Edge-computing-based environment parameter dynamic analysis energy-saving door and window intelligent control system

Abstract

The application discloses an environmental parameter dynamic analysis energy-saving door and window intelligent control system based on edge computing and relates to the technical field of edge computing, comprising: a dynamic analysis decision unit, which is used for synchronously calling preset building thermodynamic models, user behavior probability models and short-term prediction subunits based on the fusion perception data, performing heat load influence analysis, user behavior prediction and future environment trend prediction, and generating an energy-saving regulation strategy according to current analysis results and future prediction results, and calling a comfort constraint model for arbitration when there is a strategy conflict. Through the instruction checking and correcting unit, real-time checking and correcting are performed based on the latest environmental data and preset physical safety rules and logic consistency rules before the control instruction is issued, so that the execution risk or the wrong operation caused by environmental mutation, instruction conflict and the like is effectively prevented, and the safety and the robustness of the system are greatly enhanced.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, and in particular to an energy-saving intelligent control system for doors and windows based on dynamic analysis of environmental parameters using edge computing. Background Technology

[0002] With the deep integration of IoT technology, edge computing, and smart building concepts, the synergistic optimization of building energy conservation and indoor environmental quality has become an important direction for the development of green buildings. In traditional buildings, doors and windows, as a key part of the building envelope, directly affect the indoor thermal environment, air quality, and building energy consumption through their opening and closing behavior. Currently, people's demands for comfort, health, and personalization in their living and working environments are increasing, while global energy conservation and emission reduction goals are also placing higher demands on building operational energy efficiency. This makes the realization of intelligent, adaptive, and precise control over the opening and closing of doors and windows increasingly urgent.

[0003] Currently, automated or intelligent control technologies for doors and windows have made some progress, with the mainstream approaches mainly falling into two categories: one is automated control based on simple rules or preset scenarios, such as triggering window opening based on indoor temperature or CO2 concentration thresholds; the other is smart home systems relying on cloud computing platforms, which collect data from multiple sensors, perform comprehensive analysis in the cloud, and then issue control commands. While the former offers a direct response, its strategies are fixed and rigid, making it difficult to cope with complex and changing environmental factors and personalized user habits. The latter, while possessing stronger data processing capabilities, relies on network transmission and remote cloud servers for decision-making, exhibiting inherent limitations in real-time performance, reliability, and privacy protection, and struggling to achieve low-latency local rapid response and offline operation.

[0004] However, existing technical solutions still have several key issues that need to be addressed. First, most systems lack the ability to dynamically and collaboratively analyze and predict multi-dimensional environmental parameters (such as heat load, pollutants, and meteorological factors). Decision models are often static and isolated, unable to make real-time trade-offs and arbitrations among multiple objectives such as energy saving, comfort, and health. Second, systems generally lack the ability to self-learn user behavior patterns, making it difficult to align automatic control strategies with user habits and easily leading to conflicting user experiences. Third, the generation of control commands is mostly based on instantaneous states, lacking forward-looking utilization of short-term future environmental trends, and there is a lack of closed-loop verification and correction mechanisms based on real-time safety rules and logical consistency before commands are issued, which may lead to execution risks or policy contradictions. Finally, existing systems have not yet formed a closed-loop learning link from control result feedback to decision model parameter optimization, making it impossible for the system's long-term operating performance to continuously improve itself with data accumulation.

[0005] Therefore, it is essential to invent an energy-saving intelligent control system for doors and windows based on dynamic analysis of environmental parameters using edge computing to solve the above problems. Summary of the Invention

[0006] The purpose of this invention is to provide an energy-saving intelligent control system for doors and windows based on dynamic analysis of environmental parameters using edge computing, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an energy-saving intelligent control system for doors and windows based on dynamic analysis of environmental parameters using edge computing, comprising the following modules: Sensor network modules, deployed inside and outside buildings, are used to collect multi-dimensional environmental parameters; An edge computing node, communicatively connected to the sensor network module, comprises: The data fusion and preprocessing unit is used to receive and fuse the multi-dimensional environmental parameters to obtain spatiotemporally aligned fused sensing data; The dynamic analysis and decision-making unit is used to simultaneously call the pre-set building thermodynamics model, user behavior probability model and short-term prediction subunit based on the fused sensing data to perform heat load impact analysis, user behavior prediction and future environmental trend prediction. When there is a conflict between the current analysis results and the future prediction results, the unit calls the comfort constraint model for arbitration to generate energy-saving control strategies. The short-term prediction subunit is used to predict the change trend of key environmental parameters in the near future based on the historical fused sensing data sequence. The collaborative control instruction generation unit is used to generate control instructions containing specific action parameters based on the energy-saving regulation strategy. The instruction verification and correction unit is used to receive the control instructions and the latest fusion sensing data in real time, and perform verification based on preset physical security rules and logical consistency rules; if the verification fails, the correction mechanism is triggered according to the failure type, and the corrected control instructions are output; at the same time, the unit marks the verification failure event and related environmental data as optimization samples. The model optimization unit, connected to the instruction verification and correction unit and the dynamic analysis and decision unit, is used to receive the optimization sample and analyze the correlation between the root cause of the verification failure and the prediction deviation of the corresponding model in the dynamic analysis and decision unit. Based on this, it performs targeted updates on specific parameters of the model to form a closed-loop learning link from instruction verification to decision model optimization. The communication gateway module is used to distribute the verified or corrected control commands to the corresponding door and window actuator control modules. The door and window actuator control module is used to parse and execute the received control commands.

[0008] The technical effects and advantages of this invention are as follows: 1. This invention achieves low-latency acquisition, fusion, and real-time analysis of environmental parameters by deploying sensor networks and edge computing nodes locally. It avoids the network dependence and communication latency problems of traditional cloud control solutions, significantly improves system response speed, operational reliability, and user privacy security, and supports offline operation. 2. This invention uses a dynamic analysis decision-making unit to simultaneously call upon the building thermodynamics model, user behavior probability model, and short-term prediction subunit to comprehensively analyze the impact of heat load, user habits, and future environmental trends. When there is a conflict of strategies, a comfort constraint model is introduced for arbitration, which realizes dynamic trade-offs and intelligent decision-making among multiple objectives such as energy saving, comfort, and health, and solves the problem of poor adaptability of single rules or fixed strategies. 3. This invention uses a short-term forecasting subunit to predict the trends of key environmental parameters in the future, and enables a dynamic analysis and decision-making unit to make forward-looking judgments based on the forecast results, generating predictive strategies such as "delayed ventilation preparation". This allows for more accurate capture and utilization of favorable natural ventilation windows, further enhancing energy-saving potential and control foresight. 4. The present invention uses an instruction verification and correction unit to perform real-time verification and correction based on the latest environmental data and preset physical security rules and logical consistency rules before the control instructions are issued. This effectively prevents execution risks or erroneous operations caused by environmental changes, instruction conflicts, etc., and greatly enhances the security and robustness of the system. 5. This invention analyzes the correlation between failed instruction verification samples and prediction deviations of the decision model through a model optimization unit, and updates the empirical coefficients of the building thermodynamic model and the parameters of the user behavior probability model in a targeted manner, thus constructing a closed-loop learning link from execution feedback to model optimization. This enables the system to continuously evolve itself, adapt to specific building characteristics and user habits, and improve long-term control performance. 6. When the high-rise strategy is transformed into specific action parameters by the collaborative control command generation unit, the present invention constructs an optimization problem with the goal of minimizing the purification demand per unit time, and solves the problem by comprehensively considering multiple constraints such as safe opening degree and pollutant concentration. This achieves refined and optimized control of window opening angle and ventilation duration, maximizing ventilation efficiency and energy saving effect while ensuring safety and health. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the overall system structure of the present invention; Figure 2 This is a flowchart illustrating the data processing and decision-making process of the present invention. Figure 3 This is a flowchart of the instruction generation, verification, and correction process for this invention. Figure 4 The model optimization closed-loop learning link diagram of this invention; Figure 5 This is a diagram of the verification failure case library and parameter adjustment mechanism of the present invention. Detailed Implementation

[0010] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] This invention provides, for example Figure 1 The energy-saving intelligent control system for doors and windows based on dynamic analysis of environmental parameters using edge computing, as shown, includes the following modules: Sensor network modules, deployed inside and outside buildings, are used to collect multi-dimensional environmental parameters; Furthermore, in the above technical solution, the multidimensional environmental parameters include: indoor temperature, indoor humidity, indoor carbon dioxide concentration, indoor PM2.5 concentration, outdoor temperature, outdoor humidity, outdoor wind speed, outdoor wind direction, outdoor light intensity, outdoor rainfall status, and outdoor PM2.5 concentration.

[0012] It is important to understand that the sensor network module consists of multiple sensor nodes deployed both inside and outside the building, and a convergence node. The sensor nodes, depending on their monitoring parameter types, specifically include: indoor temperature and humidity sensors, indoor carbon dioxide concentration sensors, indoor PM2.5 sensors, an outdoor weather station (integrating temperature, humidity, wind speed, wind direction, light intensity, and rainfall sensors), and an outdoor PM2.5 sensor. The indoor sensor nodes typically communicate wirelessly with the convergence node deployed inside the building (e.g., Zigbee, LoRa, or Wi-Fi); the outdoor weather station and PM2.5 sensor can transmit data to the convergence node via wired or wireless means. The convergence node is responsible for initial timestamp alignment and format standardization of the raw data from each sensor node, and then sends the converged multi-dimensional environmental parameter data stream to the edge computing node via a local area network or direct interface. The data acquisition frequency of the sensor network module can be configured according to system requirements, for example, set to once per second or once every 5 seconds, to ensure that the fused sensing data obtained by the edge computing node has sufficient timeliness. The parameter "outdoor rainfall status" is provided by the rain sensor in the outdoor weather station, which can output status signals such as "no rainfall", "light rain", "moderate rain", "heavy rain" or continuous rainfall intensity data for subsequent units to make judgments.

[0013] An edge computing node, communicatively connected to the sensor network module, comprises: The data fusion and preprocessing unit is used to receive and fuse the multi-dimensional environmental parameters to obtain spatiotemporally aligned fused sensing data; It is important to understand that the data fusion and preprocessing unit is the first data processing stage in the edge computing node. This unit establishes a connection with the aggregation node of the sensor network module through a preset communication interface (such as Ethernet, RS485, or a corresponding wireless module interface) to receive the multi-dimensional environmental parameters in real time at fixed time intervals or event-triggered methods. "Fusion" refers to integrating parameters from different sensors, with different data formats and acquisition time points, into a data set with a unified time and coordinate reference. The specific process includes: time alignment, which involves assigning a unified system timestamp to all received parameter data; for data that failed to arrive in the same period due to transmission delays, interpolation or nearest neighbor methods are used to align it to a standard time point; and spatial alignment, which involves assigning a location identifier to each parameter based on the actual deployment location of each sensor in the building (e.g., the indoor or outdoor side of a window), so that subsequent analysis can be associated with specific doors and windows. The "preprocessing" includes data validity verification and cleaning: for example, checking whether the data is within a reasonable physical range, such as a temperature between -20℃ and 60℃; identifying and removing constant or abrupt abnormal values ​​caused by sensor malfunctions; and using linear interpolation or forward imputation to fill in any small amount of missing data caused by brief communication interruptions. After processing by the data fusion and preprocessing unit, the output is structured, time-series arranged fused sensing data, which can be directly used by the dynamic analysis and decision-making unit.

[0014] refer to Figure 2 The dynamic analysis and decision-making unit is used to simultaneously call the pre-set building thermodynamics model, user behavior probability model, and short-term prediction subunit based on the fused sensing data to perform heat load impact analysis, user behavior prediction, and future environmental trend prediction. Based on the current analysis results and future prediction results, it calls the comfort constraint model for arbitration when there are strategy conflicts to generate energy-saving control strategies. The short-term prediction subunit is used to predict the changing trends of key environmental parameters in the near future based on historical fused sensing data sequences. Furthermore, in the above technical solution, the execution process of the dynamic analysis and decision-making unit includes: First, based on the indoor temperature, outdoor temperature, outdoor wind speed, and outdoor light intensity from the fused sensing data, the building thermodynamic model is invoked to calculate the dynamic heat exchange load Q through doors and windows. win The calculation formula is as follows: Q win =U*A*ΔT+k s *A*Ik v *A*v*ΔT, Where U is the heat transfer coefficient of doors and windows, A is the area of ​​doors and windows, ΔT is the indoor-outdoor temperature difference, I is the outdoor light intensity, v is the outdoor wind speed, and k is the outdoor light intensity. s and k v This is an empirical coefficient; Simultaneously, based on the current fused perception data, the historical probability P of the user opening a window in the user behavior probability model is queried. hist The user behavior probability model is obtained by machine learning training based on historical fusion perception data and user window opening behavior records. Secondly, based on the first judgment rule: if Q win <-Q th (This indicates that opening windows significantly reduces the load) and P hist >P th If Q..., then an "active ventilation" strategy is generated; if Q... win Q th (Indicating that opening the window significantly increases the load), then a "keep closed" strategy is generated; where Q th P is the threshold for judging heat load. hist P represents the historical probability of a user opening a window. th This is the user behavior probability threshold. If none of the conditions in the first judgment rule are met, the arbitration step is initiated: the dynamic analysis decision unit calls its internal comfort constraint model to calculate the PMV value of the current indoor environment based on the current indoor temperature and humidity; if the PMV value exceeds the comfort zone, the predicted PMV value after opening the window for ventilation is further calculated based on the current indoor and outdoor temperature, humidity and wind speed; if the predicted PMV value falls within the preset comfort zone, a "restricted ventilation" strategy is generated; otherwise, a "close and adjust" strategy is generated.

[0015] It is important to understand that the dynamic analysis and decision-making unit is the core of the edge computing node, responsible for making intelligent decisions by integrating multiple inputs. The pre-set building thermodynamic model in this unit is a simplified calculation model based on physical formulas, where the heat transfer coefficient U of doors and windows and the area A of doors and windows are fixed parameters pre-configured according to the specific building door and window profiles and dimensions; the empirical coefficient k... s (Solar radiation heat gain coefficient) and k v The initial value of the wind-induced convective heat transfer coefficient can be obtained by consulting the building thermal engineering handbook or by fitting and calibrating the building’s historical energy consumption data with the environmental data of the same period, and can be updated by the model optimization unit in the future.

[0016] The user behavior probability model is a machine learning model, such as logistic regression, gradient boosting decision tree, or neural network. Its training data consists of long-term accumulated historical fusion sensing data (such as time, indoor and outdoor temperature and humidity, light intensity, etc.) and user manual window opening / closing behavior labels recorded synchronously. After training, the model is deployed in this unit and can output a probability value P between 0 and 1 based on the real-time input fusion sensing data. hist This indicates the likelihood that the user will actively open the window under the current environmental conditions.

[0017] Example of a specific training method for a user behavior probability model: Data Collection Phase: After system deployment, a 2-4 week "observation and learning period" begins. During this period, the system records all fused sensing data and user manual window opening / closing behaviors, forming a labeled training dataset. Each sample includes: timestamp, indoor and outdoor temperature and humidity, CO2 concentration, PM2.5 concentration, wind speed, light intensity, and other environmental parameters, as well as user behavior labels (open window = 1, closed window = 0).

[0018] Model Selection and Training: Logistic Regression or Lightweight Gradient Boosting Tree (LightGBM) will be used as the user behavior probabilistic model. Taking Logistic Regression as an example, the model form is as follows: , Where ω i These are the model parameters. The model is trained using historical data, and the optimization objective is to minimize the cross-entropy loss function.

[0019] Online update mechanism: When a logical consistency check fails, the system adds the current environment state and implicit user preferences (such as the user canceling automatic window opening) as new samples to the training set, triggering incremental learning. For example, stochastic gradient descent (SGD) is used for parameter fine-tuning, with a learning rate set to 10. -4 Each update uses only a few recent samples to avoid catastrophic forgetting. The short-term prediction subunit can be based on time series prediction algorithms, such as ARIMA and LSTM neural networks, using the fused sensing data sequence (especially outdoor temperature and wind speed) over a historical period (e.g., the past 2 hours) as input to predict the numerical trends of these key parameters over a shorter future period (e.g., the next 30 minutes). In one specific embodiment, a lightweight time series prediction method is used, with the input being the outdoor temperature and wind speed sequence (24 data points in total, one point every 5 minutes over the past 2 hours). This sequence is first standardized. The prediction model is a two-layer feedforward neural network with 16 hidden layer nodes, using the ReLU activation function. The model output is the predicted outdoor temperature and wind speed values ​​(3 time points, 6 output values) for the next 30 minutes, one point every 10 minutes. This neural network has been trained and quantized on historical datasets before deployment on edge computing nodes to adapt to the computing capabilities of the nodes. During the operation phase, a rolling prediction is triggered every decision cycle (e.g., every 5 minutes).

[0020] The comfort constraint model is based on the internationally recognized thermal comfort evaluation index PMV (Predicted Mean Vote) calculation model. It calculates the thermal sensation index by comprehensively considering factors such as air temperature, radiant temperature, air velocity, humidity, human metabolic rate, and clothing thermal resistance, according to standards such as ISO 7730. In this system, to simplify the calculation, radiant temperature is approximated as air temperature, and the standard sitting metabolic rate and typical indoor clothing thermal resistance are used as fixed parameters. Therefore, the PMV value mainly varies with the real-time collected indoor temperature, humidity, and wind speed. The "comfort zone range" is typically set to a PMV value between -0.5 and +0.5.

[0021] The initial values ​​of ks and kv can be obtained in the following ways: Consult the building thermal engineering handbook: Based on the door and window profiles, glass type, shading coefficient, etc., consult the "Building Thermal Design Code" or the product thermal performance table to obtain approximate values.

[0022] Based on historical data fitting: In the initial stage of system deployment, indoor and outdoor temperature, wind speed, light intensity, and building energy consumption data for a continuous week were collected. Multiple linear regression was used to fit the data to obtain k. s and k v The initial estimate. For example, k is obtained by fitting measured data from an office building. s ≈0.65, k v ≈0.12.

[0023] The execution logic of this unit is concurrent: it synchronously calls the building thermodynamic model, queries the user behavior probability model, and runs the short-term prediction subunit to obtain the current heat load Q. winUser window opening probability P hist And future environmental trends. Then, a preliminary decision is made according to the aforementioned "first judgment rule".

[0024] Q th The determination criteria are as follows: This threshold should be determined comprehensively based on the building type, the thermal characteristics of the building envelope, and the design load of the air conditioning system. A typical value can be obtained by simulating the heat transfer distribution per unit area through doors and windows under typical meteorological conditions using building energy simulation software (such as EnergyPlus or TRNSYS), and taking the 75th percentile as the initial threshold. For example, for a typical office building, the area of ​​a single window is approximately 2m². 2 The heat transfer coefficient U ≈ 2.5 W / (m²) 2 When K), Q th It can be initially set to 100W. In practical applications, it can be fine-tuned based on the building's actual energy consumption data and user feedback.

[0025] P th The determination of this threshold is based on the habitual intensity of users' window-opening behavior, and can be set according to the frequency distribution of users' window-opening behavior in historical data. A typical value can be calculated based on the proportion of user window-opening times during the same time period (e.g., weekdays 9:00-18:00) to the total number of observations within the past 30 days. For example, if this proportion exceeds 60%, then the initial setting is P. th =0.6. After the system is running, it can be dynamically adjusted based on feedback from the model optimization unit.

[0026] When initial decisions result in conflict (e.g., Q) win Although less than -Q th But P hist Below P th , or Q win In -Q th With Q th If a clear strategy cannot be derived (in the case of conflicting conditions), the comfort constraint model is invoked for arbitration. The arbitration process first calculates the current indoor PMV value to determine comfort level. If uncomfortable, it simulates the indoor environment after opening windows for ventilation (incorporating current outdoor temperature, humidity, and wind speed) and calculates a predicted PMV value. If the predicted PMV value falls within the comfort zone, ventilation improves comfort, thus generating a "restricted ventilation" strategy. Otherwise, current outdoor conditions do not improve or may even worsen indoor comfort, and a "close and adjust" strategy should be generated. This strategy can be interpreted by the subsequent system as requiring the activation of backup control equipment such as air conditioning and fresh air systems. Ultimately, this unit outputs a clear energy-saving control strategy, such as "active ventilation," "keep closed," "restricted ventilation," "delayed ventilation preparation," or "close and adjust," which is then passed to the collaborative control command generation unit.

[0027] Furthermore, in the above technical solution, the short-term prediction subunit is used to predict the temporal change trend of outdoor temperature and outdoor wind speed within a first preset time period in the future. When applying the first judgment rule, the dynamic analysis and decision-making unit makes a forward-looking judgment based on the output of the short-term prediction subunit: when based on the current Q... win When determining that a "keep off" strategy should be generated, if the predicted heat load Q for the second preset time period is calculated based on the predicted future temperature and wind speed... future Satisfying Q future <-Q th The dynamic analysis and decision-making unit then generates a "delayed ventilation preparation" strategy and generates predictive triggering conditions bound to this strategy; the predictive triggering conditions are sent to the collaborative control command generation unit; wherein, Q future To predict future heat load, Q th The threshold for determining heat load; After receiving a "delayed ventilation preparation" strategy bound with predictive trigger conditions, the collaborative control command generation unit monitors real-time fused sensing data. When it determines that the predictive trigger conditions are met, it performs an optimized control command generation operation based on the real-time environmental parameters at the time of determination to generate and output real-time control commands. The optimized control command generation operation includes: using window opening angle and ventilation duration as decision variables, constructing and solving an optimization problem with minimizing the comprehensive purification demand per unit time as the objective function. The objective function is related to the current indoor and outdoor pollutant concentration deviation, and the solution process must satisfy the safe opening degree constraint related to real-time wind speed and wind direction, as well as the outdoor pollutant safe concentration constraint.

[0028] It is important to understand that the short-term prediction subunit is crucial for achieving proactive energy-saving control. This subunit is typically based on time-series prediction algorithms, such as the Autoregressive Integral Moving Average (ARIMA) model or Long Short-Term Memory (LSTM) network, and is deployed within the edge computing node. Its input is the historical fused sensing data sequence, arranged chronologically, output by the data fusion and preprocessing unit, particularly continuous outdoor temperature and wind speed data. The sequence length (i.e., the historical window) is, for example, the past 2 hours, and the data sampling interval is consistent with the system's acquisition frequency. The "first preset time period in the future" refers to the time span of the prediction output, for example, the next 30 minutes. This subunit operates in a rolling prediction manner. In each decision cycle, based on the latest historical data sequence, it predicts the expected outdoor temperature and wind speed values ​​for a series of equally spaced time points (e.g., one point every 5 minutes) within the first preset time period in the future, and outputs them in the form of time-series data.

[0029] After obtaining the forecast time series data output by the short-term forecast subunit, the dynamic analysis and decision-making unit makes a forward-looking judgment. Specifically, it selects a "second preset time period in the future" (for example, the future interval from the 10th minute to the 25th minute from the current moment), and calculates the predicted outdoor temperature and wind speed values ​​for each forecast time point within this time period, along with the current indoor temperature (assuming it does not change much in the short term) and the inherent parameters of doors and windows (U, A, k). s k v Substituting these values ​​into the calculation formula of the building thermodynamic model, a series of future predicted heat load values ​​Q are calculated. future Then, determine whether there are multiple consecutive time points Q within this future time period. future All values ​​satisfy Q future <-Q th (That is, the forecast shows that opening windows in the future will significantly reduce the heat load). If this condition is met, it means that although the current Q... win Q th This suggests that the facility should "remain closed," but a favorable window of opportunity for ventilation is anticipated.

[0030] At this point, the dynamic analysis and decision-making unit will generate a "delayed ventilation preparation" strategy. Simultaneously, it needs to generate a clear "predictive trigger condition" that can be monitored and judged in real-time. The core of this "predictive trigger condition" is determining the real-time heat load Q. win Has the threshold condition for initiating ventilation been met? This is to be determined in conjunction with the judgment logic (Q). win <-Q th To maintain consistency and achieve a smooth transition from "preparation" to "execution," this condition should be set to approach -Q. th The critical value. For example, it can be set as "when the real-time calculated heat load Q..." win ≤-Q th "When" indicates that the favorable conditions for prediction have been fully met; or, for the sake of early response, it is set to "when Q". win ≤-0.9*Q th The specific threshold is a routine optimization performed by those skilled in the art based on system response delay and prediction uncertainty. After generating the strategy and triggering conditions, they are sent together to the cooperative control command generation unit.

[0031] Upon receiving a "delayed ventilation preparation" strategy bound with predictive triggering conditions, the collaborative control command generation unit does not immediately generate control commands but instead enters a monitoring standby state. It continuously receives the latest fused sensing data from the data fusion and preprocessing unit and uses parameters such as real-time outdoor temperature, wind speed, and indoor temperature to calculate the real-time Q using the same building thermodynamic model. win Value. It will give this real-time Q.win The value is compared with the received "predicted trigger condition". Once it is determined that the trigger condition is met, for example, real-time Q... win The value is ≤ -0.9*Q for the first time. th The unit immediately exits the monitoring state and, based on the latest and most accurate fused sensing data at that moment, executes the "optimized control command generation operation".

[0032] The "optimized control command generation operation" and the operation of the collaborative control command generation unit are mathematically consistent, but their objective function focuses more on "comprehensive purification needs." To handle the weighted summation of parameters with different dimensions (carbon dioxide concentration, PM2.5 concentration), parameter normalization is required in actual calculations. Specifically, the objective function is actually calculated as: [α*((C in -C target ) / N c )+β*((P in -P target ) / N p )] / t. Where, N c and N p These are normalized baseline values ​​for carbon dioxide concentration and PM2.5 concentration, respectively. For example, N c Choose C target Value, N p Take P target The value is used to convert the concentration deviation into a dimensionless relative quantity. The weighting coefficients α and β reflect the priority of different pollutant purification needs. The constraints to solve this optimization problem include: a dynamic safety opening θ obtained by looking up a table or calculating based on the real-time wind speed v and wind direction D at the trigger time. safe (v, D), and real-time outdoor PM2.5 concentration C out It must be below the safety threshold C safe By solving this constrained optimization problem, the optimal θ* and t* are obtained, and then specific control commands containing these two parameters are generated and output.

[0033] refer to Figure 3 : Cooperative control instruction generation unit, used to generate control instructions containing specific action parameters based on the energy-saving control strategy; Furthermore, in the above technical solution, when the collaborative control command generation unit is executed, if the strategy is "active ventilation" or "restricted ventilation," it performs the following operations to generate control commands: Using the window opening angle θ and ventilation duration t as decision variables, an optimization problem is constructed with the objective function of minimizing the purification demand per unit time. The objective function is [α*(C)] in -Ctarget ) / N c +β*(P in -P target ) / N p ] / t, where C in P represents the indoor carbon dioxide concentration. in For indoor PM2.5 concentration, C target and P target These are their target concentration values, and α and β are preset positive weighting coefficients; The constraints that the optimization problem needs to satisfy include: θ ≤ θ safe (v, D), where θ safe (v, D) is a dynamic safety opening function, which determines the maximum permissible safe window opening angle based on the outdoor wind speed v and outdoor wind direction D; and the outdoor PM2.5 concentration C. out C must be satisfied out ≤C safe C safe The preset safe concentration threshold; By solving this optimization problem, the optimal window opening angle θ and ventilation duration t are obtained, and control commands containing the parameters θ and t are generated.

[0034] It is important to understand that the collaborative control command generation unit is a crucial step in transforming abstract strategies into executable actions. When it receives an "active ventilation" or "restricted ventilation" strategy from the dynamic analysis and decision-making unit, this unit immediately initiates an optimization calculation process based on the latest fused sensing data to generate precise control commands.

[0035] Regarding the construction of the optimization problem: The objective function aims to quantify the "comprehensive purification requirement" per unit ventilation time. To ensure that the parameters with different physical dimensions (carbon dioxide concentration and PM2.5 concentration) in the objective function can be effectively weighted and summed, the objective function is normalized in the actual calculation, specifically as follows: [α*((C in -C target ) / N c )+β*((P in -P target ) / N p )] / t, Among them, C target and P target To set a target concentration value based on indoor air quality standards (e.g., C). target It can be set to 1000ppm, P target It can be set to 35μg / m 3 N c and Np are the corresponding normalized reference values (for example, N c =C target , N p =P target ), which are used to convert the concentration deviation into a dimensionless relative ratio. The weight coefficients α and β are positive numbers, which are used to balance the relative importance of reducing the carbon dioxide concentration and reducing the PM2.5 concentration in the system optimization goal, and their specific values can be determined through simulation or on-site debugging. For example, the initial values can be set as α = 1.0 and β = 0.7.

[0036] Regarding the constraint conditions: The dynamic safety opening function θ safe (v, D): This is the core constraint to ensure the physical safety of the window under various wind conditions. The specific corresponding relationship of this function needs to be pre-calibrated for the profile, size and mechanical strength of each window. A typical implementation method is to establish a look-up table based on the wind speed v and the wind direction D. For example, for a certain window: when the wind speed v ≤ 5m / s, θ safe = 90° (fully open); when 5m / s < v ≤ 10m / s, if the wind direction D is head-on wind, then θ safe = 30°, if it is side wind, then θ safe = 45°; when v > 10m / s, θ safe = 0° (opening prohibited).

[0037] Outdoor pollutant safety constraint: The outdoor PM2.5 concentration C out needs to satisfy C out ≤C safe , where C safe is the preset safety concentration threshold, such as 75μg / m 3 . If this condition is not satisfied, it is determined that the current outdoor air quality is not suitable for ventilation, and there is no feasible solution for this optimization.

[0038] Regarding the solution of the optimization problem: This is an optimization problem with boundary constraints. In the embedded edge computing environment, to meet the real-time requirements, the discretized grid search method can be used for efficient solution. Specifically: The decision variable window opening angle θ is discretized at a fixed interval (such as 5°) within its domain [0, θ safe (v, D)]; the ventilation duration t is discretized at a fixed interval (such as 60 seconds) within the preset reasonable range [t min , t max (for example, t min = 300 seconds, t max = 1800 seconds). Traverse all discrete point combinations (θ, t), and for each combination, check whether C out ≤C safeThe constraints are defined as follows: Among all combinations that satisfy the constraints, the value of the normalized objective function is calculated, and the combination (θ*, t*) that minimizes the objective function value is selected as the optimal solution. If no combination satisfies the constraints, no windowing command is generated.

[0039] Regarding the generation of control commands: After obtaining the optimal solution (θ*, t*), the unit generates a structured control command. This command contains at least the following fields: the identifier (ID) of the target door / window, the action command (open window), the target angle θ*, and the expected holding time t*. For example: {"window_id": "W_01", "command": "OPEN", "angle": 45", "duration": 600}. This command is then sent to the command verification and correction unit for final security and logical consistency verification.

[0040] The instruction verification and correction unit is used to receive the control instructions and the latest fusion sensing data in real time, and perform verification based on preset physical security rules and logical consistency rules; if the verification fails, the correction mechanism is triggered according to the failure type, and the corrected control instructions are output; at the same time, the unit marks the verification failure event and related environmental data as optimization samples. Furthermore, in the above technical solution, the instruction verification and correction unit includes: The physical safety rules include: whether the window opening angle in the instruction exceeds the dynamic safe maximum opening degree determined based on the characteristics of the currently controlled doors and windows and the real-time wind speed and direction; and, when the outdoor rainfall status in the fused sensing data is "raining", the execution of the window opening instruction is prohibited or restricted. The logical consistency rule includes: whether there is a mutually exclusive instruction sequence for the same door or window within an overlapping time window; The correction mechanism includes: for security rule failures, adjusting parameters to security boundary values ​​or canceling instructions; for consistency rule failures, selecting instructions based on policy priority and timeliness.

[0041] It is important to understand that the instruction verification and correction unit is the last line of security and logical defense before control instructions are issued. This unit monitors the control instruction stream from the collaborative control instruction generation unit in real time and simultaneously acquires the latest round of fused sensing data from the data fusion and preprocessing unit.

[0042] Regarding the implementation of physical security rules: Window opening angle safety verification: This unit parses the target window opening angle θ from the command to be verified. inst At the same time, the current outdoor wind speed v is obtained from the latest fusion sensing data. now Wind direction Dnow It queries or calculates the current availability of that window type based on the model number of the target door / window. now D now Dynamic safety maximum opening θ under the condition safe，now This θ safe，now The determination method and the θ used in the cooperative control command generation unit safe The (v, D) function is consistent. The verification rule is: if θ inst ≤θ safe，now If it passes, then it is approved; otherwise, the security rule is considered to have failed.

[0043] Rainfall Status Safety Verification: This unit reads the "Outdoor Rainfall Status" parameter from the latest fused sensing data. If the status is clearly "Raining" (e.g., determined by a rain sensor to be continuous precipitation), the verification rule is: prohibit any window opening command (i.e., regardless of the command angle, it is considered a failure). In some implementations, it can also be set to allow small-angle window opening in "light rain" conditions, while completely prohibiting it in "moderate rain" or "heavy rain" conditions.

[0044] Regarding the implementation of logical consistency rules: This unit maintains a cache queue of recently sent instructions. When a new instruction is received, it checks the target door / window ID of the instruction and searches the cache queue for a previously sent but incomplete instruction targeting the same door / window with overlapping execution time windows. For example, one instruction requires "open the window at 10:00:00 for 600 seconds," while another instruction targeting the same window requires "close the window at 10:05:00." Since the execution time windows of these two instructions ([10:00:00, 10:10:00] and [10:05:00, 10:05:00 + window closing action time]) overlap and their actions are mutually exclusive ("open window" and "close window"), the logical consistency check is considered to have failed.

[0045] Regarding the triggering and execution of the correction mechanism: For failures related to physical safety rules: a "correction mechanism" is triggered. For example, for a failure due to "window opening angle exceeding limits," the correction logic is: adjust the window opening angle θ in the instruction. inst Corrected to the current maximum allowed safe opening θ safe，now If the corrected angle θ safe，now If the value is 0° (i.e., not allowed to be turned on), or the reason for failure is "raining", the correction result is to cancel the instruction and not distribute it further.

[0046] When logical consistency rules fail, a "correction mechanism" is triggered. The correction logic prioritizes and adjusts based on preset "policy priority" and "instruction timeliness." Generally, policies like "active ventilation" and "restricted ventilation" have higher priority than "keep closed"; newer instructions, those closer to the current time, have higher timeliness priority. The correction may result in canceling the new instruction, canceling the old instruction from the cache, or, in rare reconcilable cases, modifying instruction parameters (such as postponing the start time of the new instruction) to avoid conflicts. Ultimately, only the retained or corrected instructions are output.

[0047] Regarding the labeling of optimized samples: Whenever a verification fails (regardless of whether the correction passes), this unit creates an "optimized sample" record. This record contains at least the following information: (1) Failure timestamp; (2) Content of the original control instructions; (3) The latest fusion-sensing data snapshot when the trigger fails; (4) Failure types, such as "safety angle exceeded", "rainfall prohibited", "command sequence conflict"; (5) Correction operations performed, such as "adjust angle to X degrees" or "cancel command".

[0048] These labeled optimization samples will be temporarily stored and sent to the model optimization unit periodically or in batches for subsequent analysis and learning.

[0049] refer to Figure 4 The model optimization unit, connected to the instruction verification and correction unit and the dynamic analysis and decision unit, is used to receive the optimization sample and analyze the correlation between the root cause of the verification failure and the prediction deviation of the corresponding model in the dynamic analysis and decision unit. Based on this, it performs targeted updates on specific parameters of the model to form a closed-loop learning link from instruction verification to decision model optimization. Furthermore, in the above technical solution, the targeted update process of the model optimization unit is specifically as follows: When the optimized sample indication fails due to "environmental mutation verification", the model optimization unit extracts environmental data before and after the mutation and calculates the effect of the building thermodynamic model on load Q in the dynamic analysis decision unit. win The prediction error is used to adjust the empirical coefficient k in the model in reverse. s or k v ; When the optimized sample indication fails due to "logical consistency check", the model optimization unit analyzes the user behavior probability P corresponding to the conflicting instruction. hist And fine-tune the parameters of the user behavior probability model.

[0050] It is important to understand that the model optimization unit is the core of achieving system adaptation and continuous optimization. It constructs a closed-loop learning chain from the feedback of instruction execution results to the parameters of the decision model. This unit typically runs offline or as a low-priority background task, and initiates an analysis and update process periodically (e.g., every 24 hours) or after accumulating a certain number of optimization samples.

[0051] Analysis and updates regarding the failure of "Environmental Mutation Validation": The failure of the "environmental mutation verification" specifically refers to a situation where environmental parameters (mainly wind speed and outdoor temperature) undergo a short-term, drastic change after the command is generated but before execution, causing the original command to violate physical safety rules (such as exceeding angle limits). Upon receiving such an optimization sample, the model optimization unit performs the following steps: Data extraction: Extract environmental data "before the mutation" from the samples (i.e., fused sensing data at instruction generation time t0, including wind speed v0 and outdoor temperature T). out0 Indoor temperature T in0 Illumination I0) and the data after the "mutation" (i.e., the real-time data v1 at the time of verification failure t1, T out1 T in1 ,I1).

[0052] Error calculation: Using the aforementioned building thermodynamic model, calculate the predicted load Q at time t0. win，pred0 and the predicted load Q at time t1 win，pred1 Meanwhile, based on the actual environmental data at time t1, a more accurate reference value for the actual heat load Q can be estimated. win，actual1 (This estimate may involve simplified instantaneous energy balance calculations, or assume that T in the short term...) in The changes are minor; the main differences lie in v and T. out (Caused by a sudden change). The prediction error E of the model at time t1 can be characterized as E=Q. win，pred1 -Q win，actual1 .

[0053] Parameter-oriented updates: the main source of analysis error E. If the abrupt change is mainly caused by wind speed v, then the error may be mainly related to the term -k which includes wind speed. v *A*v*ΔT is related, and at this point, the gradient descent approach is used, adjusting the empirical coefficient k along the direction that reduces the error E. v Make fine adjustments: k v，new =k v，old -η k *(∂E / ∂k v ), where η kThe learning rate is initially set to 0.001 and decays with each update. To avoid model oscillation caused by frequent updates, updates are limited to once per day, and only performed when the accumulated error exceeds a set threshold (e.g., three consecutive failures of the same type). Similarly, if the mutation is mainly caused by changes in solar radiation (illuminance I), then k is adjusted. s This adjustment makes the model more accurate in predicting loads for similar environmental upheavals in the future, thus allowing the generated instructions to have a greater safety margin or be more predictable.

[0054] Analysis and updates regarding the failure of "logical consistency check": Such failures indicate that the P predicted by the user behavior probability model is... hist The (user's window-opening tendency) deviated significantly from the actual situation, causing the system to generate instructions that conflicted with the user's potential intentions. The processing flow of the model optimization unit is as follows: Conflict scenario analysis: Extract the instructions that cause the conflict and their corresponding environmental context (i.e., the environmental data at the time the instruction is generated) from the samples, as well as the operations that the user may have expected when the instruction is overridden or canceled (implied in the conflict).

[0055] Probability bias correction: This unit inputs environmental data into the current user behavior probability model to obtain the predicted probability P given by the model. model If a conflict indicates that the system command is overly aggressive—for example, the system determines "active ventilation" when the user typically does not intervene—then it suggests that P... model The target probability may be overestimated in this context; conversely, it may be underestimated in the context of a less specific target. The unit generates a "corrected target probability" P. target For example, by appropriately adjusting towards 0 or 1, and using this as a target, the core parameters (such as neural network weights or logistic regression coefficients) of the user behavior probability model can be fine-tuned using retraining with a small number of samples or online learning algorithms (such as incremental updates of logistic regression). This would make the model output a probability closer to P when given similar input contexts. target .

[0056] Examples of validation failure case libraries and proactive parameter adjustments: Case library structure: Each case record includes fields such as failure type, environment parameter snapshot, decision strategy, verification failure reason, and correction operation.

[0057] Weight accumulation mechanism: Similarity judgment of similar cases is based on key parameter range matching (such as temperature ±2℃, wind speed ±1m / s). For each new similar case, its weight accumulation value increases by 1. If a case causes the instruction to be canceled, an additional 1 is added.

[0058] Threshold Triggering and Parameter Adjustment Example: Suppose the cumulative weight of the "sudden wind speed change causing safe opening to exceed the limit" type of case reaches 15 (threshold), the system triggers an adjustment to Q. th The adjustment strategy is to adjust Q. th Increasing the absolute value by 10% (e.g., from 1 million to 1.1 million) makes the decision-making more conservative, reserving a larger safety margin. After the adjustment, the weight of this type of case is reset to 5, and it enters the observation period. Regarding the update activation mechanism: After completing the parameter update, the model optimization unit will update the coefficients (k) s k v The updated user behavior probability model file is securely synchronized to the corresponding model storage location in the dynamic analysis and decision-making unit. At the start of the next decision cycle, the dynamic analysis and decision-making unit will automatically load and use the updated model parameters, thereby achieving iterative optimization of decision-making capabilities. The entire closed-loop learning process requires no manual intervention, realizing the system's self-evolution.

[0059] refer to Figure 5 Furthermore, in the above technical solution, the model optimization unit maintains a database of failed verification cases and accumulates weights for recurring similar failed cases; when the accumulated weight of a certain type of case exceeds a threshold, it triggers the threshold (Q) of the corresponding decision rule in the dynamic analysis and decision unit. th P th Or the active adjustment of the optimization problem weight parameters (α, β) in the collaborative control instruction generation unit.

[0060] It's important to know that the "verification failure case library" maintained in the model optimization unit is a structured database used to systematically track and analyze recurring control problems, and to optimize system-level parameters accordingly. Its workflow is as follows: Identification and clustering of similar failure cases: Whenever a new optimization sample is generated, the unit first extracts key features to form a feature vector. Features include: failure type (e.g., "safety angle exceeded - sudden wind speed change", "logic conflict - user behavior deviation"), key environmental parameter ranges (e.g., wind speed range at the time of the incident: 5~8m / s, outdoor temperature range: 20~25℃), and the decision-making process involved (e.g., which decision rule was triggered). The unit performs clustering by calculating the similarity between the new case feature vector and the features of existing cases in the case library (e.g., based on Euclidean distance or key field matching). If the similarity with an existing case class exceeds a preset threshold, it is determined to be a "similar failure case" and assigned to that class; otherwise, a new case class is created starting from it.

[0061] Weight Accumulation and Threshold Triggering: Each case class has an associated "cumulative weight" value. When a new case is assigned to an existing class, the cumulative weight of that class increases. The weight increment is not fixed at 1, but is weighted according to the severity of the failure. For example, a failure that causes the instruction to be completely canceled may have a greater weight increment (e.g., +2) than a failure that only causes parameter fine-tuning (e.g., +1). Each case class also has a "trigger threshold," which can be preset according to the importance of the parameter (e.g., for the core threshold Q). th The trigger threshold for adjustment may be set to 10, and the threshold for adjusting the weight parameter α may be set to 15. When the cumulative weight of a certain case class exceeds its corresponding trigger threshold, it means that the problem of this type has occurred so frequently that it is necessary to adjust the underlying parameters of the system, and the unit will then trigger the "active adjustment" process.

[0062] Active parameter adjustment strategy: For decision threshold (Q) th P th Adjustment of ) : For example, if the case class of cumulative weights indicates that the system is under "moderate negative heat load (Q) win Slightly less than 0) and medium user probability (P) hist In scenarios with a resolution of approximately 0.5, frequent "logical consistency" conflicts occur. Analysis indicates that this is due to the current Q... th and P th The combination is too aggressive. A possible adjustment strategy is to: [change Q] th The absolute value is slightly increased, for example, from -100W to -120W, making the triggering conditions for the "active ventilation" strategy more stringent; or P th Slightly increasing the value, for example from 0.6 to 0.65, makes the decision more responsive to user habits. Adjustments are typically made in small, incremental steps, such as adjusting the current value by 5% to 10% each time.

[0063] Adjustment of the optimization weight parameters (α, β): For example, if the case study shows that the system frequently experiences fluctuations in outdoor PM2.5 concentration due to C... out >C safe The cancellation of the instruction was found to be due to an insufficient weighting of β (corresponding to PM2.5) in the objective function, resulting in the optimization process not adequately "avoiding" the risk of high outdoor PM2.5. A possible adjustment strategy is to appropriately increase the relative value of β (e.g., adjusting α:β from 1.0:0.7 to 1.0:0.9), so that the generated instructions, under the same conditions, are more inclined to choose times with lower outdoor PM2.5 or smaller window opening angles. The adjustment must maintain an overall balance of weights, avoiding excessive bias towards any one indicator.

[0064] Adjustment Implementation and Verification: After parameter adjustment, the unit updates the configuration of the dynamic analysis and decision-making unit or the collaborative control instruction generation unit with the new parameter values. Simultaneously, the accumulated weight of this case class is reset to a base value (not cleared to zero) and enters an "observation period." During the observation period, if similar cases continue to accumulate weight rapidly, it may indicate an incorrect adjustment direction or insufficient magnitude. The unit will then perform a reverse or more forceful adjustment upon the next trigger. This mechanism ensures the robustness of parameter adjustment and the system's adaptive convergence capability.

[0065] The communication gateway module is used to distribute the verified or corrected control commands to the corresponding door and window actuator control modules. This module acts as a communication bridge between the edge computing nodes and the actuators distributed at each door and window. Its core functions are protocol conversion and command routing. It connects to the command verification and correction unit via an internal network (such as Ethernet, RS485 bus, or CAN bus) to receive control commands that have passed verification or have been corrected. The control commands are structured data messages, such as JSON or a custom binary format. The communication gateway module maintains a door and window address mapping table, which records the unique logical or physical address of each door and window actuator control module in the network. When the gateway receives a command, it first parses the "target door / window ID" field in the command, then queries the address mapping table to find the corresponding target address. Next, according to the communication protocol supported by the target actuator control module, such as Zigbee, LoRa, Bluetooth, or RS485-based Modbus RTU, it encapsulates the received command data into a corresponding downlink communication frame and sends it out through the corresponding physical communication interface. This module is also responsible for managing the connection status of the communication link and briefly buffering and retrying commands in case of network anomalies.

[0066] The door and window actuator control module is used to parse and execute the received control commands.

[0067] This module is an embedded hardware unit installed on each controlled door / window, typically including a microcontroller, motor drive circuitry, position sensors, and a communication interface. Its workflow is as follows: Command Reception and Parsing: The module receives downlink data frames from the communication gateway module through its communication interface. After verification, the data frames are passed to the microcontroller for parsing. The microcontroller extracts the valid control command portion from the data frame, identifying key parameters such as "action command" (open / close window), "target angle θ", and "expected holding time t".

[0068] Instruction execution: The microcontroller calculates the direction and number of pulses required for the motor to rotate based on the target angle θ* and the current window angle (obtained via an angle encoder or potentiometer) either built-in or read in real time. Subsequently, the motor drive circuit (such as an H-bridge circuit) drives the window opening motor (usually a DC motor or a stepper motor) to rotate, thereby moving the window opening and closing mechanism until the target angle is reached.

[0069] Status Feedback and Maintenance: After the window moves into position, the position sensor sends a confirmation signal to the microcontroller. The module can (optionally) send a status report containing the "window ID" and "current angle" to the communication gateway module for system monitoring. After reaching the "expected holding time t*", if no new instructions are received, the module can perform subsequent operations according to the preset default behavior (such as automatic closing) or the next received strategy instruction.

[0070] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An energy-saving intelligent control system for doors and windows based on dynamic analysis of environmental parameters using edge computing, characterized in that: Includes the following modules: Sensor network modules, deployed inside and outside buildings, are used to collect multi-dimensional environmental parameters; An edge computing node, communicatively connected to the sensor network module, comprises: The data fusion and preprocessing unit is used to receive and fuse the multi-dimensional environmental parameters to obtain spatiotemporally aligned fused sensing data; The dynamic analysis and decision-making unit is used to simultaneously call the pre-set building thermodynamics model, user behavior probability model and short-term prediction subunit based on the fused sensing data to perform heat load impact analysis, user behavior prediction and future environmental trend prediction. When there is a conflict between the current analysis results and the future prediction results, the unit calls the comfort constraint model for arbitration to generate energy-saving control strategies. The short-term prediction subunit is used to predict the change trend of key environmental parameters in the near future based on the historical fused sensing data sequence. The collaborative control instruction generation unit is used to generate control instructions containing specific action parameters based on the energy-saving regulation strategy. The instruction verification and correction unit is used to receive the control instructions and the latest fusion sensing data in real time, and perform verification based on preset physical security rules and logical consistency rules; if the verification fails, the correction mechanism is triggered according to the failure type, and the corrected control instructions are output; at the same time, the unit marks the verification failure event and related environmental data as optimization samples. The model optimization unit, connected to the instruction verification and correction unit and the dynamic analysis and decision unit, is used to receive the optimization sample and analyze the correlation between the root cause of the verification failure and the prediction deviation of the corresponding model in the dynamic analysis and decision unit. Based on this, it performs targeted updates on specific parameters of the model to form a closed-loop learning link from instruction verification to decision model optimization. The communication gateway module is used to distribute the verified or corrected control commands to the corresponding door and window actuator control modules. The door and window actuator control module is used to parse and execute the received control commands.

2. The energy-saving intelligent control system for doors and windows based on dynamic analysis of environmental parameters using edge computing as described in claim 1, characterized in that, The multidimensional environmental parameters include: indoor temperature, indoor humidity, indoor carbon dioxide concentration, indoor PM2.5 concentration, outdoor temperature, outdoor humidity, outdoor wind speed, outdoor wind direction, outdoor light intensity, outdoor rainfall status, and outdoor PM2.5 concentration.

3. The energy-saving intelligent control system for doors and windows based on dynamic analysis of environmental parameters using edge computing as described in claim 1, characterized in that, The execution process of the dynamic analysis and decision-making unit includes: First, based on the indoor temperature, outdoor temperature, outdoor wind speed, and outdoor light intensity from the fused sensing data, the building thermodynamic model is invoked to calculate the dynamic heat exchange load Q through doors and windows. win The calculation formula is as follows: Q win =U*A*ΔT+k s *A*I-k v *A*v*ΔT, Where U is the heat transfer coefficient of doors and windows, A is the area of ​​doors and windows, ΔT is the indoor-outdoor temperature difference, I is the outdoor light intensity, v is the outdoor wind speed, and k is the outdoor light intensity. s and k v This is an empirical coefficient; Simultaneously, based on the current fused perception data, the historical probability P of the user opening a window in the user behavior probability model is queried. hist The user behavior probability model is obtained by machine learning training based on historical fusion perception data and user window opening behavior records. Secondly, based on the first judgment rule: if Q win <-Q th And P hist >P th If Q, then an "active ventilation" strategy is generated; if Q win Q th Then, a "keep off" policy is generated; where Q th P is the threshold for judging heat load. hist P represents the historical probability of a user opening a window. th This is the user behavior probability threshold. If none of the conditions in the first judgment rule are met, the arbitration step is initiated: the dynamic analysis decision unit calls its internal comfort constraint model to calculate the PMV value of the current indoor environment based on the current indoor temperature and humidity; if the PMV value exceeds the comfort zone, it further simulates and calculates the predicted PMV value after opening the window for ventilation based on the current indoor and outdoor temperature, humidity and wind speed; if the predicted PMV value falls within the preset comfort zone, a "restricted ventilation" strategy is generated; otherwise, a "close and adjust" strategy is generated.

4. The energy-saving intelligent control system for doors and windows based on dynamic analysis of environmental parameters using edge computing as described in claim 1, characterized in that, The short-term prediction subunit is used to predict the temporal variation trend of outdoor temperature and outdoor wind speed within a first preset time period in the future. When applying the first judgment rule, the dynamic analysis and decision-making unit makes a forward-looking judgment based on the output of the short-term prediction subunit: when based on the current Q... win When determining that a "keep off" strategy should be generated, if the predicted heat load Q for the second preset time period is calculated based on the predicted future temperature and wind speed... future Satisfying Q future <-Q th The dynamic analysis and decision-making unit then generates a "delayed ventilation preparation" strategy and generates predictive triggering conditions bound to this strategy; the predictive triggering conditions are sent to the collaborative control command generation unit; wherein, Q future To predict future heat load, Q th The threshold for determining heat load; After receiving a "delayed ventilation preparation" strategy bound with predictive trigger conditions, the collaborative control command generation unit monitors real-time fused sensing data. When it determines that the predictive trigger conditions are met, it performs an optimized control command generation operation based on the real-time environmental parameters at the time of determination to generate and output real-time control commands. The optimized control command generation operation includes: using the window opening angle and ventilation duration as decision variables, constructing and solving an optimization problem with minimizing the comprehensive purification demand per unit time as the objective function. The objective function is related to the current indoor and outdoor pollutant concentration deviation, and the solution process must satisfy the safe opening degree constraint related to real-time wind speed and wind direction, as well as the outdoor pollutant safe concentration constraint.

5. The energy-saving intelligent control system for doors and windows based on dynamic analysis of environmental parameters using edge computing as described in claim 1, characterized in that, When the collaborative control command generation unit is executed, if the strategy is "active ventilation" or "restricted ventilation", it performs the following operations to generate control commands: Using the window opening angle θ and ventilation duration t as decision variables, an optimization problem is constructed with the objective function of minimizing the purification demand per unit time. The objective function is [α*(C)] in -C target ) / N c +β*(P in -P target ) / N p ] / t, where C in P represents the indoor carbon dioxide concentration. in For indoor PM2.5 concentration, C target and P target These represent their target concentration values, α and β are preset positive weighting coefficients, and N... c and N p These are the corresponding normalized baseline values; The constraints that the optimization problem needs to satisfy include: θ ≤ θ safe (v, D), where θ safe (v, D) is a dynamic safety opening function, which determines the maximum permissible safe window opening angle based on the outdoor wind speed v and outdoor wind direction D; and the outdoor PM2.5 concentration C. out C must be satisfied out ≤C safe C safe The preset safe concentration threshold; By solving this optimization problem, the optimal window opening angle θ and ventilation duration t are obtained, and control commands containing the parameters θ and t are generated.

6. The energy-saving intelligent control system for doors and windows based on dynamic analysis of environmental parameters using edge computing as described in claim 1, characterized in that, In the instruction verification and correction unit: The physical safety rules include: whether the window opening angle in the instruction exceeds the dynamic safe maximum opening degree determined based on the characteristics of the currently controlled doors and windows and the real-time wind speed and direction; and, when the outdoor rainfall status in the fused sensing data is "raining", the execution of the window opening instruction is prohibited or restricted. The logical consistency rule includes: whether there is a mutually exclusive instruction sequence for the same door or window within an overlapping time window; The correction mechanism includes: for security rule failures, adjusting parameters to security boundary values ​​or canceling instructions; for consistency rule failures, selecting instructions based on policy priority and timeliness.

7. The energy-saving intelligent control system for doors and windows based on dynamic analysis of environmental parameters using edge computing as described in claim 1, characterized in that, The targeted update process of the model optimization unit is as follows: When the optimized sample indicator fails due to "environmental mutation verification", the model optimization unit extracts environmental data before and after the mutation and calculates the effect of the building thermodynamic model on load Q in the dynamic analysis decision unit. win The prediction error is used to adjust the empirical coefficient k in the model in reverse. s or k v ; When the optimized sample indication fails due to "logical consistency check", the model optimization unit analyzes the user behavior probability P corresponding to the conflicting instruction. hist And fine-tune the parameters of the user behavior probability model.

8. The energy-saving intelligent control system for doors and windows based on dynamic analysis of environmental parameters using edge computing according to claim 7, characterized in that, The model optimization unit maintains a database of failed verification cases and accumulates weights for recurring cases of the same type. When the accumulated weight of a certain type of case exceeds a threshold, it triggers the threshold (Q) of the corresponding decision rule in the dynamic analysis and decision-making unit. th P th Or the active adjustment of the optimization problem weight parameters (α, β) in the collaborative control instruction generation unit.

Images