Purification control method and system based on neural network

Abstract

The invention is suitable for the technical field of air purification, and provides a purification control method and system based on a neural network, and the method comprises the steps: obtaining air quality data and gas handling capacity data, taking the air quality data and the gas handling capacity data as input dimensions of an input layer at the same time, and inputting the air quality data and the gas handling capacity data into a multi-layer sensing network; and on the basis of the gas handling capacity target, the actual gas handling capacity of each fresh air device is controlled. The system is based on a multi-layer sensing network, through input and output control, the purification system learns the complex association between the air quality and the actual handling capacity of each device from historical data, the control strategy is optimized in real time, the airflow path can be adaptively optimized, and the balance between energy saving and deep optimization is maintained. Not only is the change of air quality considered, but also the purification capacity can be adjusted according to the performance of the equipment, so that the control process is more refined and intelligent.

Description

Technical Field

[0001] The present invention belongs to the technical field of air purification, and in particular relates to a purification control method and system based on a neural network. Background Art

[0002] For large indoor venues, such as hospitals, it is particularly important to maintain air circulation and provide stable and efficient continuous air purification. Existing air purification systems generally use a number of air intake and exhaust equipment to form air convection inside the venue, and set a sterilizing filter device upstream of the air return pipeline for sterilization.

[0003] However, as the internal structure of buildings becomes more and more complex, the flow path of airflow inside buildings is becoming increasingly difficult to predict. Therefore, for scenarios with higher cleanliness requirements, the gas in spaces where airflow forms vortices, backflows or dead corners is difficult to be effectively purified with the airflow, making it difficult to quickly remove pollutants in the air.

[0004] However, existing air purification systems can often only supply air according to simple preset modes, and the comprehensiveness of air purification is difficult to ensure, and the purification effect needs to be improved. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide a purification control method based on a neural network, aiming to solve the problem that the existing air purification system can often only supply air according to a simple preset mode, it is difficult to ensure the comprehensiveness of air purification, and the purification effect needs to be improved.

[0006] The embodiment of the present application is implemented by providing a purification control method based on a neural network, the method comprising: Acquire air quality data obtained by a plurality of air quality sensors; Acquire gas processing volume data of several fresh air devices, wherein the fresh air devices are used to absorb air from the environment into the air purification device for purification or to output the purified air to the environment; The air quality data and the gas throughput data are simultaneously used as input dimensions of the input layer and input into a multi-layer perception network, wherein the multi-layer perception network is used to obtain the correlation between the air quality data and the gas throughput data; the multi-layer perception network also includes an output layer, which is used to output the gas throughput target of each fresh air device based on the air quality target; Based on the gas processing volume target, the actual gas processing volume of each of the fresh air devices is controlled.

[0007] Preferably, the method for obtaining the input dimension of the multi-layer perception network is: Obtain the gas processing volume data of several fresh air devices, normalize the gas processing volume data and convert it into a device input vector:

[0008] The air quality data of each air quality sensor is obtained, and the air quality data is normalized and converted into an air quality vector:

[0009] The input dimensions are:

[0010] Wherein, N is the number of fresh air devices, and M is the number of air quality sensors.

[0011] Preferably, the multi-layer perception network contains at least a first hidden layer and a second hidden layer, the number of nodes in the first hidden layer is set to twice the number of nodes in the second hidden layer, and the activation functions of the first hidden layer and the second hidden layer are both ReLU; The number of nodes in the output layer of the multi-layer perception network is set to be equal to the number of the fresh air devices.

[0012] Preferably, the method further comprises: Two Dropout layers are arranged between the first hidden layer and the second hidden layer to prevent the multi-layer perception network from overfitting; the two Dropout layers are only applied to the forward propagation stage, and the setting range of the dropout rate is 0.2~0.4.

[0013] Preferably, the training method of the multi-layer perception network is: Constructing an objective function, setting the objective function as a loss function of an air quality target so that the output gas processing volume target can minimize the loss; The gas volume data and the corresponding air quality data are used as historical data, and the historical data are used as a training group. The multi-layer perception network is trained based on a back propagation method until a preset number of training times is reached.

[0014] Preferably, the training method of the multi-layer perception network further includes: The multi-layer perception network is trained by gradient descent based on the RMSprop optimizer to optimize its network parameters.

[0015] Preferably, the method for acquiring historical data further includes: Acquire a noise data set, and perform noise injection and / or offset processing on the air quality data based on the noise data set to obtain a first extended data set; Acquire a pollutant experimental data set, where the pollutant experimental data set is used to characterize the change over time of the data collected by the air quality sensor under the condition of sudden introduction of artificial high pollutants, to obtain a second extended data set; The first extended data set and the second extended data set are used as historical data sets.

[0016] Another object of an embodiment of the present application is to provide a purification control system based on a neural network, which, when running, executes the steps of the purification control method based on the neural network as described above.

[0017] Preferably, the system further comprises: a plurality of fresh air devices; the fresh air devices comprise air intake devices and air exhaust devices, the air intake devices are used to intake gas into the air purification system, and the air exhaust devices are used to exhaust gas from the air purification system.

[0018] The purification control method based on neural network provided in the embodiment of the present application has the advantages that in order to improve the degree of air purification, the traditional fresh air system can only adopt a high-intensity air supply strategy, but under this strategy, local eddies, circulation dead corners and other situations may occur, resulting in the actual purification effect not necessarily meeting the requirements, but also causing problems such as excessive energy consumption. This scheme constructs an adaptive air purification and adjustment system based on a neural network, which uses the actual air quality index detected by the air quality detector and the gas processing volume data of each fresh air device as the input dimension of the neural network, so that the actual air quality at different locations can be analyzed in multiple dimensions, and the relationship between the fresh air devices set at different locations can be analyzed, and based on the preset target air quality requirements, the system can adaptively adjust the actual air supply or air intake of each device to achieve automatic and precise control. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic diagram of a conventional air purification system provided in an embodiment of the present application; Figure 2 A flowchart of a purification control method based on a neural network provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of a multi-layer perception network provided in an embodiment of the present application; Figure 4 FIG. 4 is a block diagram of the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0021] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, a first script may be referred to as a second script, and similarly, a second script may be referred to as a first script without departing from the scope of this application.

[0022] like Figure 2 As shown, in one embodiment, a purification control method based on a neural network is proposed, which may specifically include the following steps: Step S10, acquiring air quality data obtained by a plurality of air quality sensors.

[0023] Step S20, obtaining gas processing volume data of several fresh air devices, wherein the fresh air devices are used to absorb air in the environment into the air purification device for purification or to output the purified air to the environment.

[0024] Step S30, the air quality data and the gas processing volume data are simultaneously input as input dimensions of the input layer into a multi-layer perception network, wherein the multi-layer perception network is used to obtain the correlation between the air quality data and the gas processing volume data; the multi-layer perception network also includes an output layer, which is used to output the gas processing volume target of each fresh air device based on the air quality target.

[0025] In this embodiment, in this embodiment, as Figure 3 As shown, a structural diagram of a multi-layer perception network is given. In this embodiment, air quality data and gas processing capacity data of fresh air equipment are used as input dimensions of the neural network, so that the relationship between air quality and gas processing flow of each fresh air equipment can be analyzed in multiple dimensions. This comprehensive analysis can effectively optimize the operating parameters of the fresh air equipment, making air purification more efficient and avoiding the problems of inefficiency or waste of resources caused by single-dimensional control. By introducing neural networks, the system can handle complex changes in air quality in different environments, including sudden pollution or long-term environmental changes. Through learning from historical data, the system can automatically adjust the processing capacity of the ventilation equipment according to the concentration of pollutants in different times, different regions or environments, so as to keep the air quality stable within the target range.

[0026] Step S40: Based on the gas processing volume target, controlling the actual gas processing volume of each of the fresh air devices.

[0027] Those skilled in the art will know that conventional indoor air purification systems, such as Figure 1As shown, it is often only possible to perform simple air purification operations based on fresh air equipment (including gas intake equipment, gas exhaust equipment and air purification components). Through the intake and exhaust of gas performed by several fresh air equipment, the gas generates convection in the environment to achieve indoor ventilation and air purification. Its working mode is often only basic air intake and air output control. However, the researchers of this application found that in large indoor places, such as various large-scale event venues, conference halls, and indoor sports halls, the indoor environment is full of various obstacles and moving people. Therefore, the airflow direction inside is complicated. In many cases, there are a large number of vortices, dead corners and other locations in these venues that are difficult to fully ventilate. Therefore, in these local locations, it is difficult to achieve sufficient air purification, resulting in incomplete air purification and potential health risks.

[0028] In the embodiment of the present application, a neural network-based intelligent control system is constructed to address the above problems. The system is built on a multi-layer perception network. Through the control of input and output, the air purification system can learn the complex relationship between air quality and the actual processing capacity of each device from historical data, optimize the control strategy in real time, avoid the problem that traditional control systems are difficult to achieve deep purification and cleaning even with large air volumes, and can adaptively optimize the airflow path to maintain a balance between energy saving and deep optimization. Therefore, the system not only takes into account changes in air quality, but also adjusts the purification capacity according to the performance of the equipment, making the control process more refined and intelligent.

[0029] In a preferred embodiment, the method for obtaining the input dimension of the multi-layer perception network is: Obtain the gas processing volume data of several fresh air devices, normalize the gas processing volume data and convert it into a device input vector:

[0030] The air quality data of each air quality sensor is obtained, and the air quality data is normalized and converted into an air quality vector:

[0031] The input dimensions are:

[0032] Wherein, N is the number of fresh air devices, and M is the number of air quality sensors.

[0033] In an embodiment of the present application, several fresh air devices may refer to the processing flow rate, throughput, etc. of the gas. For example, m³ / min may be used as the default unit for each device, and the numerical value of the exhaust volume and the opposite number of the intake volume, i.e., the negative numerical value, may be substituted into the above matrix. In this embodiment, it is necessary to first summarize the gas processing volume data of different devices into a unified format. It should be noted here that the device data needs to be preprocessed, outliers removed, and time-aligned with the air quality data to ensure that the air quality data at each time point corresponds to the device processing volume data one by one. For the collection of air quality data, the collected raw data also needs to be preprocessed, including operations such as removing outliers, data smoothing, and missing value filling. The data is then standardized or normalized so that the neural network can better learn the relationship between different features. Finally, the air quality vector and the device processing volume vector are used together as input dimensions and input into the input layer of the multi-layer perception network. The air quality can select one of the air quality parameters, for example, PM2.5 particulate matter data.

[0034] In a preferred embodiment, the multi-layer perception network contains at least a first hidden layer and a second hidden layer, the number of nodes in the first hidden layer is set to twice the number of nodes in the second hidden layer, and the activation functions of the first hidden layer and the second hidden layer are both ReLU; The number of nodes in the output layer of the multi-layer perception network is set to be equal to the number of the fresh air devices.

[0035] In an embodiment of the present application, the first hidden layer is used to receive data from the input layer. In this model, the input data can be, for example, PM2.5, CO2 concentration, etc. In this embodiment, the first hidden layer uses 128 nodes, and the second hidden layer only uses 64 nodes. The reason is: it can help the model reduce the risk of overfitting and ensure that the model has more generalization ability. Reducing the number of nodes can be regarded as a feature compression process of the data, which helps the model further refine the most critical features for the control system output.

[0036] It should be noted that this embodiment adopts the ReLU (Rectified Linear Unit) activation function. The reason is that the relationship between air quality and the airflow throughput of each fresh air device may be nonlinear. The nonlinear characteristics of ReLU allow the model to learn complex nonlinear relationships from the input data, which is crucial to the decision-making of the overall system. In addition, ReLU does not have the problem of gradient vanishing in the positive region, which can effectively promote the model to update weights in deep networks. In this application, He normal initialization is used for the ReLU activation function, even if the weight value is drawn from a normal distribution with a mean of zero and a variance of 2 / p (p is the number of input units in the previous layer). This can avoid the problem of gradient vanishing or exploding when the network starts training.

[0037] In a preferred embodiment, two Dropout layers are arranged between the first hidden layer and the second hidden layer to prevent the multi-layer perceptual network from overfitting; the two Dropout layers are only applied to the forward propagation stage, and the setting range of the dropout rate is 0.2~0.4.

[0038] In the embodiment of the present application, in order to further improve the performance of the model, two Dropout layers are added between hidden layer 1 and hidden layer 2. The Dropout regularization method randomly "turns off" some neurons in each training, thereby forcing the model to generalize better. In the present application, the setting range of the dropout rate is 0.2~0.4, preferably, 0.2, that is, 20% of the neurons in the hidden layer will be "shielded" and will not transmit any signal. In order to keep the total amount of network activation output consistent, the output of non-failed neurons is usually scaled proportionally. For example, if the dropout rate is 0.2, the network will amplify the output value of the remaining 80% of the neurons to the original 1 / (1 - dropout rate), that is, multiply by 1 / 0.8 = 1.25. In this process, Dropout is only applied in the forward propagation stage, so during back propagation, the weights of the activated neurons need to be updated. Dropout does not directly change the back propagation algorithm, but indirectly determines which neurons' weights will be updated by affecting the selection of neurons in the forward propagation.

[0039] In a preferred embodiment, the training method of the multi-layer perception network is: Construct an objective function, and set the objective function as a loss function of the air quality target so that the output gas processing volume target can minimize the loss; use the gas volume data and the corresponding air quality data as historical data, use the historical data as a training group, and train the multi-layer perception network based on the back propagation method until the preset number of training times is reached.

[0040] In an embodiment of the present application, the data generated during the operation of the system can be used as historical measurement data to train the system, so the longer the system runs, the more accurate the system control is.

[0041] In a preferred embodiment, the training method of the multi-layer perception network further includes: The multi-layer perception network is trained by gradient descent based on the RMSprop optimizer to optimize its network parameters.

[0042] The reason for using the RMSprop optimizer in the embodiment of the present application is that the optimizer dynamically adjusts the learning rate according to the update history of different parameters. It maintains a specific learning rate for each parameter and balances the learning rate by accumulating the squared gradient of each parameter. This adaptive learning rate mechanism can effectively cope with the scale differences between different dimensions of the gradient, avoiding the problem of too large or too small learning rate.

[0043] In a preferred embodiment, the method for obtaining historical data further includes: Acquire a noise data set, and perform noise injection and / or offset processing on the air quality data based on the noise data set to obtain a first extended data set; acquire a pollutant experimental data set, and the pollutant experimental data set is used to characterize the changes over time of the data collected by the air quality sensor under the condition of sudden introduction of artificial high pollutants, to obtain a second extended data set; use the first extended data set and the second extended data set as historical data sets.

[0044] In the embodiment of the present application, two methods for expanding the data set are given, and the system can be further trained based on the above expanded training set. The introduction of these expanded data sets enhances the robustness and generalization ability of the model, making it more adaptable to various emergencies in practical applications.

[0045] In one embodiment, a purification control system based on a neural network is provided. When the system is running, the system executes the steps of the purification control method based on the neural network as described above.

[0046] In the embodiment of the present application, the system can be a computer software or program system, which runs in the hardware containing the processing device, and when the above system is running, the corresponding method is executed. For the description of the above purification control method based on neural network, please refer to the above text, which will not be repeated here.

[0047] In the embodiment of the present application, based on this system, the advantage is that in order to improve the degree of air purification, the traditional fresh air system can only adopt a high-intensity air supply strategy, but this strategy may cause local eddies, circulation dead corners and other situations, resulting in the actual purification effect not necessarily meeting the requirements, but also causing problems such as excessive energy consumption. This scheme constructs an adaptive air purification and adjustment system based on a neural network, which uses the actual air quality index detected by the air quality detector and the gas processing volume data of each fresh air device as the input dimension of the neural network, so that the actual air quality at different locations can be analyzed in multiple dimensions, and the relationship between the fresh air devices set at different locations, and based on the preset target air quality requirements, the system can adaptively adjust the actual air supply or air intake of each device to achieve automatic and precise control.

[0048] In one embodiment, the system further includes: a plurality of fresh air devices; the fresh air devices include air intake devices and air exhaust devices, the air intake devices are used to intake gas into the air purification system, and the air exhaust devices are used to exhaust gas from the air purification system.

[0049] In this embodiment, the system is an integrated system including a number of intelligent fresh air devices and their control software. The control software can run in an independent dedicated terminal control device and can run in computers, cloud servers and other devices.

[0050] Figure 4 FIG. 1 shows an internal structure diagram of a computer device in an embodiment. The computer device may specifically refer to a computer device on which the above method is executed. Figure 4As shown, the computer device includes a processor, a memory, a network interface, an input device and a display screen connected through a system bus. Among them, the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program. When the computer program is executed by the processor, the processor can implement a purification control method based on a neural network. The internal memory may also store a computer program. When the computer program is executed by the processor, the processor can execute a purification control method based on a neural network. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0051] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0052] It should be understood that, although each step in the flow chart of each embodiment of the present application is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is clear explanation in this article, the execution of these steps does not have strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0053] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0054] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0055] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A purification control method based on neural network, characterized in that: The method comprises: Acquire air quality data obtained by a plurality of air quality sensors; Acquire gas processing volume data of several fresh air devices, wherein the fresh air devices are used to absorb air from the environment into the air purification device for purification or to output the purified air to the environment; The air quality data and the gas throughput data are simultaneously used as input dimensions of the input layer and input into a multi-layer perception network, wherein the multi-layer perception network is used to obtain the correlation between the air quality data and the gas throughput data; the multi-layer perception network also includes an output layer, which is used to output the gas throughput target of each fresh air device based on the air quality target; Based on the gas processing volume target, the actual gas processing volume of each of the fresh air devices is controlled.

2. A purification control method based on neural network according to claim 1, characterized in that: The method for obtaining the input dimension of the multi-layer perception network is: Obtain the gas processing volume data of several fresh air devices, normalize the gas processing volume data and convert it into a device input vector: The air quality data of each air quality sensor is obtained, and the air quality data is normalized and converted into an air quality vector: The input dimensions are: Wherein, N is the number of fresh air devices, and M is the number of air quality sensors.

3. A purification control method based on neural network according to claim 1, characterized in that: The multi-layer perception network comprises at least a first hidden layer and a second hidden layer, the number of nodes in the first hidden layer is set to be twice the number of nodes in the second hidden layer, and the activation functions of the first hidden layer and the second hidden layer are both ReLU; The number of nodes in the output layer of the multi-layer perception network is set to be equal to the number of the fresh air devices.

4. A purification control method based on neural network according to claim 3, characterized in that: The method further comprises: Two Dropout layers are arranged between the first hidden layer and the second hidden layer to prevent the multi-layer perception network from overfitting; the two Dropout layers are only applied to the forward propagation stage, and the setting range of the dropout rate is 0.2~0.

4.

5. The purification control method based on neural network according to claim 1, characterized in that: The training method of the multi-layer perception network is: Constructing an objective function, setting the objective function as a loss function of an air quality target so that the output gas processing volume target can minimize the loss; The gas volume data and the corresponding air quality data are used as historical data, and the historical data are used as a training group. The multi-layer perception network is trained based on a back propagation method until a preset number of training times is reached.

6. A purification control method based on neural network according to claim 5, characterized in that: The training method of the multi-layer perception network also includes: The multi-layer perception network is trained by gradient descent based on the RMSprop optimizer to optimize its network parameters.

7. A purification control method based on neural network according to claim 5, characterized in that: The method for obtaining historical data also includes: Acquire a noise data set, and perform noise injection and / or offset processing on the air quality data based on the noise data set to obtain a first extended data set; Acquire a pollutant experimental data set, where the pollutant experimental data set is used to characterize the change over time of the data collected by the air quality sensor under the condition of sudden introduction of artificial high pollutants, to obtain a second extended data set; The first extended data set and the second extended data set are used as historical data sets.

8. A purification control system based on neural network, characterized in that: When the system is running, the steps of the purification control method based on neural network as described in any one of claims 1 to 7 are executed.

9. A purification control system based on a neural network according to claim 8, characterized in that: The system also includes: a number of fresh air devices; The fresh air equipment includes an air suction device and an air discharge device. The air suction device is used to suck gas into the air purification system, and the air discharge device is used to discharge gas from the air purification system.

Images