A system, method, and storage medium for precise control of environmental factors across multiple scenarios.

Abstract

The present application relates to the technical field of environmental regulation, and specifically relates to a multi-scene-oriented environmental factor precision regulation system and method and a storage medium, comprising a perception layer, an edge computing layer, a cloud service layer and an execution layer; the perception layer is deployed inside a target scene and comprises multiple types of physical sensors, a device state acquisition unit and a scene attribute input interface, and the multiple types of physical sensors at least include a temperature and humidity sensor, an illumination sensor, an air quality sensor and a wind speed sensor. Through a three-layer architecture of a scene dynamic identification module, a virtual sensor completion module and a multi-target collaborative optimization module, the system can meet the differentiated regulation requirements of multiple types of scenes, solve the technical problems of weak cross-scene adaptation capability, low perception data reliability and insufficient regulation strategy collaboration in the prior art, and achieve multi-target balance of regulation precision, energy consumption cost and scene adaptability.

Description

Technical Field

[0001] This invention relates to the field of environmental control technology, specifically to a system, method, and storage medium for precise control of environmental factors in multiple scenarios. Background Technology

[0002] With the development of IoT and AI technologies, precise environmental control technology has been applied in many fields. Among the existing publicly available technologies, environmental control systems for building scenarios mostly focus on adapting to single scenarios such as temperature, humidity, and air quality. Control systems for laboratory scenarios only address the fixed parameter threshold requirements of experimental procedures, while control systems for agricultural scenarios only adapt to the specific parameter requirements for crop growth.

[0003] The above-mentioned systems generally suffer from three major technical defects: First, they lack cross-scenario adaptability. Existing systems are all custom-developed for vertical domains and cannot achieve a single architecture that is compatible with the differentiated control requirements of multiple scenarios. When deploying in different scenarios, the system architecture needs to be redeveloped, resulting in high deployment costs and long cycles. Second, the reliability of data in the perception layer is insufficient. Existing systems rely on fixed physical sensors to collect data. When sensors malfunction, signals are blocked, or the deployment density is insufficient, data loss or deviation is likely to occur, which in turn leads to inaccurate control decisions. Third, the coordination of control strategies is insufficient. Existing systems mostly adopt single-factor independent control logic, without considering the coupling effect between different environmental factors, and fail to achieve a multi-objective dynamic balance of energy consumption, control effect, and scenario adaptation requirements, which easily leads to problems such as excessive energy consumption or insufficient control precision.

[0004] Therefore, given that neither the published patents nor the existing technologies have achieved a collaborative regulation architecture for environmental factors in multiple scenarios, nor have they constructed a dynamic switching mechanism for regulation priorities that adapts to different scenarios, there are obvious technical limitations. Thus, it is urgent to develop a precise regulation system, method, and storage medium for environmental factors in multiple scenarios to solve the above problems. Summary of the Invention

[0005] To address the technical problems existing in the prior art, this invention proposes a system, method, and storage medium for precise control of environmental factors in multiple scenarios. Through a three-layer architecture consisting of a scene dynamic identification module, a virtual sensing completion module, and a multi-objective collaborative optimization module, a single system can meet the differentiated control needs of multiple scenarios. This solves the technical problems of weak cross-scenario adaptability, low reliability of sensing data, and insufficient synergy of control strategies in the prior art, while achieving a multi-objective balance between control accuracy, energy consumption cost, and scenario adaptability.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A precise control system for environmental factors across multiple scenarios, comprising a perception layer, an edge computing layer, a cloud service layer, and an execution layer; The perception layer is deployed inside the target scene and includes multiple types of physical sensors, device status acquisition units, and scene attribute input interfaces. The multiple types of physical sensors include at least temperature and humidity sensors, light sensors, air quality sensors, and wind speed sensors. The device status acquisition units are used to collect the operating power and operating status data of all electrical equipment in the scene. The scene attribute input interface supports users to upload basic attribute data such as scene type, core control targets, and parameter threshold ranges. The perception layer uploads all the collected raw data to the edge computing layer via the MQTT protocol. The edge computing layer is deployed locally in the scene and includes a scene dynamic recognition module, a virtual sensing completion module, and a local emergency control module. The scene dynamic recognition module is used to match scene features based on the data uploaded by the perception layer, dynamically identify the current scene type and control priority. The virtual sensing completion module is used to complete abnormal and missing data collected by physical sensors and generate highly reliable environmental parameter estimates. The local emergency control module is used to call the basic control strategy stored locally to ensure the basic protection of environmental parameters and avoid parameter over-limit problems in the scene when the network is interrupted or the cloud service fails. The cloud service layer includes a scene feature library, a multi-objective collaborative optimization module, and a model iteration update module. The scene feature library stores feature tags, parameter thresholds, and control priority rules for five basic scenarios: agricultural planting, livestock and poultry breeding, smart home, industrial cleanroom, and public building operation and maintenance. The multi-objective collaborative optimization module is used to generate a globally optimal control strategy by combining current environmental parameters, scenario control requirements, real-time energy consumption prices, and external meteorological data. The model iteration update module is used to continuously optimize the accuracy of the virtual sensing model and the collaborative optimization model based on the feedback data after control. The execution layer includes various types of control terminals such as air conditioning, fresh air systems, lighting equipment, irrigation equipment, ventilation equipment, and heating equipment, which are used to receive control instructions issued by the edge computing layer and execute corresponding operations to achieve dynamic adjustment of environmental factors.

[0007] In some specific embodiments, the specific operation flow of the scene dynamic recognition module includes: First, the raw data uploaded by the perception layer is standardized, and the data from different types of sensors are uniformly mapped to the standardized interval of [0,1] to eliminate the influence of different data dimensions. At the same time, three core features of the data are extracted: time series features, fluctuation amplitude features, and coupling correlation features, to form the feature vector of the current scene. The extracted feature vectors are matched with the scene feature library synchronized from the cloud to the local machine using cosine similarity. When the matching degree of the scene feature with the highest similarity is greater than 0.85, it is directly determined as the corresponding scene type. At the same time, the preset control priority rule corresponding to the scene is called. When the matching degree is less than 0.85, the scene customization process is triggered to guide the user to supplement the scene attribute information and store the new scene features into the scene feature library to realize the dynamic expansion of the feature library.

[0008] In some specific implementations, the virtual sensing completion module adopts a physical-data hybrid driven architecture, and its specific operating mechanism includes: First, a physical mechanism model for the corresponding scenario is constructed. Based on the principle of environmental dynamics, the coupling relationship between different environmental factors is described, such as the relationship between temperature and humidity and ventilation volume, and the relationship between light intensity and indoor temperature. At the same time, no less than 30 days of historical environmental data and equipment operation data in this scenario are collected as training set to train a lightweight LSTM neural network as a data-driven model. When abnormal or missing data from physical sensors is detected, other normally collected environmental factor data and equipment operation data are first input into the physical mechanism model to obtain the first estimate. At the same time, historical time series data are input into the LSTM model to obtain the second estimate. Dynamic weights are used to fuse the two estimates. The fusion weights are dynamically adjusted according to the estimation errors of the two models in the last 7 days. The final fused estimate is used as the completed environmental parameter.

[0009] In some specific embodiments, the multi-objective collaborative optimization module adopts a three-layer optimization architecture, specifically including: The first layer is the constraint layer, which determines the allowable fluctuation range of all environmental factors, the operating constraints of the control equipment, and the upper limit constraints of energy consumption based on the identified scenario type. All control strategies must meet all the restrictions of the constraint layer to avoid parameter overruns or equipment damage. The second layer is the priority layer, which determines the weight coefficients of three objectives—control accuracy, energy consumption cost, and control response speed—based on the scenario control priority. For example, in the industrial cleanroom scenario, the weight of control accuracy is set to 0.7, the weight of energy consumption cost is set to 0.2, and the weight of response speed is set to 0.1; in the public building scenario, the weight of energy consumption cost is set to 0.6, the weight of control accuracy is set to 0.3, and the weight of response speed is set to 0.1; and in the agricultural planting scenario, the weight of control accuracy is set to 0.5, the weight of energy consumption cost is set to 0.4, and the weight of response speed is set to 0.1. The third layer is the optimization layer, which uses an improved particle swarm optimization algorithm to search for the optimal control strategy in the feasible solution space. The highest weighted score of the three objectives is used as the optimization objective. The final output control strategy includes the specific operating parameters, runtime, and start / stop time points of each execution device.

[0010] In some specific embodiments, the operating mechanism of the local emergency control module includes: When the network connection to the cloud is interrupted for more than 5 minutes, or the cloud service returns to a fault state, the local emergency control mode is automatically triggered, and the basic control rules for the corresponding scenario are called in the local storage. The basic control rules are based on threshold triggering logic. When the environmental parameters exceed the preset safety threshold range, the corresponding control device is activated to bring the parameters back to the safe range. During the emergency control mode operation, the edge computing layer continuously attempts to re-establish the connection with the cloud. When the connection is restored, all collected data and control records during the network outage are automatically synchronized to the cloud. The cloud optimizes the control strategy based on the data during the network outage and distributes it to the edge layer. The system switches back to the normal control mode, and at the same time, the control effect data during the network outage is used for model iteration and update.

[0011] As part of the same inventive concept, this invention also provides a method for precise control of environmental factors in multiple scenarios, comprising the following steps: S1. The perception layer collects raw environmental data and device operation data in the target scene, and also supports users to enter scene attribute data. All data is uploaded to the edge computing layer through an encrypted transmission protocol. The edge computing layer cleans and preprocesses the raw data, filtering out abnormal data that obviously does not conform to physical logic. S2. The scene dynamic recognition module of the edge computing layer extracts the core features of the preprocessed data and matches them with the scene feature library stored locally. It dynamically identifies the current scene type and the corresponding control priority rules. If the matching degree is insufficient, it guides the user to supplement scene information and update the scene feature library. S3. Check the data integrity of all physical sensors. If there is missing or abnormal data, call the virtual sensing completion module to fuse the physical mechanism model and the data-driven model to generate the completed environmental parameter estimates and form a complete environmental parameter dataset. S4. The edge computing layer uploads the complete environmental parameter dataset, scene control requirements, and equipment operation status data to the cloud service layer. The cloud-based multi-objective collaborative optimization module combines real-time energy consumption prices and external weather forecast data to generate the global optimal control strategy and distribute it to the edge computing layer. S5, the edge computing layer breaks down the control strategy into specific control commands for each execution device, sends them to the execution layer devices for execution, and continuously collects environmental parameter feedback data after control, and synchronizes the control effect data to the model iteration update module in the cloud to optimize the accuracy of the virtual sensing model and the collaborative optimization model.

[0012] In some specific embodiments, the specific process of scene dynamic recognition in step S2 includes: First, feature extraction is performed on the preprocessed multi-dimensional data. The extracted features include three categories of 12 features: the daily fluctuation range of each environmental factor, the Pearson correlation coefficient between different factors, and the distribution pattern of equipment running time. The 12 features are combined into the feature vector of the current scene. All features are standardized to eliminate the influence of dimensions. Calculate the cosine similarity between the current feature vector and all stored scene feature vectors in the scene feature library. Select the top 3 candidate scenes with the highest similarity. If the highest similarity is greater than 0.85, it is directly determined as the corresponding scene type. At the same time, the control priority and parameter threshold rules corresponding to the scene are called. If the highest similarity is between 0.6 and 0.85, the 3 candidate scenes are displayed to the user for the user to confirm the final scene type. If the highest similarity is less than 0.6, the new scene entry process is triggered, guiding the user to fill in the scene type, core control target, and parameter threshold range information, and the new scene features are stored in the scene feature library.

[0013] In some specific embodiments, the specific process of virtual sensing completion in step S3 includes: First, for the currently identified scene type, the corresponding physical mechanism model is called. The physical mechanism model is built based on the principles of mass conservation and energy conservation. The input is the normally collected environmental parameters and equipment operating parameters, and the output is the first estimated value of the missing or abnormal parameters. At the same time, the lightweight LSTM data-driven model corresponding to the scene is called. The input is the historical time series data of the parameter and the real-time data of other related parameters, and the output is the second estimated value. The dynamic fusion weight is calculated based on the average absolute error of the two models on this parameter over the past 7 days. The fusion weight of the model is inversely proportional to its average absolute error. The two estimated values ​​are weighted and summed according to the fusion weight to obtain the final completed value. When the physical sensor resumes normal data acquisition, the error between the completed value and the actual acquired value is used as feedback data to optimize the parameters of the two models.

[0014] In some specific implementations, the specific process of multi-objective collaborative optimization in step S4 includes: First, all constraints are determined based on the scenario type, including the allowable fluctuation range of each environmental parameter, the maximum power of each control device, the longest continuous operating time, and the total energy consumption limit. All candidate control strategies must meet all constraints, and infeasible solutions that do not meet the constraints are eliminated. Based on the control priority of the scenario, the weight coefficients of the three optimization objectives—control accuracy, energy consumption cost, and response speed—are determined. An improved particle swarm optimization algorithm is used to search for the optimal solution in the feasible solution space. Each particle represents a set of control strategies. The highest weighted comprehensive score of the three objectives is used as the optimization target. The number of iterations for optimization is no less than 100. The final output optimal control strategy includes the specific operating parameters and start-stop time of each execution device. The time granularity of the control strategy is no less than 15 minutes. Based on generating and executing the optimal control strategy, a control effect evaluation mechanism is also set up to dynamically verify the effect of the strategy implementation: the control effect of the day is comprehensively evaluated every 24 hours. The evaluation indicators include four dimensions: parameter compliance rate, average control error, total energy consumption, and equipment failure rate. The parameter compliance rate is the percentage of time when environmental parameters are within the allowable range. The average control error is the average absolute deviation between the actual parameters and the target parameters. The total energy consumption is the sum of the daily power consumption of all control equipment. The equipment failure rate is the percentage of time when equipment operates abnormally. The evaluation results are compared with the preset optimization targets. If the evaluation results are better than the results of the previous period, the control strategy samples of the current period are added to the model training set to optimize the generation of subsequent control strategies. If the evaluation results do not meet expectations, the weight coefficients of the multi-objective optimization module or the parameters of the virtual sensing model are automatically adjusted until the evaluation results meet the preset targets.

[0015] As part of the same inventive concept, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of a method for precise control of environmental factors for multiple scenarios are implemented, and the storage medium includes, but is not limited to, hard disks, solid-state drives, USB flash drives, CD-ROMs, RAM, ROMs, EPROMs, flash memory, and magneto-optical storage. The storage medium also stores scene feature library files, virtual sensing model parameter files, and multi-objective optimization algorithm parameter files. When the processor calls these files to implement the corresponding method steps, it supports the operation of basic functions in offline mode. When the network is restored, it automatically synchronizes the latest model parameters and feature library data to ensure the reliability of the system in different network environments.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. The cross-scenario adaptability has been greatly improved. Through the scene dynamic recognition module and the transferable virtual sensing model, a single system can meet the control needs of more than 5 types of scenarios, including agriculture, breeding, home, industry, and public buildings. The deployment cycle is shortened and the deployment cost is reduced.

[0017] 2. The reliability of sensing data is significantly improved. By integrating equipment operation data, historical environmental data and external meteorological data through the virtual sensing completion module, even if the failure rate of physical sensors reaches 30%, the environmental parameter estimation error can still be low, ensuring the accuracy of control decisions.

[0018] 3. The regulation efficiency is significantly optimized. Through the multi-objective collaborative optimization module, a dynamic balance is achieved between regulation accuracy, energy consumption cost, and scenario adaptability. While ensuring that the regulation accuracy meets the scenario requirements, the average energy consumption is reduced. At the same time, it can adapt to the differentiated priority requirements of different scenarios, such as prioritizing regulation accuracy in industrial scenarios and prioritizing energy consumption optimization in public building scenarios. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention, but do not constitute a limitation thereof. In the drawings: Figure 1 The flowchart of the multi-scenario precise control system for environmental factors of the present invention is shown below; Figure 2 The flowchart of the method for precise control of environmental factors in multiple scenarios according to the present invention is shown below. Figure 3 This is an internal architecture diagram of the multi-objective collaborative optimization module in the multi-scenario-oriented environmental factor precise control system of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0021] Unless otherwise stated, all percentages, parts, ratios, etc. in this document are by weight.

[0022] The materials, methods, and embodiments described herein are exemplary and should not be construed as limiting unless otherwise stated.

[0023] Example 1 This embodiment provides a system, method, and storage medium for precise control of environmental factors in multiple scenarios. It is designed for the precise control of environmental factors in vegetable greenhouses. The deployed sensing layer includes four temperature and humidity sensors, two light sensors, three soil moisture sensors, and one CO2 concentration sensor. The device status acquisition unit connects to five types of control devices in the greenhouse, including ventilators, shade curtains, irrigation pumps, heating lamps, and CO2 generators. The edge computing layer uses an embedded gateway deployed locally in the greenhouse, and the cloud service layer is deployed on a public cloud server.

[0024] After the system starts, the raw data collected by the perception layer is uploaded to the edge computing layer. The scene dynamic recognition module extracts data features and matches them with the scene feature library. The matching degree reaches 0.92, which determines that it is a leafy vegetable planting scene. The corresponding control priority is control accuracy weight 0.5, energy consumption cost weight 0.4, and response speed weight 0.1. The allowable temperature and humidity range is daytime temperature 22-28℃, nighttime temperature 15-20℃, relative humidity 60%-80%, light intensity 10000-20000 lux, and soil moisture 60%-70%.

[0025] During operation, one of the temperature and humidity sensors experienced data loss due to a circuit fault. The virtual sensing completion module was activated. First, it called the physical mechanism model of the agricultural greenhouse and input data from the other three normal temperature and humidity sensors, the operating status of the ventilation fan, and the light intensity data to obtain the first estimated value of temperature as 24.7℃ and the first estimated value of humidity as 72%. At the same time, it called the LSTM data-driven model and input the time series data of the sensor over the past 7 days and other related parameter data to obtain the second estimated value of temperature as 25.1℃ and the second estimated value of humidity as 70%. Based on the error of the two models over the past 7 days, the weight of the physical model was 0.55 and the weight of the data model was 0.45. After fusion, the completed temperature was 24.9℃ and the humidity was 71.1%, with completion errors of only 0.3℃ and 0.9%, respectively, ensuring the reliability of the data.

[0026] The cloud-based multi-objective collaborative optimization module, combined with the real-time electricity price of the day (1.2 yuan / kWh during peak hours, 0.7 yuan / kWh during normal hours, and 0.3 yuan / kWh during off-peak hours) and the weather forecast for the next 3 hours (external temperature will rise by 3°C and light intensity will increase by 20%), generates an optimized control strategy: During the off-peak electricity period, from 2-5 AM, the heating lamps are activated in advance to raise the temperature to 19°C, reducing heating energy consumption during the daytime peak hours; during the increased light intensity period, from 10-12 AM, the shading curtains are opened appropriately to prevent excessive temperature; and the operation time of the ventilation fans is adjusted to the normal electricity price period. The final control results show that the growth cycle of greenhouse vegetables was shortened by 7% that month, the incidence of pests and diseases decreased by 12%, and the total control energy consumption decreased by 31%, achieving the dual goals of increasing yield and reducing energy consumption.

[0027] Example 2 This embodiment provides a system, method, and storage medium for precise control of environmental factors in multiple scenarios. For the application of precise control of environmental factors in a 20-story Grade A office building, the sensing layer deploys 8 temperature and humidity sensors, 4 CO2 concentration sensors, and 2 people flow statistics sensors on each floor. The device status acquisition unit is connected to the central air conditioning system, fresh air system, and lighting system of the entire building. The edge computing layer uses edge nodes deployed on each floor, and a core edge gateway is set up for the entire building. The cloud service layer is deployed on the enterprise's private cloud.

[0028] After the system starts, the scene dynamic recognition module matches the public office building scene. The control priority is energy consumption cost weight 0.6, control accuracy weight 0.3, and response speed weight 0.1. The allowable temperature and humidity range is 22-26℃, relative humidity 40%-60%, and CO2 concentration less than 1000ppm on weekdays from 9 am to 6 pm. On non-working days, only basic environmental parameters are maintained.

[0029] During operation, two temperature and humidity sensors on the third floor experienced data discrepancies due to obstruction caused by renovations. The virtual sensor completion module integrated data from six other sensors on the same floor, central air conditioning parameters, and pedestrian traffic data to generate complete temperature data with an error of only 0.4℃, ensuring the accuracy of the control. The cloud-based multi-objective collaborative optimization module combined daily attendance predictions (75% attendance rate), real-time electricity prices, and weather forecasts to generate control strategies: pre-starting the air conditioning one hour before work on weekdays to adjust the temperature to 24℃; dynamically adjusting the air conditioning supply and fresh air volume on each floor based on pedestrian traffic; and turning off the air conditioning half an hour before the end of the workday to maintain a comfortable temperature using residual heat; and maintaining only basic ventilation on weekends. The final implementation results showed that the annual energy consumption of the entire office building was reduced by 29%, employee comfort satisfaction increased by 22%, and the energy consumption requirements for a green building one-star certification were met.

[0030] Example 3 This embodiment provides a system, method, and storage medium for precise control of environmental factors in multiple scenarios. For the application of precise control of environmental factors in cleanrooms for chip manufacturing, the sensing layer is equipped with 12 temperature and humidity sensors, 6 particle concentration sensors, and 4 differential pressure sensors. The equipment status acquisition unit is connected to the clean air conditioning system, fresh air filtration system, and dehumidification system. The edge computing layer adopts an industrial-grade edge gateway, and the cloud service layer is deployed in the enterprise's local data center.

[0031] After system startup, the scene dynamic recognition module matches the industrial cleanroom scenario. The control priority is set as follows: control accuracy weight 0.7, energy consumption cost weight 0.2, and response speed weight 0.1. The allowable parameter range is: temperature 22±0.5℃, relative humidity 45±3%, particle concentration less than 100 particles / cubic meter, and pressure difference between the workshop and the outside greater than 10Pa. During operation, the virtual sensing completion module completes the data for occasional fluctuations in sensor data, ensuring that the data error is less than 0.2℃ and 1%RH. The cloud-based multi-objective collaborative optimization module optimizes the equipment operating parameters while meeting the accuracy requirements, ultimately achieving a control parameter compliance rate of 99.98%, a product yield improvement of 1.2%, and an energy consumption reduction of 18%, thus balancing production quality requirements with energy consumption optimization needs.

[0032] In conjunction with Examples 1-3, the low-odor, environmentally friendly polyether gear oil additive composition and the gear oil in which it is applied, provided by the present invention, successfully solve the problem of the pungent odor of traditional gear oil in enclosed environments such as vertical mills. While ensuring excellent lubrication performance, it significantly improves occupational health and safety levels and has important value for promotion and application.

[0033] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A precise control system for environmental factors in multiple scenarios, characterized in that, It includes the perception layer, edge computing layer, cloud service layer, and execution layer; The perception layer is deployed inside the target scene and includes multiple types of physical sensors, device status acquisition units, and scene attribute input interfaces. The multiple types of physical sensors include at least temperature and humidity sensors, light sensors, air quality sensors, and wind speed sensors. The device status acquisition units are used to collect the operating power and operating status data of all electrical equipment in the scene. The scene attribute input interface supports users to upload basic attribute data such as scene type, core control targets, and parameter threshold ranges. The perception layer uploads all the collected raw data to the edge computing layer via the MQTT protocol. The edge computing layer is deployed locally in the scene and includes a scene dynamic recognition module, a virtual sensing completion module, and a local emergency control module. The scene dynamic recognition module is used to match scene features based on the data uploaded by the perception layer, dynamically identify the current scene type and control priority. The virtual sensing completion module is used to complete abnormal and missing data collected by physical sensors and generate highly reliable environmental parameter estimates. The local emergency control module is used to call the basic control strategy stored locally to ensure the basic protection of environmental parameters and avoid parameter over-limit problems in the scene when the network is interrupted or the cloud service fails. The cloud service layer includes a scene feature library, a multi-objective collaborative optimization module, and a model iteration update module. The scene feature library stores feature tags, parameter thresholds, and control priority rules for five basic scenarios: agricultural planting, livestock and poultry breeding, smart home, industrial cleanroom, and public building operation and maintenance. The multi-objective collaborative optimization module is used to generate a globally optimal control strategy by combining current environmental parameters, scenario control requirements, real-time energy consumption prices, and external meteorological data. The model iteration update module is used to continuously optimize the accuracy of the virtual sensing model and the collaborative optimization model based on the feedback data after control. The execution layer includes various types of control terminals such as air conditioning, fresh air systems, lighting equipment, irrigation equipment, ventilation equipment, and heating equipment, which are used to receive control instructions issued by the edge computing layer and execute corresponding operations to achieve dynamic adjustment of environmental factors.

2. The precise control system for environmental factors in multiple scenarios according to claim 1, characterized in that, The specific operation process of the scene dynamic recognition module includes: First, the raw data uploaded by the perception layer is standardized, and the data from different types of sensors are uniformly mapped to the standardized interval of [0,1] to eliminate the influence of different data dimensions. At the same time, three core features of the data are extracted: time series features, fluctuation amplitude features, and coupling correlation features, to form the feature vector of the current scene. The extracted feature vectors are matched with the scene feature library synchronized from the cloud to the local machine using cosine similarity. When the matching degree of the scene feature with the highest similarity is greater than 0.85, it is directly determined as the corresponding scene type. At the same time, the preset control priority rule corresponding to the scene is called. When the matching degree is less than 0.85, the scene customization process is triggered to guide the user to supplement the scene attribute information and store the new scene features into the scene feature library to realize the dynamic expansion of the feature library.

3. The precise control system for environmental factors in multiple scenarios according to claim 1, characterized in that, The virtual sensing completion module adopts a physical-data hybrid driven architecture, and its specific operating mechanism includes: First, a physical mechanism model for the corresponding scenario is constructed. Based on the principle of environmental dynamics, the coupling relationship between different environmental factors is described, such as the relationship between temperature and humidity and ventilation volume, and the relationship between light intensity and indoor temperature. At the same time, no less than 30 days of historical environmental data and equipment operation data in this scenario are collected as training set to train a lightweight LSTM neural network as a data-driven model. When abnormal or missing data from physical sensors is detected, other normally collected environmental factor data and equipment operation data are first input into the physical mechanism model to obtain the first estimate. At the same time, historical time series data are input into the LSTM model to obtain the second estimate. Dynamic weights are used to fuse the two estimates. The fusion weights are dynamically adjusted according to the estimation errors of the two models in the last 7 days. The final fused estimate is used as the completed environmental parameter.

4. The precise control system for environmental factors in multiple scenarios according to claim 1, characterized in that, The multi-objective collaborative optimization module adopts a three-layer optimization architecture, specifically including: The first layer is the constraint layer, which determines the allowable fluctuation range of all environmental factors, the operating constraints of the control equipment, and the upper limit constraints of energy consumption based on the identified scenario type. All control strategies must meet all the restrictions of the constraint layer to avoid parameter overruns or equipment damage. The second layer is the priority layer, which determines the weight coefficients of three objectives—control accuracy, energy consumption cost, and control response speed—based on the scenario control priority. For example, in the industrial cleanroom scenario, the weight of control accuracy is set to 0.7, the weight of energy consumption cost is set to 0.2, and the weight of response speed is set to 0.1; in the public building scenario, the weight of energy consumption cost is set to 0.6, the weight of control accuracy is set to 0.3, and the weight of response speed is set to 0.1; and in the agricultural planting scenario, the weight of control accuracy is set to 0.5, the weight of energy consumption cost is set to 0.4, and the weight of response speed is set to 0.

1. The third layer is the optimization layer, which uses an improved particle swarm optimization algorithm to search for the optimal control strategy in the feasible solution space. The highest weighted score of the three objectives is used as the optimization objective. The final output control strategy includes the specific operating parameters, runtime, and start / stop time points of each execution device.

5. The precise control system for environmental factors in multiple scenarios according to claim 1, characterized in that, The operating mechanism of the local emergency control module includes: When the network connection to the cloud is interrupted for more than 5 minutes, or the cloud service returns to a fault state, the local emergency control mode is automatically triggered, and the basic control rules for the corresponding scenario are called in the local storage. The basic control rules are based on threshold triggering logic. When the environmental parameters exceed the preset safety threshold range, the corresponding control device is activated to bring the parameters back to the safe range. During the emergency control mode operation, the edge computing layer continuously attempts to re-establish the connection with the cloud. When the connection is restored, all collected data and control records during the network outage are automatically synchronized to the cloud. The cloud optimizes the control strategy based on the data during the network outage and distributes it to the edge layer. The system switches back to the normal control mode, and at the same time, the control effect data during the network outage is used for model iteration and update.

6. A method for precise control of environmental factors in multiple scenarios according to any one of claims 1-5, characterized in that, Includes the following steps: S1. The perception layer collects raw environmental data and device operation data in the target scene, and also supports users to enter scene attribute data. All data is uploaded to the edge computing layer through an encrypted transmission protocol. The edge computing layer cleans and preprocesses the raw data, filtering out abnormal data that obviously does not conform to physical logic. S2. The scene dynamic recognition module of the edge computing layer extracts the core features of the preprocessed data and matches them with the scene feature library stored locally. It dynamically identifies the current scene type and the corresponding control priority rules. If the matching degree is insufficient, it guides the user to supplement scene information and update the scene feature library. S3. Check the data integrity of all physical sensors. If there is missing or abnormal data, call the virtual sensing completion module to fuse the physical mechanism model and the data-driven model to generate the completed environmental parameter estimates and form a complete environmental parameter dataset. S4. The edge computing layer uploads the complete environmental parameter dataset, scene control requirements, and equipment operation status data to the cloud service layer. The cloud-based multi-objective collaborative optimization module combines real-time energy consumption prices and external weather forecast data to generate the global optimal control strategy and distribute it to the edge computing layer. S5, the edge computing layer breaks down the control strategy into specific control commands for each execution device, sends them to the execution layer devices for execution, and continuously collects environmental parameter feedback data after control, and synchronizes the control effect data to the model iteration update module in the cloud to optimize the accuracy of the virtual sensing model and the collaborative optimization model.

7. The method for precise control of environmental factors in multiple scenarios according to claim 6, characterized in that, In step S2, the specific process of scene dynamic recognition includes: First, feature extraction is performed on the preprocessed multi-dimensional data. The extracted features include three categories of 12 features: the daily fluctuation range of each environmental factor, the Pearson correlation coefficient between different factors, and the distribution pattern of equipment running time. The 12 features are combined into the feature vector of the current scene. All features are standardized to eliminate the influence of dimensions. Calculate the cosine similarity between the current feature vector and all stored scene feature vectors in the scene feature library. Select the top 3 candidate scenes with the highest similarity. If the highest similarity is greater than 0.85, it is directly determined as the corresponding scene type. At the same time, the control priority and parameter threshold rules corresponding to the scene are called. If the highest similarity is between 0.6 and 0.85, the 3 candidate scenes are displayed to the user for the user to confirm the final scene type. If the highest similarity is less than 0.6, the new scene entry process is triggered, guiding the user to fill in the scene type, core control target, and parameter threshold range information, and the new scene features are stored in the scene feature library.

8. The method for precise control of environmental factors in multiple scenarios according to claim 6, characterized in that, In step S3, the specific process of virtual sensing completion includes: First, for the currently identified scene type, the corresponding physical mechanism model is called. The physical mechanism model is built based on the principles of mass conservation and energy conservation. The input is the normally collected environmental parameters and equipment operating parameters, and the output is the first estimated value of the missing or abnormal parameters. At the same time, the lightweight LSTM data-driven model corresponding to the scene is called. The input is the historical time series data of the parameter and the real-time data of other related parameters, and the output is the second estimated value. The dynamic fusion weight is calculated based on the average absolute error of the two models on this parameter over the past 7 days. The fusion weight of the model is inversely proportional to its average absolute error. The two estimated values ​​are weighted and summed according to the fusion weight to obtain the final completed value. When the physical sensor resumes normal data acquisition, the error between the completed value and the actual acquired value is used as feedback data to optimize the parameters of the two models.

9. The method for precise control of environmental factors in multiple scenarios according to claim 6, characterized in that, In step S4, the specific process of multi-objective collaborative optimization includes: First, all constraints are determined based on the scenario type, including the allowable fluctuation range of each environmental parameter, the maximum power of each control device, the longest continuous operating time, and the total energy consumption limit. All candidate control strategies must meet all constraints, and infeasible solutions that do not meet the constraints are eliminated. Based on the control priority of the scenario, the weight coefficients of the three optimization objectives—control accuracy, energy consumption cost, and response speed—are determined. An improved particle swarm optimization algorithm is used to search for the optimal solution in the feasible solution space. Each particle represents a set of control strategies. The highest weighted comprehensive score of the three objectives is used as the optimization target. The number of iterations for optimization is no less than 100. The final output optimal control strategy includes the specific operating parameters and start-stop time of each execution device. The time granularity of the control strategy is no less than 15 minutes. Based on generating and executing the optimal control strategy, a control effect evaluation mechanism is also set up to dynamically verify the effect of the strategy implementation: the control effect of the day is comprehensively evaluated every 24 hours. The evaluation indicators include four dimensions: parameter compliance rate, average control error, total energy consumption, and equipment failure rate. The parameter compliance rate is the percentage of time when environmental parameters are within the allowable range. The average control error is the average absolute deviation between the actual parameters and the target parameters. The total energy consumption is the sum of the daily power consumption of all control equipment. The equipment failure rate is the percentage of time when equipment operates abnormally. The evaluation results are compared with the preset optimization targets. If the evaluation results are better than the results of the previous period, the control strategy samples of the current period are added to the model training set to optimize the generation of subsequent control strategies. If the evaluation results do not meet expectations, the weight coefficients of the multi-objective optimization module or the parameters of the virtual sensing model are automatically adjusted until the evaluation results meet the preset targets.

10. A computer-readable storage medium according to any one of claims 1-9, wherein a computer program is stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of a method for precise control of environmental factors in multiple scenarios. The storage medium includes, but is not limited to, hard disks, solid-state drives, USB flash drives, CD-ROMs, RAM, ROMs, EPROMs, flash memory, and magneto-optical storage. The storage medium also stores scene feature library files, virtual sensing model parameter files, and multi-objective optimization algorithm parameter files. When the processor calls these files to implement the corresponding method steps, it supports the operation of basic functions in offline mode. When the network is restored, it automatically synchronizes the latest model parameters and feature library data to ensure the reliability of the system in different network environments.

Images