Energy-saving control method and system of environment-friendly ventilation equipment

Abstract

The present application relates to the technical field of energy-saving control of ventilation equipment, and particularly relates to an energy-saving control method and system of environment-friendly ventilation equipment. The present application first acquires various monitoring data of the ventilation equipment; further compares the abnormal deviation characteristics of each module with respect to the corresponding preset standard value, combines the fluctuation characteristics of each monitoring data, acquires the total fluctuation intensity of each module, and determines the master-slave control relationship of two modules; further acquires the prediction data and the predicted air pressure value of each monitoring dimension, compares the prediction data with the preset target value, and controls and adjusts the active module; finally, according to the difference between the current air pressure and the preset standard air pressure, the difference between the predicted air pressure value and the preset standard air pressure, and the comparison between the prediction data and the preset target value, the driving module is controlled and adjusted, which solves the problems of weak module cooperation, response lag and high energy consumption under the traditional unified control strategy, and realizes precise, stable and system-level energy saving.

Description

Technical Field

[0001] This invention relates to the field of energy-saving control technology for ventilation equipment, and specifically to an energy-saving control method and system for environmentally friendly ventilation equipment. Background Technology

[0002] Ventilation equipment serves as a crucial infrastructure in industries such as industrial and civil buildings, and nuclear power plants. Its energy consumption accounts for a significant portion of total building energy consumption, making energy conservation in ventilation systems a core industry demand. Currently, ventilation systems generally employ a unified control strategy, failing to fully consider the differences in functional positioning and data sensitivity between supply and exhaust modules.

[0003] While traditional methods have incorporated variable frequency speed control and neural network technology to achieve dynamic airflow adjustment, they still aim at general load regulation and lack differentiated control logic for different drive modules. This results in lag in system adjustment when dealing with complex operating conditions and low inter-module coordination efficiency, making it difficult to meet the high-level requirements of precise air supply, efficient smoke extraction, and system-level energy saving in industrial and civil building scenarios. Furthermore, existing methods often rely on real-time feedback control, failing to effectively address the lag between data acquisition and execution, which can easily lead to over-regulation and energy waste. Summary of the Invention

[0004] To address the technical problems of insufficient energy efficiency optimization and limited adjustment accuracy in current ventilation equipment control when multiple modules are coordinated, the present invention aims to provide an energy-saving control method and system for environmentally friendly ventilation equipment. The specific technical solution adopted is as follows:

[0005] An energy-saving control method for an environmentally friendly ventilation device, the method comprising:

[0006] The monitoring data of the air supply module and the air exhaust module of the ventilation equipment are acquired separately within a preset time window;

[0007] Within the current time window, based on the abnormal deviation characteristics of each monitoring data item of each module relative to the corresponding preset standard value, the total abnormality coefficient of each module is obtained; based on the drastic fluctuation characteristics of each monitoring data item of each module, the total fluctuation intensity of each module is obtained; based on the difference between the total abnormality coefficients of the two modules and the difference between the total fluctuation intensity, the master-slave control relationship between the two modules is determined.

[0008] Based on the pre-trained LSTM model, the predicted data and predicted air pressure values ​​for each monitoring dimension are output. The active module is controlled and adjusted by comparing the predicted data with the preset target value. The drive module is controlled and adjusted by comparing the difference between the current air pressure and the preset standard air pressure, the difference between the predicted air pressure value and the preset standard air pressure, and the predicted data with the preset target value.

[0009] Furthermore, the method for obtaining the total anomaly coefficient includes:

[0010] Within the current time window, for each monitoring dimension, the sub-anomaly coefficient is obtained based on the difference ratio between the monitoring data at each time point and the corresponding preset standard value;

[0011] The sub-anomaly coefficients of each module are merged using preset attention weights to obtain the total anomaly coefficient.

[0012] Furthermore, the method for obtaining the total fluctuation intensity includes:

[0013] The total fluctuation intensity is obtained by fusing the standard deviation of the data of each monitoring dimension of each module within the current time window using preset attention weights.

[0014] Furthermore, the method for determining the master-slave control relationship between the two modules includes:

[0015] Based on the difference in the total anomaly coefficients of the two modules, an anomaly difference coefficient is obtained; when the anomaly difference coefficient is greater than or equal to a preset difference coefficient threshold, the module corresponding to the largest total anomaly coefficient is determined to be the active module, and the other module is the driving module; when the anomaly difference coefficient is less than the preset difference coefficient threshold, the module corresponding to the largest total fluctuation intensity is determined to be the active module, and the other module is the driving module.

[0016] Furthermore, methods for controlling and adjusting the active module include:

[0017] For the active module, the first weighted error term is obtained based on the deviation between the predicted data of each monitoring dimension and the corresponding preset target value, combined with the corresponding preset attention weight.

[0018] Based on the first weighted error term, the first cumulative error term and the first differential error term are obtained, and the active module is adjusted by PID control.

[0019] Furthermore, the methods for controlling and adjusting the drive module include:

[0020] By integrating the difference between the current air pressure and the preset standard air pressure, as well as the difference between the predicted air pressure value and the preset standard air pressure, a comprehensive air pressure difference is obtained, and a comprehensive differential air pressure difference is also obtained; based on the comprehensive air pressure difference and the comprehensive differential air pressure difference, PD control is used to obtain the first following control quantity;

[0021] For the drive module, based on the deviation between the predicted data of each monitoring dimension and the corresponding preset target value, and combined with the corresponding preset attention weight, a second weighted error term is obtained; based on the second weighted error term, a second cumulative error term is obtained, and a second follow-up control quantity is obtained by using PI control;

[0022] The drive module is controlled and adjusted based on the first follow control quantity and the second follow control quantity.

[0023] Furthermore, the monitoring dimensions of the air supply module should include at least: indoor carbon dioxide concentration, indoor temperature and humidity, and indoor air pressure;

[0024] The monitoring dimensions of the ventilation module should include at least: smoke concentration and volatile organic compound concentration.

[0025] Furthermore, the parameters of the PID controller are obtained through the Ziegler-Nichols tuning method.

[0026] Furthermore, the method for obtaining the preset attention weight includes:

[0027] The random forest algorithm is used to train based on historical data to obtain the preset attention weights of each dimension of data on the module load.

[0028] The present invention also proposes an energy-saving control system for an environmentally friendly ventilation device. The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the steps of the energy-saving control method for the environmentally friendly ventilation device described above.

[0029] The present invention has the following beneficial effects:

[0030] This invention first acquires various monitoring data of the ventilation equipment to provide a foundation for subsequent data analysis. Then, within the current time window, based on the abnormal deviation characteristics of each module's monitoring data relative to the corresponding preset standard value, it obtains the total abnormality coefficient of each module, characterizing the severity of the overall deviation of the environmental parameters corresponding to the module from the health or comfort benchmark. Furthermore, based on the drastic fluctuation characteristics of each module's monitoring data, it obtains the total fluctuation intensity of each module, characterizing the drastic dynamic changes in the environmental load that the module needs to cope with. It further determines the master-slave control relationship between the two modules, ensuring that the system can always focus on adjusting the most urgent or unstable environmental conflicts, improving adjustment efficiency and energy-saving potential at the system level. Furthermore, based on the predicted data and predicted air pressure values ​​of each monitoring dimension, it compares the predicted data with preset target values ​​to control and adjust the active module, effectively avoiding the lag control of traditional measured feedback-based systems, enabling precise and energy-saving active regulation. Finally, based on the difference between the current air pressure and the preset standard air pressure, the difference between the predicted air pressure value and the preset standard air pressure, and by comparing the predicted data with the preset target value, it controls and adjusts the drive module, performing follow-up control to ensure the stability of the ventilation system. By employing dual-module differentiated perception and dynamic master-slave determination, combined with LSTM predictive feedforward and pressure-parameter collaborative following control, the problems of weak module collaboration, slow response, and high energy consumption under traditional unified control strategies are solved, achieving precise, stable, and system-level energy saving. Attached Figure Description

[0031] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 A flowchart illustrating an energy-saving control method for an environmentally friendly ventilation device provided in one embodiment of the present invention;

[0033] Figure 2 This is a flowchart illustrating a method for controlling and adjusting a drive module according to an embodiment of the present invention. Detailed Implementation

[0034] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an energy-saving control method and system for an environmentally friendly ventilation device proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0036] The following description, in conjunction with the accompanying drawings, details the specific scheme of an energy-saving control method and system for an environmentally friendly ventilation device provided by the present invention.

[0037] Please see Figure 1 The flowchart illustrates an energy-saving control method for an environmentally friendly ventilation device according to an embodiment of the present invention, specifically including:

[0038] Step S1: Acquire various monitoring data of the air supply module and the exhaust module of the ventilation equipment within a preset time window.

[0039] In one embodiment of the present invention, the ventilation equipment is divided into an air supply module and an exhaust module. Based on the installation scenario of the ventilation equipment, the core sensitive data dimensions (data monitoring dimensions) are determined. Taking a hospital ward as an example, the monitoring dimensions of the air supply module include at least: indoor carbon dioxide concentration, indoor temperature and humidity, and indoor air pressure; the monitoring dimensions of the exhaust module include at least: smoke concentration and volatile organic compound concentration.

[0040] In the central area of ​​the ward, below the ceiling and about 0.3-0.5 meters from the ceiling, install IoT sensors such as infrared absorption CO2 sensors and integrated temperature and humidity sensors; in the ward, near the door or on the wall connecting to the corridor, about 0.5 meters from the ground, install IoT sensors such as barometric pressure sensors.

[0041] Install IoT sensors such as photoelectric smoke sensors and photoionization VOCs sensors along the airflow path of the exhaust vent. This ensures that polluted air flows past the sensors before being drawn away by the exhaust vent, sensitively detecting smoke and volatile organic compound concentrations. Avoid installing all sensors in corners (airflow dead zones) or areas where people frequently walk and bump into them.

[0042] The preset time window length is set to 1 minute, the acquisition frequency is 1Hz, all IoT sensors synchronously acquire data to obtain various monitoring data, provide a basis for subsequent data analysis, and connect the IoT sensors to the edge computing node to achieve local real-time processing.

[0043] Before data analysis, the process also includes routine noise reduction preprocessing of the data. For example, first-order low-pass filtering is used for slowly changing CO2, temperature and humidity data to smooth random high-frequency noise; median filtering is first used to remove pulses from air pressure data that is susceptible to transient interference, and then low-pass smoothing is performed; median filtering is directly used for smoke and VOCs concentration data that may have sudden spikes in order to preserve trends while suppressing abnormal pulses.

[0044] It should be noted that the installation and setup of various sensors are existing technologies. In other embodiments of the present invention, the implementer can adjust the sensor installation position, sensor type, acquisition frequency and preset time window length according to the actual scenario, and use existing interpolation and resampling methods to unify the acquisition frequency of different sensors, which will not be elaborated further.

[0045] Step S2: Within the current time window, based on the abnormal deviation characteristics of each module's monitoring data relative to the corresponding preset standard value, obtain the total abnormality coefficient of each module; based on the drastic fluctuation characteristics of each module's monitoring data, obtain the total fluctuation intensity of each module; based on the difference in the total abnormality coefficients of the two modules, combined with the difference in the total fluctuation intensity, determine the master-slave control relationship between the two modules.

[0046] Traditional control strategies often pre-set master-slave relationships or adopt rotation mechanisms, which cannot adapt to dynamically changing operating conditions. In one embodiment of the present invention, by analyzing abnormal deviations and severe fluctuations in monitoring data, the master-slave control relationship of the modules is dynamically determined, thereby ensuring that the system can always focus on adjusting the most urgent or unstable environmental conflicts, thus improving adjustment efficiency and energy-saving potential at the system level.

[0047] First, within the current time window (the current preset time window), based on the abnormal deviation characteristics of each module's monitoring data relative to the corresponding preset standard value, the degree and persistence of deviation of each dimension's data relative to its preset standard value are analyzed, and the total abnormality coefficient of each module is obtained, characterizing the severity of the overall deviation of the environmental parameters corresponding to the module from the healthy or comfortable baseline; further, based on the drastic fluctuation characteristics of each module's monitoring data, the drastic change of each dimension's data in time series is analyzed, and the total fluctuation intensity of each module is obtained, characterizing the drastic dynamic change of the environmental load that the module needs to cope with;

[0048] The difference in the total anomaly coefficient between the two modules reflects the urgency gap between them in meeting their core environmental objectives, and the difference in the total fluctuation intensity reflects the difference in the initiative required for intervention in terms of environmental stability. Therefore, based on the difference in the total anomaly coefficient between the two modules and the difference in the total fluctuation intensity, the master-slave control relationship between the two modules is determined.

[0049] Preferably, in one embodiment of the present invention, within the current time window, for each monitoring dimension, a sub-anomaly coefficient is obtained based on the difference ratio between the monitoring data at each time point and the corresponding preset standard value, and the abnormal deviation characteristics of the monitoring data relative to the corresponding preset standard value are expressed in the form of the difference ratio.

[0050] The sub-anomaly coefficients of each module are merged using preset attention weights to obtain the total anomaly coefficient.

[0051] As an example, within the current time window, each monitoring dimension is selected one by one. For each monitoring dimension, if it is a two-way indicator (both too high or too low are considered abnormal, such as temperature, humidity and air pressure): the absolute value of the difference between the monitoring data at each time point and the corresponding preset standard value is used as the numerator, the corresponding preset tolerance deviation is used as the denominator, the fraction ratio is used as the sub-anomaly factor, and the average value of the sub-anomaly factors at all time points is used as the sub-anomaly coefficient.

[0052] For unidirectional indicators (aiming to be below a specific upper limit, the smaller the value the better, such as carbon dioxide concentration, smoke concentration, and volatile organic compound concentration): when the monitoring data at a time point is less than the corresponding preset standard value, the sub-anomaly factor is 0; when the monitoring data at a time point is greater than or equal to the corresponding preset standard value, the difference between the monitoring data and the corresponding preset standard value is used as the numerator, the corresponding preset tolerance deviation is used as the denominator, the ratio of the fractions is used as the sub-anomaly factor, and the average of the sub-anomaly factors at all time points is used as the sub-anomaly coefficient.

[0053] Using preset attention weights, the total anomaly coefficient is obtained by linearly weighting and fusing all sub-anomaly coefficients of all monitoring dimensions of a single module.

[0054] The absolute difference is represented by the absolute value of the difference, the proportion of the difference is represented by the ratio, and the abnormal deviation characteristics of a single dimension are represented by the average. Since the contribution of the various data monitored by the module to the module load is different, linear weighted fusion is also performed by using preset attention weights.

[0055] It should be noted that the monitoring data are based on existing standards for the installation scenario of the ventilation equipment, and corresponding preset standard values ​​are set. For example, referring to the "Indoor Air Quality Standard", the carbon dioxide concentration corresponds to 1000 ppm, and the volatile organic compound (VOCs) concentration corresponds to 0.6 mg / L. Referring to the *Code for Design of General Hospital Buildings*, the indoor temperature should be 22 degrees Celsius in winter (median of 20-24 degrees Celsius) and 25 degrees Celsius in summer (median of 24-26 degrees Celsius); the relative humidity should be 30% in winter and 65% in summer; referring to the *Code for Design of Clean Operating Rooms in Hospitals*, the indoor air pressure should be +10 Pa (median of +5 Pa to +15 Pa relative to standard atmospheric pressure); referring to the *Code for Design of Automatic Fire Alarm Systems*, the smoke concentration should be 0.075%obs / m³ (median of 0.05%-0.1%obs / m³); for range-based standards, the preset tolerance deviation is half the width of the range-based standard, such as when the temperature range is 24-26 degrees Celsius, the tolerance... 1 degree Celsius, the preset tolerance deviation is 1 degree Celsius; when some data corresponds to a one-sided threshold standard, such as carbon dioxide concentration, the background value of the scene is obtained. For example, if the carbon dioxide concentration is 400 ppm, the preset tolerance deviation is the preset standard value minus the background value, which is 600 ppm; for point value standards (without a clear range), the default setting is 20% of the preset standard value.

[0056] Among them, the various reference standards are already known technologies. The background value of the scene can be obtained by statistically analyzing the minimum value of historical data during the period of least human activity. For example, the carbon dioxide concentration is taken as the minimum value between 3 am and 4 am. The average value of the minimum value over the past 30 days is taken as the background value of the environment. In other embodiments of the present invention, the implementer can adjust the various preset standard values ​​and the preset tolerance deviation according to the actual scene, referring to expert experience or relevant standards. For the range threshold, the median value is taken, which will not be elaborated further.

[0057] Preferably, in one embodiment of the present invention, considering that the standard deviation can reflect the characteristics of data fluctuation, the larger the standard deviation, the more drastic the data fluctuation, the standard deviation of each monitoring dimension of each module in the current time window is fused with a preset attention weight to obtain the total fluctuation intensity.

[0058] Specifically, the total fluctuation intensity is obtained by linearly weighting and fusing the standard deviation of all monitoring dimensions of a single module within the current time window using preset attention weights.

[0059] It should be noted that the methods for obtaining the preset attention weights used in linear weighted fusion include: using the random forest algorithm to train based on historical data, obtaining the preset attention weights of each dimension of data on the module load, and each module corresponds to a set of preset attention weights.

[0060] As an example, updates are performed regularly (e.g., weekly or monthly). Each update retrieves historical environmental parameter data (historical monitoring data) within a preset historical period (e.g., the last 90 days) as feature input and retrieves the module power change rate during the same period as the target label. A random forest regression model is constructed and trained. After training, the importance scores of each feature output by the model are extracted. The scores are normalized so that the sum of the weights of each dimension of a module is 1, thus obtaining the preset attention weight.

[0061] The setting and updating methods for the preset attention weights are the same for each module, and the analysis process for each time window is the same. Only one example will be described here, and the details will not be repeated. The update frequency and preset historical duration of the preset attention weights can be adjusted by the implementer according to the scenario. Using a random forest regression model to analyze the importance of data is a technique well-known to those skilled in the art, and will not be elaborated upon here.

[0062] Preferably, in one embodiment of the present invention, considering that when the difference in the degree of abnormality between the two modules is significant, it indicates that one of the modules faces a more urgent adjustment task and needs to prioritize satisfying the module corresponding to the largest total abnormality coefficient; when the difference in the degree of abnormality between the two modules is not significant, it indicates that the urgency of their adjustment tasks is similar, and at this time, the source of instability is dealt with first to prevent the oscillation from spreading, that is, the module corresponding to the largest total fluctuation intensity is prioritized.

[0063] Based on this, the abnormal difference coefficient is obtained according to the difference in the total abnormal coefficients of the two modules; when the abnormal difference coefficient is greater than or equal to the preset difference coefficient threshold, the module corresponding to the largest total abnormal coefficient is determined to be the active module and the other module is the driving module; when the abnormal difference coefficient is less than the preset difference coefficient threshold, the module corresponding to the largest total fluctuation intensity is determined to be the active module and the other module is the driving module.

[0064] As an example, with a preset difference coefficient threshold of 20%, the absolute value of the difference between the total anomaly coefficients of the two modules is used as the numerator, and the average value of the total anomaly coefficients of the two modules is used as the denominator. After the fractional ratio is converted into a percentage, it is used as the anomaly difference coefficient to characterize the difference between the total anomaly coefficients.

[0065] Initially, the air supply module is the default active module. When the total fluctuation intensity is the same, the master-slave control relationship remains unchanged, that is, the relationship between the active module and the drive module remains unchanged. A preset positive parameter for division by zero, such as 1e-6, can be added to the denominator. The preset positive parameter for division by zero has the same unit as the denominator and is used for denominator correction to prevent the denominator from being zero. The preset difference coefficient threshold is set based on the system control hysteresis and stability requirements. It is recommended to take a value range of 15%-25%. Implementers can adjust it according to the needs of the scenario. The smaller the value, the higher the sensitivity and the more frequent the master-slave switching may be.

[0066] Step S3: Based on the pre-trained LSTM model, output the predicted data and predicted air pressure values ​​for each monitoring dimension, compare the predicted data with the preset target value, and control and adjust the active module; based on the difference between the current air pressure and the preset standard air pressure, the difference between the predicted air pressure value and the preset standard air pressure, and compare the predicted data with the preset target value, control and adjust the drive module.

[0067] Because there is a certain lag between real-time adjustment and data feedback when ventilation equipment is adjusted, the control module often takes a period of time to obtain effective data results. If real-time data is used for control, it will lead to control lag, resulting not only in poor control effect, but also in over-control and a significant waste of energy.

[0068] Since LSTM models can directly process multivariate input sequences without the need for manual modeling of coupling equations, and automatically learn the cross-influences between variables during training to output predicted values, in one embodiment of this invention, the predicted data and predicted air pressure values ​​of each monitoring dimension are output based on the pre-trained LSTM model, and feedback adjustment and control are performed based on the prediction results. Considering the difference between the predicted data of the active module and the preset target value, which reflects the degree of deviation between the active module and the target if the current operating state is maintained, the active module is controlled and adjusted by comparing the predicted data with the preset target value, effectively avoiding overshoot, oscillation or repetitive adjustment caused by the lag control based on measured feedback in traditional methods, so as to achieve precise and energy-saving active regulation.

[0069] In a preferred embodiment of the present invention, a pre-trained LSTM model predicts the predicted data and predicted air pressure values ​​for each monitoring dimension at a future preset time step (e.g., 1 minute later) while maintaining the current operating trend.

[0070] The training process includes:

[0071] 1. Dataset Construction and Preprocessing:

[0072] Data source: Comprehensive monitoring data of the ventilation system over a historical operating period (e.g., the past 180 days). The data collection frequency remains consistent with that in step S1, for example, 1 Hz.

[0073] Feature variables: Air supply module data: Monitoring data of the air supply module obtained in step S1 (indoor carbon dioxide concentration, indoor temperature and humidity, indoor air pressure), and the real-time operating frequency of the air supply fan. Exhaust module data: Monitoring data of the exhaust module obtained in step S1 (smoke concentration, volatile organic compound concentration), and the real-time operating frequency of the exhaust fan. Auxiliary environmental data: Timestamp features (such as hour, weekday, used to capture periodic patterns).

[0074] Target Labels:

[0075] Data values ​​of each monitoring dimension of the active module at a preset time step in the future (representing the trend of environmental load).

[0076] Data values ​​of each monitoring dimension of the slave module with a preset time step in the future (representing the trend of environmental load).

[0077] Indoor air pressure value at a future preset time step (characterizing air pressure evolution under current operating conditions).

[0078] Data partitioning and standardization: The dataset, after preprocessing (filtering and noise reduction) in step S1, is partitioned into a training set (70%), a validation set (20%), and a test set (10%) in time series order. Z-score standardization (subtracting the mean and dividing by the standard deviation) is performed on all feature variables and the target variable to eliminate dimensional differences and accelerate model convergence.

[0079] Time series sample construction: The sliding window method is used to construct samples. The length of the input time window is set to T_in (which is consistent with the preset time window length in step S1, such as 60 seconds, and the windows are connected end to end). The preset time step of the predicted output is T_out (such as the data point in the next 60 seconds). That is, based on the data in the current time window, the data at the time point after the preset time step is predicted.

[0080] 2. Model Architecture Design:

[0081] Input layer: Multivariate time series data (i.e., time series segments containing air supply module data, exhaust module data and auxiliary environment data generated in the dataset construction and preprocessing steps).

[0082] LSTM encoding layer: Set up a multi-layer (e.g., 2-3 layers) stacked LSTM network, with 64 or 128 hidden layer units. The LSTM layer is responsible for capturing long-term and short-term temporal dependencies and dynamic features in the input sequence.

[0083] Attention Layer: A self-attention mechanism is introduced after the LSTM output layer. This layer automatically calculates the contribution weights of features at different times within the input time window to future prediction results, strengthens the influence of key time nodes (such as the moment when the smoke concentration suddenly increases), and suppresses irrelevant noise.

[0084] Fully Connected Output Layer (Dense Layer): The output of the attention layer is mapped to the final prediction dimension through the fully connected layer. The output layer contains multiple neurons, corresponding to: the predicted values ​​of each monitoring dimension of the active module, the predicted values ​​of each monitoring dimension of the passive module, and the predicted value of indoor air pressure, thereby obtaining the predicted data and predicted air pressure values ​​for each monitoring dimension.

[0085] 3. Model training strategy:

[0086] Loss function: The mean squared error (MSE) is used as the main loss function to measure the deviation between the predicted and actual values. Mean absolute error (MAE) can also be introduced as an auxiliary evaluation metric.

[0087] Optimizer: The Adam optimizer is selected, with an initial learning rate set to 0.001. A learning rate decay strategy is also used, which automatically reduces the learning rate if the validation set loss does not decrease within a preset number of rounds (e.g., patience=5).

[0088] Training process: Input the training set into the model for iterative training, set the batch size to 32 or 64, and set the maximum number of iterations (Epochs) to 100.

[0089] Early stopping mechanism: During training, the validation set loss is monitored in real time. If the validation set loss does not improve within 10 consecutive epochs, the early stopping mechanism is triggered, and the model weights at the lowest validation set loss are saved as the optimal model.

[0090] 4. Model Deployment and Inference:

[0091] After training, the optimal model is deployed to edge computing nodes. During the real-time control phase, real-time monitoring data within the current preset time window is input into the model, and the model immediately outputs the predicted values ​​of each monitoring dimension of the active module, the predicted values ​​of each monitoring dimension of the passive module, and the predicted air pressure value for future times.

[0092] 5. Regular updates:

[0093] The LSTM model is updated by setting the update frequency to once every 30 days.

[0094] It should be noted that the training and use of LSTM models are common techniques used by those skilled in the art, and are only briefly described here. In other embodiments of the present invention, the implementer can adjust the training process, update frequency and future preset time step by himself.

[0095] Preferably, in one embodiment of the present invention, for the active module, firstly, based on the deviation between the predicted data of each monitoring dimension and the corresponding preset target value, and combined with the corresponding preset attention weight, a first weighted error term is obtained to obtain the degree of deviation between the overall data and the expected target.

[0096] As an example, for the active module, the difference between the predicted data of each monitoring dimension and the corresponding preset target value is used as the numerator, the corresponding preset tolerance deviation is used as the denominator, and the ratio of the fractions is used as the first error sub-item; all the first error sub-items are linearly weighted and summed with the corresponding preset attention weights, and the result is used as the first weighted error term.

[0097] The first cumulative error term and the first differential error term are obtained based on the first weighted error term. The parameters of the PID controller are obtained through the Ziegler-Nichols tuning method, and the active module is adjusted by PID control.

[0098] In this example, the preset target value is the preset standard value of the same dimension; by adjusting the overlap length between time windows, the PID adjustment can be set to be performed once every 10-60 seconds, and 60 seconds is taken here; when calculating the first error sub-item (and the subsequent second error sub-item), for unidirectional indicators, when the predicted data (or monitoring data) is less than the corresponding preset target value, the numerator corresponding to the first error sub-item (and the subsequent second error sub-item) is set to 0.

[0099] In other embodiments of the present invention, a preset standard value can be used as the basis. The preset target value can be flexibly configured by the implementer according to the actual energy-saving needs and the comfort tolerance range of the application scenario. For example, when cooling in summer, the preset standard value is 25 degrees Celsius, and the preset target value is set to 26.5 degrees Celsius.

[0100] It should be noted that the Ziegler-Nichols tuning method, PID control, and the method of obtaining the cumulative error term (such as the first cumulative error term) and the differential error term (such as the first differential error term) based on the error term (such as the first weighted error term) are well-known technologies and will not be described in detail here. In other embodiments of the present invention, the implementer may use other existing methods to obtain the parameters of the PID controller, such as conducting experimental debugging and setting parameters.

[0101] When the active module performs control and adjustment, it will affect the indoor air pressure and the parameters of another module. Therefore, the drive module needs to follow and control the drive module from two aspects: air pressure and predicted monitoring data. Thus, based on the difference between the current air pressure and the preset standard air pressure, the difference between the predicted air pressure value and the preset standard air pressure, and by comparing the predicted data with the preset target value, the drive module is controlled and adjusted to ensure the stability of the ventilation system.

[0102] Preferably, in one embodiment of the present invention, please refer to Figure 2 The diagram illustrates a flowchart of a method for controlling and adjusting a drive module according to an embodiment of the present invention, specifically including:

[0103] Step S301: Combine the difference between the current air pressure and the preset standard air pressure, as well as the difference between the predicted air pressure value and the preset standard air pressure, to obtain the comprehensive air pressure difference and the comprehensive differential air pressure difference; based on the comprehensive air pressure difference and the comprehensive differential air pressure difference, use PD control to obtain the first following control quantity.

[0104] The difference between the current air pressure and the preset standard air pressure reflects the actual deviation of the current indoor air pressure. Analyzing the difference between the predicted air pressure value and the preset standard air pressure is an estimate of the future air pressure deviation. Therefore, the two are combined to obtain the comprehensive air pressure difference, which takes into account the overall air pressure regulation needs of both the current steady-state deviation and the future dynamic trend, and obtains the comprehensive differential air pressure difference.

[0105] As an example, the difference between the current air pressure (the average air pressure in the current time window) and the preset standard air pressure is obtained, and the difference between the predicted air pressure and the preset standard air pressure is obtained. The sum of these two differences is used as the comprehensive air pressure difference. After differentiation, the comprehensive differential air pressure difference is obtained.

[0106] By employing a proportional-derivative (PD) control strategy, the proportional term (P) provides a proportional adjustment action based on the overall pressure difference, aiming to quickly reduce the current and future overall pressure deviation; the derivative term (D) provides a proportional adjustment action based on the overall differential pressure difference, aiming to suppress excessively rapid pressure changes, improve system damping, and reduce overshoot. Therefore, based on the overall pressure difference and the overall differential pressure difference, PD control is used to obtain the first following control quantity, which characterizes the instantaneous and rapid response control component applied to the drive module to maintain stable indoor pressure.

[0107] Step S302: For the drive module, based on the deviation between the predicted data of each monitoring dimension and the corresponding preset target value, and combined with the corresponding preset attention weight, a second weighted error term is obtained; based on the second weighted error term, a second cumulative error term is obtained, and a second follow-up control quantity is obtained by using PI control.

[0108] The deviation between the predicted data of each monitoring dimension and the corresponding preset target value reflects the degree to which the module's core environmental parameters (such as pollutant concentration) deviate from the expected target. At the same time, the second weighted error term is obtained by weighted fusion with preset attention weights, which characterizes the comprehensive predicted deviation level of the overall task target of the driving module in the future.

[0109] As an example, the difference between the predicted data of each monitoring dimension of the driving module and the corresponding preset target value is used as the numerator, the corresponding preset tolerance deviation is used as the denominator, and the fractional ratio is used as the second error sub-item; all the second error sub-items are linearly weighted and summed with the corresponding preset attention weights, and the result is used as the second weighted error term, and the second cumulative error term is obtained based on the second weighted error term;

[0110] By employing a proportional-integral (PI) control strategy, the proportional term (P) provides a fast response based on the second weighted error term, aiming to reduce future prediction deviations; the integral term (I) eliminates steady-state errors by accumulating historical deviations based on the second cumulative error term, aiming to achieve accurate, zero-steady-state error tracking of the drive module's own environmental targets. Therefore, based on the second weighted error term and the second cumulative error term, PI control is used to obtain the second following control quantity, which characterizes the fine, steady-state adjustment component required to continuously optimize the drive module's own task objectives during the following process.

[0111] It should be noted that in this example, the target value is also set to a preset standard value of the same dimension, which is the same as the adjustment frequency of the active module. In other embodiments of the present invention, based on the preset standard value of the same dimension, the preset target value can be flexibly configured by the implementer according to the actual energy-saving needs and the comfort tolerance range of the application scenario, without being limited to a specific value.

[0112] Step S303: Based on the first follow control quantity and the second follow control quantity, control and adjust the drive module.

[0113] The first follow control variable reflects the mandatory and rapid requirements for maintaining system stability, while the second follow control variable reflects the optimization and accuracy requirements for achieving the module-level task objectives (such as pollutant control). Therefore, by controlling and adjusting the drive module based on both, a composite control objective that balances system stability and local task completion is achieved in cooperative following is realized.

[0114] As an example, when both the first and second follower control values ​​are greater than or equal to 0, the maximum value of the two is taken as the final control value, which satisfies the requirements of both control values ​​and avoids over-adjustment caused by superposition.

[0115] When both the first and second follower control values ​​are less than 0, the minimum of the two is taken as the final control value, which satisfies the requirements of both control values ​​and avoids over-adjustment caused by superposition.

[0116] When the first and second following control variables have opposite signs, the air pressure deviation rate R is calculated. The absolute value of the difference between the current window's average air pressure and the preset standard air pressure is used as the numerator, and the preset safety red line deviation is used as the denominator. The ratio of the fractions is R. The minimum value of R and 1 is taken as the dynamic weight r. The smaller R is, the smaller the current air pressure deviation, and the more it is biased towards pollutant control. Therefore, r is the weight of the first following control variable, and 1-r is the weight of the second following control variable. The weighted sum is used as the final control variable to control and adjust the drive module.

[0117] In this example, the parameters of the PID controller are also obtained using the Ziegler-Nichols tuning method. The preset standard atmospheric pressure is +10 Pa, the allowable fluctuation range is +5 Pa to +15 Pa relative to the standard atmospheric pressure, and the preset safety redline deviation is 5 Pa.

[0118] It should be noted that the air supply / exhaust valves of both modules are normally open (or fixed). The adjustment commands from the active module and the drive module are input to their respective frequency converters. Based on these command signals, the frequency converters adjust the frequency and voltage of the AC power output to the drive motor, thereby steplessly changing the fan motor speed and achieving continuous and precise control of the ventilation volume. When an adjustment command is sent to the frequency converter, the corresponding output frequency is subject to upper and lower limits. If the target frequency calculated from the adjustment command exceeds the upper or lower limit range, the target frequency is clamped to the corresponding upper or lower limit value. In this example, the lower limit is 5Hz to maintain minimum ventilation, and the upper limit is 50Hz. The upper and lower limits can be adjusted according to the actual motor model, fan characteristics, and application scenario.

[0119] It should be noted that, under normal circumstances, the preset standard value and the preset target value are not zero. However, considering the variability of the scenario, the implementer can add a preset division-by-zero positive parameter, such as 1e-6, to the denominator. The preset division-by-zero positive parameter has the same unit as the denominator and is used to correct the denominator to prevent the denominator from being zero. In other embodiments of the present invention, the implementer can adjust the preset safety red line deviation according to the actual scenario.

[0120] An embodiment of the present invention also provides an energy-saving control system for an environmentally friendly ventilation device. The system includes a memory, a processor, and a computer program. The memory is used to store the corresponding computer program, and the processor is used to run the corresponding computer program. When the computer program runs in the processor, it can implement the energy-saving control method for an environmentally friendly ventilation device described in steps S1-S3.

[0121] In summary, to address the technical problems of insufficient energy efficiency optimization and limited adjustment accuracy in the current control of ventilation equipment when multiple modules are coordinated, the present invention aims to provide an energy-saving control method and system for environmentally friendly ventilation equipment. The invention first acquires various monitoring data of the ventilation equipment; further, within the current time window, based on the abnormal deviation characteristics of each module's monitoring data relative to the corresponding preset standard values, it acquires the total anomaly coefficient of each module; further, based on the drastic fluctuation characteristics of each module's monitoring data, it acquires the total fluctuation intensity of each module; further, based on the difference in the total anomaly coefficients of two modules, combined with the difference in the total fluctuation intensity, it determines the master-slave control relationship between the two modules; further, based on the pre-trained LSTM model, it outputs predicted data and predicted air pressure values ​​for each monitoring dimension, compares the predicted data with preset target values, and controls and adjusts the active module; finally, based on the difference between the current air pressure and the preset standard air pressure, the difference between the predicted air pressure value and the preset standard air pressure, and comparing the predicted data with the preset target values, it controls and adjusts the drive module. By employing dual-module differentiated perception and dynamic master-slave determination, combined with LSTM predictive feedforward and pressure-parameter collaborative following control, the problems of weak module collaboration, slow response, and high energy consumption under traditional unified control strategies are solved, achieving precise, stable, and system-level energy saving.

[0122] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0123] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. An energy-saving control method for an environmentally friendly ventilation device, characterized in that, The method includes: The monitoring data of the air supply module and the air exhaust module of the ventilation equipment are acquired separately within a preset time window; Within the current time window, based on the abnormal deviation characteristics of each monitoring data item of each module relative to the corresponding preset standard value, the total abnormality coefficient of each module is obtained; based on the drastic fluctuation characteristics of each monitoring data item of each module, the total fluctuation intensity of each module is obtained; based on the difference between the total abnormality coefficients of the two modules and the difference between the total fluctuation intensity, the master-slave control relationship between the two modules is determined. Based on the pre-trained LSTM model, the predicted data and predicted air pressure values ​​for each monitoring dimension are output. The active module is controlled and adjusted by comparing the predicted data with the preset target value. The drive module is controlled and adjusted based on the difference between the current air pressure and the preset standard air pressure, the difference between the predicted air pressure value and the preset standard air pressure, and by comparing the predicted data with the preset target value. The method for determining the master-slave control relationship between the two modules includes: Based on the difference in the total anomaly coefficients of the two modules, an anomaly difference coefficient is obtained; when the anomaly difference coefficient is greater than or equal to a preset difference coefficient threshold, the module corresponding to the largest total anomaly coefficient is determined to be the active module, and the other module is the driving module; when the anomaly difference coefficient is less than the preset difference coefficient threshold, the module corresponding to the largest total fluctuation intensity is determined to be the active module, and the other module is the driving module. The absolute value of the difference between the total anomaly coefficients of the two modules is used as the numerator, and the average value of the total anomaly coefficients of the two modules is used as the denominator. The fractional ratio is then converted to a percentage and used as the anomaly difference coefficient. The monitoring dimensions of the air supply module should include at least: indoor carbon dioxide concentration, indoor temperature and humidity, and indoor air pressure; The monitoring dimensions of the ventilation module should include at least: smoke concentration and volatile organic compound concentration.

2. The energy-saving control method for an environmentally friendly ventilation device according to claim 1, characterized in that, The method for obtaining the total anomaly coefficient includes: Within the current time window, for each monitoring dimension, the sub-anomaly coefficient is obtained based on the difference ratio between the monitoring data at each time point and the corresponding preset standard value; The sub-anomaly coefficients of each module are merged using preset attention weights to obtain the total anomaly coefficient.

3. The energy-saving control method for an environmentally friendly ventilation device according to claim 1, characterized in that, The method for obtaining the total fluctuation intensity includes: The total fluctuation intensity is obtained by fusing the standard deviation of the data of each monitoring dimension of each module within the current time window using preset attention weights.

4. The energy-saving control method for an environmentally friendly ventilation device according to claim 1, characterized in that, Methods for controlling and adjusting the active module include: For the active module, the first weighted error term is obtained based on the deviation between the predicted data of each monitoring dimension and the corresponding preset target value, combined with the corresponding preset attention weight. Based on the first weighted error term, the first cumulative error term and the first differential error term are obtained, and the active module is adjusted by PID control.

5. The energy-saving control method for an environmentally friendly ventilation device according to claim 1, characterized in that, Methods for controlling and adjusting the drive module include: By integrating the difference between the current air pressure and the preset standard air pressure, as well as the difference between the predicted air pressure value and the preset standard air pressure, a comprehensive air pressure difference is obtained, and a comprehensive differential air pressure difference is also obtained; based on the comprehensive air pressure difference and the comprehensive differential air pressure difference, PD control is used to obtain the first following control quantity; For the drive module, based on the deviation between the predicted data of each monitoring dimension and the corresponding preset target value, and combined with the corresponding preset attention weight, a second weighted error term is obtained; based on the second weighted error term, a second cumulative error term is obtained, and a second follow-up control quantity is obtained by using PI control; The drive module is controlled and adjusted based on the first follow control quantity and the second follow control quantity.

6. The energy-saving control method for an environmentally friendly ventilation device according to claim 4 or 5, characterized in that, The parameters of the PID controller are obtained using the Ziegler-Nichols tuning method.

7. An energy-saving control method for an environmentally friendly ventilation device according to any one of claims 2, 3, 4 or 5, characterized in that, The method for obtaining the preset attention weight includes: The random forest algorithm is used to train based on historical data to obtain the preset attention weights of each dimension of data on the module load.

8. An energy-saving control system for an environmentally friendly ventilation device, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the energy-saving control method for an environmentally friendly ventilation device as described in any one of claims 1 to 7.

Images