constructing the intelligent agent's control action set according to the executable control quantity type of the unit corresponding to the intelligent agent, wherein the control action set includes control actions corresponding to purification power, fan speed, valve opening, air supply volume, return air volume, airflow path, and differential pressure mode.

5. The dynamic control method for a medical purification system based on swarm intelligence according to claim 1, characterized in that, The generation of the local control target set specifically includes: constructing a global control target set based on the regional state vector and local observation vector. The construction process involves extracting key state variables such as cleanliness risk, energy consumption, airflow organization, and pressure difference from the regional state vector of the medical purification area and the local observation vectors of each agent according to pre-defined target items, generating corresponding cleanliness risk targets, energy consumption targets, airflow organization targets, and pressure difference targets, thus forming the global control target set; and constructing an operational constraint set based on the regional state vector and local observation vector. The construction process involves reading the regional state vector and local observation vector in each control cycle, locating and... The state variables corresponding to control actions, differential pressure, supply air volume, return air volume, and equipment operating status are categorized and assigned upper and lower limits according to pre-set constraints and threshold ranges, resulting in control action boundary constraints, area differential pressure constraints, supply air volume constraints, return air volume constraints, and equipment operating status constraints, forming an operating constraint set. Based on the control action set of each intelligent agent's medical purification area and the corresponding unit of each intelligent agent, a one-to-one correspondence is established between the control action set and the global control target set. According to the one-to-one correspondence, the global control target set is decomposed into local control target sets corresponding to each intelligent agent.

6. The dynamic control method for a medical purification system based on swarm intelligence according to claim 1, characterized in that, The specific steps for obtaining the candidate control action set include: within each control cycle, establishing a search population for each agent in the agent set using an improved SCSO algorithm; randomly generating a predetermined number of search individuals within the control action boundary constraints corresponding to the agent's control action set, forming an initial search individual set for the agent; the improved SCSO algorithm includes an adaptive search scheduling unit, a neighborhood cooperative interaction unit, and a constraint robust processing unit; the adaptive search scheduling unit uses the initial search individual set as an iteration object to obtain a step size adjustment factor; and the neighborhood cooperative interaction unit introduces a chaotic mapping cooperative perturbation mechanism to form a cooperative perturbation mechanism. With the same update step size, the updated search individuals are obtained. The constraint robust processing unit refers to performing runtime constraint set verification and control action boundary constraint verification on each control action corresponding to the search individual to obtain the corrected search individual and recalculate the fitness value. In the adaptive search scheduling unit, the initial set of search individuals is used as the iteration object. Within each iteration step, the search strength coefficient is calculated based on the correspondence between the current iteration step size and the preset maximum iteration step size, and the step size adjustment factor of each search individual is obtained based on the search strength coefficient. Based on the local control target set and the runtime constraint set, the fitness value is calculated for each search individual of each agent. The fitness value is obtained by weighting and summing the local target values ​​corresponding to the cleanliness risk target, energy consumption target, airflow organization target, and pressure difference target according to preset weights. In the neighborhood cooperative interaction unit, for each agent, the neighborhood optimal search individual with the best fitness value in the neighborhood communication set is obtained within each iteration step, and the global optimal search individual with the best fitness value is obtained among all search individuals of all agents. A chaotic mapping cooperative perturbation mechanism is introduced to generate a perturbation factor, and the step size adjustment factor is coupled with the perturbation factor to form a cooperative update step size, resulting in the updated search individual. In constraint robust processing... In this unit, feasibility constraint processing is performed on the updated search individuals. This feasibility constraint processing refers to performing operation constraint set verification and control action boundary constraint verification on each control action corresponding to the search individual. Control action components that do not meet the operation constraint set or control action boundary constraints are corrected according to the corresponding constraint thresholds, and the fitness value of the corrected search individuals is recalculated. The search population is iteratively updated. When the number of iterations reaches a preset maximum value, the control action corresponding to the search individual with the best fitness value for each agent is selected as the candidate control action for the agent. The candidate control actions of each agent are then aggregated to form a candidate control action set.

7. The dynamic control method for a medical purification system based on swarm intelligence according to claim 1, characterized in that, The specific steps for obtaining the collaborative control action set include: aligning and comparing the control quantities corresponding to the control action set in the candidate control action set item by item to generate a consistency evaluation value for the agent's candidate control actions; for each agent's candidate control action, performing constraint verification item by item based on the set of operational constraints and the boundary constraints of the control action set corresponding to the agent's control action set to generate a constraint verification result for the agent's candidate control action; for each agent, selecting the candidate control action with the optimal consistency evaluation value from the candidate control actions whose constraint verification result is passed to obtain the agent's collaborative control action; and compiling all the agent's collaborative control actions according to the agent's sequence number to form a collaborative control action set.

8. The dynamic control method for a medical purification system based on swarm intelligence according to claim 1, characterized in that, The formation of the control execution feedback data specifically includes: assigning the collaborative control actions in the collaborative control action set to the corresponding intelligent agents according to the one-to-one correspondence between intelligent agents and corresponding units; unpacking the collaborative control actions of each intelligent agent to obtain control quantity setting values ​​corresponding to each item in the intelligent agent's control action set, the control quantity setting values ​​including purification power setting value, fan speed setting value, valve opening setting value, supply air volume setting value, return air volume setting value, airflow path switching setting value, and differential pressure mode setting value; sending the control quantity setting values ​​to the execution interface of the corresponding unit of the intelligent agent to execute purification power adjustment, fan speed adjustment, valve opening adjustment, supply air volume adjustment, return air volume adjustment, airflow path switching, and differential pressure mode switching; after the corresponding unit completes the execution, collecting the actual execution result data of the corresponding unit, the actual execution result data including actual purification power data, actual fan speed data, actual valve opening data, actual supply air volume data, actual return air volume data, actual airflow path status data, and actual differential pressure mode status data; combining the actual execution result data of each intelligent agent's corresponding unit to form control execution feedback data.

9. The dynamic control method for a medical purification system based on swarm intelligence according to claim 1, characterized in that, The update of the local control target set specifically includes: aggregating control execution feedback data according to agent number to obtain a subset of feedback data for each agent's corresponding unit; comparing the subset of feedback data with the corresponding collaborative control action of the agent to generate control deviation data for the agent; based on the control deviation data and the device operating status indication information in the subset of feedback data, performing a failure determination on each agent and generating a failure determination result; when the failure determination result indicates that the agent has failed, marking the agent as a failed agent and freezing it in the candidate control action within the current control cycle; and based on the neighborhood communication set of the failed agent, reallocating the local control target set corresponding to the failed agent according to a predetermined reallocation rule, thereby updating the local control target set.

Images