A method and system for dynamic positioning of indoor mobile pollution sources

Abstract

The application provides a kind of indoor mobile pollution source dynamic positioning method and system, it is related to indoor environment safety technical field, comprising: the pollution concentration field distribution sequence collected is sequentially input into the dynamic positioning prediction model trained, and the dynamic position coordinates of mobile pollution source are output;The training method of dynamic positioning prediction model includes: using fluid dynamics real-time simulation to generate the pollution concentration field distribution sequence of mobile pollution source in steady state;The pollution concentration field distribution sequence is sequentially coded as gray scale image, segmented into multiple sub-images and screened;Each sub-image after screening is subjected to data enhancement to generate multiple derivative images, and derivative position coordinates are generated in indoor space coordinates;Multiple derivative images and corresponding derivative position coordinates are input into neural network model for training to obtain dynamic positioning prediction model.The application solves the problems of strong data dependency and poor dynamic positioning accuracy in pollution source positioning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of indoor environment safety, and in particular to a dynamic positioning method and system for an indoor mobile pollution source based on a small sample neural network. BACKGROUND

[0002] In the field of indoor gas pollution source positioning, traditional static positioning methods can generally only locate fixed positions and mainly rely on sensor networks. Gradient optimization methods based on fixed sensor networks are used to realize pollution source positioning. Such methods usually require the deployment of dense sensors, and the positioning accuracy is not only limited by the number of sensors but also affected by the quality of the data collected by the sensors. In addition, error accumulation may occur due to local optimal solutions. It can be seen that traditional static positioning technology has obvious limitations.

[0003] Currently, mobile robots with dynamic search capabilities are also used for dynamic positioning, but the path planning efficiency is low in large-scale or complex airflow environments, and real-time performance is difficult to guarantee. Dynamic models are also used to realize dynamic positioning of indoor mobile pollution sources. For example, the dynamic positioning method represented by the long short-term memory network (LSTM) not only relies on continuous time step data, but also has high model training and inference calculation costs, making it difficult to adapt to mobile pollution sources in discrete scenarios. In addition, traditional machine learning methods require a large amount of experimental data, but it is difficult to obtain actual data, and insufficient data will result in low positioning accuracy. Moreover, existing methods are mostly designed for static pollution sources and cannot distinguish spatial features at different steps of mobile pollution sources, resulting in accumulated trajectory prediction errors, sensitivity to spatial layout, weak cross-scene generalization ability, and difficulty in adapting to complex indoor environments of different sizes and shapes. SUMMARY

[0004] To solve the above problems, the present application provides a dynamic positioning method and system for an indoor mobile pollution source based on a small sample neural network, which can solve the problems of strong dependence on data quantity, poor dynamic positioning accuracy, and weak cross-scene generalization ability in current pollution source positioning.

[0005] In one aspect, the present application provides a dynamic positioning method for an indoor mobile pollution source, which comprises:

[0006] A plurality of sensors are arranged in the indoor environment to collect the pollution concentration field distribution of the current mobile pollution source in the entire indoor area in real time. The collected pollution concentration field distribution sequence is sequentially input into a trained dynamic positioning prediction model, and the dynamic position coordinates of the mobile pollution source are output.

[0007] The training method of the dynamic positioning prediction model comprises:

[0008] The indoor space model is constructed, indoor space coordinates are established, and a pollutant concentration field distribution sequence of the moving pollution source in a steady state is generated by using fluid dynamics real-time simulation;

[0009] The pollutant concentration field distribution sequence is sequentially coded into a gray-scale image, the gray-scale image is segmented into a plurality of sub-images and screened to obtain a plurality of sub-images covering key feature regions of the moving pollution source;

[0010] Data augmentation is performed on each screened sub-image to generate a plurality of derivative images, and derivative position coordinates of each derivative image are generated in the indoor space coordinates;

[0011] A plurality of sets of training sample data sets are generated based on the plurality of derivative images and the corresponding derivative position coordinates, and the plurality of sets of training sample data sets are input into a neural network model for training to obtain a dynamic positioning prediction model.

[0012] Further, the convection-diffusion equation of fluid dynamics real-time simulation is:

[0013] ;

[0014] wherein, denotes the simulated concentration at the position at the moment; denotes the flow field in which the pollution source is located; denotes the concentration diffusion coefficient of the pollutant; is a source term, which denotes the instantaneous emission of the pollutant at the position at the moment.

[0015] Further, when the pollutant release rate is constant and the pollution source is a point source:

[0016] ;

[0017] wherein, denotes the Dirac function, and denotes a pollution source with a release rate of only at the position.

[0018] Further, the manner of generating a plurality of derivative images is that each sub-image is subjected to rotation step processing.

[0019] Further, the pollutant concentration field distribution sequence collected every set time length is input into the dynamic positioning prediction model to sequentially obtain dynamic position coordinates of a plurality of moving pollution sources; and the sequentially obtained dynamic position coordinates of the plurality of moving pollution sources are combined to obtain a dynamic movement path of the moving pollution source.

[0020] Further, the performance of the dynamic positioning prediction model is improved by using the coefficient of determination, the accuracy rate of the weighted accuracy rate based on the probability density, and the accuracy rate of the Euclidean distance between the coordinate position predicted by the dynamic positioning prediction model and the true coordinate position.

[0021] Further, the gray-scale image is segmented into a plurality of sub-images based on a grid cutting formula.

[0022] Further, the grid cutting formula is:

[0023] ;

[0024] wherein, represents the spatial size, used for dynamically adjusting the clipping ratio .

[0025] In another aspect, the present application provides a dynamic positioning system for indoor mobile pollution sources, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method according to any one of the above aspects.

[0026] Overall, the present application provides a dynamic positioning method and system for indoor mobile pollution sources, which can achieve the following beneficial effects through the technical solutions conceived by the present application:

[0027] (1) The present application inputs the pollution concentration field distribution sequence collected every set time length into the trained dynamic positioning prediction model in turn, outputs the dynamic position coordinates of the mobile pollution source, models the discrete path of the mobile pollution source moving every set time, captures the spatial distribution characteristics of the pollution source at different moving stages, not only realizes high-precision dynamic positioning of the pollution source during movement, locks the dynamic moving position of the mobile pollution source, but also accurately obtains the moving path of the mobile pollution source, effectively solves the problems of traditional static positioning methods, such as inability to capture moving characteristics, high calculation cost of existing models, strong dependence on data volume, poor dynamic positioning accuracy, weak cross-scene generalization ability, etc., and provides a new scheme for efficient monitoring of dynamic pollution sources.

[0028] (2) The present application increases data samples by segmentation, screening, rotation, etc. to expand small sample data to obtain derivative images, and trains using the derivative images and corresponding derivative position coordinates; this small sample data enhancement method significantly reduces the dependence on large-scale measured data and improves the cross-scene generalization ability.

[0029] (3) The present application can dynamically adjust the cutting ratio by adjusting the space size, can adapt to different indoor space sizes, and is flexible to adapt to various complex indoor scenes such as factory assembly lines, laboratories, hospital corridors and warehouse environments with mobile pollution sources, provides an efficient and reliable solution for building environment safety, industrial workshop dynamic pollution monitoring and other fields, and has a wide application prospect. BRIEF DESCRIPTION OF DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0031] Figure 1 is a method principle schematic diagram of a dynamic positioning method and system for indoor mobile pollution sources provided by the present application;

[0032] Figure 2 is a dynamic positioning prediction model training step schematic diagram of a dynamic positioning method and system for indoor mobile pollution sources provided by the present application;

[0033] Figure 3 is an indoor space model schematic diagram of a dynamic positioning method and system for indoor mobile pollution sources provided by the present application;

[0034] Figure 4 is a comparison schematic diagram of prediction accuracy before and after optimization of a dynamic positioning method and system for indoor mobile pollution sources provided by the present application;

[0035] Figure 5 is a verification space schematic diagram of a dynamic positioning method and system for indoor mobile pollution sources provided by the present application;

[0036] Figure 6 is a Euclidean distance and accuracy relationship schematic diagram of a dynamic positioning method and system for indoor mobile pollution sources provided by the present application. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical scheme and advantages of the present application more clear, the technical scheme in the present application will be described clearly and completely below in combination with the drawings and embodiments in the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application.

[0038] It should be noted that in the description of the embodiments of the present application, the terms "comprising", "including", or any other variant thereof are intended to cover a non-exclusive inclusion, so that a process, step, or apparatus that comprises a list of elements does not only include those elements, but also includes other elements that are not expressly listed, or further includes elements inherent in such process, step, or apparatus. Without more limitations, an element defined by the phrase "comprising a" does not exclude the presence of additional identical elements in the process, step, or apparatus that includes the element.

[0039] The present application proposes a small sample neural network-based indoor mobile pollution source dynamic positioning method and system, which realizes high-precision dynamic positioning of the pollution source in the moving process by modeling the discrete path of the mobile pollution source moving every set time. Specifically, the method comprises: setting multiple sensors in the indoor to collect the pollution concentration field distribution of the current mobile pollution source in the entire indoor area in real time; the collected pollution concentration field distribution sequence is input into the trained dynamic positioning prediction model in turn, and the dynamic position coordinates of the mobile pollution source are output.

[0040] As shown in Figure 1 16 sensors are set in the indoor to collect the pollution concentration field distribution of the entire indoor area, the pollution concentration field distribution collected by each sensor is spliced to obtain the pollution concentration field distribution of the entire indoor area. It can be collected every set time, such as 1 minute, or it can be collected every fixed step, such as 0.5 meters. The pollution concentration field distribution U1 is collected at t1, the pollution concentration field distribution U2 is collected at t2, and the pollution concentration field distribution U3 is collected at t3. According to the pollution concentration field distribution sequence [U1, U2, U3] obtained at the collection time; the pollution concentration field distribution in the pollution concentration field distribution sequence [U1, U2, U3] is input into the dynamic positioning prediction model in turn, and A(x1, y1, z1), B(x2, y2, z2), C(x3, y3, z3) are obtained in turn, so that the dynamic positioning of the mobile pollution source is realized.

[0041] Further, the collected pollution concentration field distribution sequence is input into the dynamic positioning prediction model to obtain the dynamic position coordinates of multiple mobile pollution sources in turn; the dynamic position coordinates of multiple mobile pollution sources obtained in turn are combined to obtain the dynamic moving path of the mobile pollution source.

[0042] For example, A(x1, y1, z1), B(x2, y2, z2), and C(x3, y3, z3) are obtained in sequence, and A-B-C is combined, so that the moving pollution source is first moved from A to B, and then from B to C, and the dynamic moving path of the moving pollution source is obtained. It should be noted that the shorter the set time length or the smaller the set fixed step length, the more accurate the dynamic moving path of the moving pollution source.

[0043] As shown in Figure 2 , the training method of the dynamic positioning prediction model comprises:

[0044] Step 101: Construct an indoor space model, establish an indoor space coordinate, and generate a pollution concentration field distribution sequence of the moving pollution source in a steady state by fluid dynamics real-time simulation.

[0045] As shown in Figure 3 , an indoor space model is constructed as x×y×z=30m×40m×4m, and a steady concentration field distribution of the moving pollution source under a discrete moving path is generated by fluid dynamics simulation. The moving pollution source moves under the discrete moving path; it can pause after moving a fixed step length, such as 0.5m; or pause after moving a set time length, such as 1min; run fluid dynamics simulation of the current pollution source diffusion state until the diffusion is stable, and generate the steady pollution concentration field distribution corresponding to each fixed step length by solving the convection-diffusion equation of fluid dynamics real-time simulation. For example, the pollution concentration field distribution is collected once for each movement, and 45 pollution concentration field distributions are obtained after 45 movements.

[0046] As an embodiment, in order to better describe the dynamic diffusion process of the moving pollution in the fluid, the convection-diffusion equation of fluid dynamics real-time simulation is used in the present application as:

[0047] ;

[0048] Among them, represents the simulation concentration at the moment position; represents the flow field where the pollution source is located; represents the concentration diffusion coefficient of the pollution; is a source term, which represents the instantaneous emission of the pollution at the position at the moment.

[0049] Further, when the pollution release rate is constant and the pollution source is a point source:

[0050] ;

[0051] wherein, represents the Dirac function, and represents a pollution source with a release rate of only at the coordinate .

[0052] Step 102: sequentially encode the pollution concentration field distribution sequence into a gray-scale image, segment the gray-scale image into multiple sub-images, and screen to obtain multiple sub-images covering the key feature regions of the moving pollution source.

[0053] For example, after generating 45 pollution concentration field distributions corresponding to 45 pollution source positions, sequentially encode the 45 pollution concentration field distributions into gray-scale images according to the generation order. The pixels of the gray-scale images reflect the concentration difference through linear mapping, thereby ensuring the prominent features of the high-concentration regions.

[0054] As an embodiment, the pixels of the gray-scale image are :

[0055] ;

[0056] wherein, represents the maximum value of the pollution concentration at all positions in the simulated indoor area; represents the minimum value of the pollution concentration at all positions in the simulated indoor area; represents the pollution concentration at the position coordinate in the simulated indoor area.

[0057] As an embodiment, the gray-scale image is segmented into multiple sub-images based on grid cutting. The grid cutting formula is: ;

[0058] wherein, represents the spatial size, which is used to dynamically adjust the cropping ratio .

[0059] It should be noted that the cropping ratio is dynamically adjusted through the grid cutting formula, thereby adapting to different spatial sizes and flexibly adapting to various complex scenarios.

[0060] Step 103: data augmentation is performed on each screened sub-image to generate multiple derivative images, and the derivative position coordinates of each derivative image are generated in the indoor space coordinates.

[0061] As another embodiment, the way to generate multiple derivative images is to perform rotation step processing on each sub-image.

[0062] In order to enhance the data, the application preferably divides the gray image into multiple sub-images based on a grid cutting formula, screens the sub-images, rotates the screened sub-images to obtain multiple derivative images, thereby eliminating invalid data and only rotating the sub-images with concentration distribution, while avoiding information dispersion or loss and increasing the diversity and effectiveness of the samples.

[0063] For example, the size of the gray image is 40x30, based on the grid cutting strategy, such as the cutting ratio α =0.75, the gray image is divided into 100 sub-images with a size of 4x3, after screening, the sub-images of the area where the pollution source has not spread are eliminated to obtain 45 sub-images to cover the key feature area of the pollution source, and finally through the rotation step θ=10°, 36 derivative images are generated for each sub-image, and 45x36=1620 derivative images are obtained. In this way, the concentration distribution sequence of the pollutant under different observation angles can be simulated, and the adaptability of the dynamic positioning prediction model to complex spatial layout can be significantly improved.

[0064] The data samples are increased by segmentation, screening, rotation and other methods to expand the derived images from the small sample data, and the derived images and corresponding derived position coordinates are used for training; this small sample data enhancement method significantly reduces the dependence on large-scale measured data and improves the cross-scene generalization ability.

[0065] Step 104: Based on the multiple derived images and corresponding derived position coordinates, multiple sets of training sample data sets are generated, and the multiple sets of training sample data sets are input into the neural network model for training to obtain a dynamic positioning prediction model.

[0066] As an embodiment, a nine-layer residual network (ResNet-9) can be used, the screened sub-images are sequentially input into the image input branch for rotation processing by increasing the image input branch, the spatial features are extracted by a 4x4 convolution kernel (channel number is 64), the gradient vanishing problem is avoided by using a skip connection mechanism, and the position coordinates are output by a single-layer fully connected layer (FC=1), thereby ensuring effective training of the deep network.

[0067] Further, the image input branch encodes the different screened sub-images, converts them into corresponding multi-dimensional vectors through an embedding layer, and then rotates them for rotation processing to extract spatial features, which are then input into a single-layer fully connected layer after being spliced with image features. For example, 45 sub-images are step- encoded to obtain 1, 2, 3,..., 45, which are converted into 45-dimensional vectors through an embedding layer.

[0068] The application can distinguish the position characteristics of the pollution source in different moving stages by introducing the step coding, while retaining the high efficient feature extraction capability of the original Res Net; in addition, by optimizing the network parameters (such as the convolution kernel size, the activation function) and simplifying the fully connected layer, the feature redundancy is reduced, and the training efficiency is improved.

[0069] Preferably, the application uses the coefficient of determination, the weighted accuracy rate based on the probability density, and the accuracy rate of the Euclidean distance between the coordinate position predicted by the dynamic positioning prediction model and the real coordinate position to improve the performance of the dynamic positioning prediction model.

[0070] Specifically, the coefficient of determination , the weighted accuracy rate based on the probability density , and the Euclidean distance The formulas are as follows:

[0071] ;

[0072] ;

[0073] ;

[0074] According to the results of the coefficient of determination , the weighted accuracy rate based on the probability density , and the Euclidean distance , the model parameters of the dynamic positioning prediction model are constantly optimized and adjusted to improve the accuracy and efficiency of the model.

[0075] The model parameters include the clipping ratio , the rotation step , the convolution kernel size, the number of fully connected layers, etc. These parameters have important influence on the generalization ability and positioning accuracy of the dynamic positioning prediction model; by comprehensively analyzing the experimental data and model performance, the optimal parameter combination is determined to improve the accuracy and efficiency of the model.

[0076] To verify the effectiveness of the application, experiments and simulations are carried out: in the scene verification, the training data is generated by fluid dynamics simulation, the CNN model is trained, and the verification is carried out on the test set, wherein the data is used for training, testing and prediction according to the ratio of 7:2:1.

[0077] As shown in Figure 4 , by comparing the baseline model (NL=1, KS=1×1), the optimized model performs well on the 45-step linear moving path, the weighted accuracy rate is improved by 17%, the single-step positioning accuracy within 1 meter radius is 88%, the cross-scene error is <0.3 meters, which is improved by 13% compared with the static model, the coefficient of determination , indicating that the model is highly fitted to the data, and the Euclidean distance is 0.18.

[0078] To verify the generalization ability, as shown in Figure 5 and Figure 6 , the application constructs a verification space of 10*5*3 meters, and the pollution source moves according to a random jumping path (8 key points). When the clipping ratio is adjusted to and the feature retention rate is 92.7%, the dynamic positioning prediction model can achieve an average positioning error of 0.28 meters without retraining, the positioning accuracy within a radius of 1 meter is 100%, and the Euclidean distance is reduced by 21.43%.

[0079] After the dynamic positioning prediction model is trained, the dynamic positioning prediction model realizes real-time tracking of the dynamic pollution source through a sliding window mechanism. The current steady-state concentration field distribution is collected every certain period of time and input into the dynamic positioning prediction model to capture the spatial distribution characteristics of the pollution source in different moving stages, and the position coordinates of the moving pollution source are output, and the historical prediction results are combined to optimize the trajectory continuity. The elastic weight consolidation (EWC) algorithm is used to incrementally update the model parameters to prevent catastrophic forgetting and ensure that the model maintains high accuracy in long-term operation. In addition, by adaptively adjusting the clipping ratio and rotation enhancement strategy, the model can flexibly cope with different spatial sizes and pollution source moving modes, significantly expanding the application range.

[0080] In the second aspect, the application provides a dynamic positioning system for an indoor moving pollution source, which comprises a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to realize the steps of the method described in any one of the preceding aspects. The technical features of the system are consistent with the method, and will not be described here.

[0081] In summary, the application combines computational fluid dynamics (CFD) simulation data with a small-sample optimized residual network (ResNet) to train a dynamic positioning prediction model, improves the generalization ability and dynamic positioning accuracy of the model under small sample conditions, and realizes efficient and accurate dynamic positioning of the position of the moving pollution source. Compared with traditional methods, the application shows significant advantages in positioning accuracy, calculation time, and calculation complexity.

[0082] It should be noted that, for the foregoing embodiments, in order to simply describe, they are all expressed as a series of action combinations, but those skilled in the art should know that the application is not limited by the order of the described actions, because according to the application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions and modules involved are not necessarily required by the application.

[0083] In the above embodiments, the description of each embodiment focuses on different aspects, and the parts not described in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0084] In several embodiments provided in the present application, it should be understood that the disclosed method or system can be implemented in other ways. For example, the above-described embodiments are merely illustrative, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0085] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place or distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0086] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0087] When the integrated unit is realized in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for making a computer device (which can be a personal computer, a server or a network device, etc.) execute all or part of the steps of the methods described in each embodiment of the present application.

[0088] Those of ordinary skill in the art can understand that all or part of each circuit in the above embodiments can be instructed by a program to complete the relevant hardware, and the program can be stored in a computer readable storage medium, which can include a flash disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0089] The above-described embodiments are merely exemplary embodiments of the present disclosure, and the present disclosure is not limited thereto. That is, any equivalent changes and modifications made in accordance with the teachings of the present disclosure are still included in the scope of the present disclosure. Embodiments of the present disclosure will be readily apparent to those skilled in the art in view of the disclosure herein with the description and practice of the present disclosure. The present application is intended to cover any variations, uses, or adaptations of the present disclosure following the general principles thereof and including such insubstantial variations and modifications thereof as come within the scope of the present disclosure. The scope of the present disclosure is defined by the appended claims rather than the description and embodiments thereof.

[0090] Any of the technical features of the above embodiments can be combined, and for the sake of brevity, not all possible combinations of the various technical features described above are repeated, however, any combination of the technical features should be considered as within the scope of the present disclosure, as long as the combination does not result in a contradiction.

[0091] Those skilled in the art easily understand that the above only describes preferred embodiments of the present application and is not intended to limit the present application. Any modifications, equivalent replacements and improvements made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A dynamic positioning method for indoor mobile pollution sources, characterized in that: The method comprises: Multiple sensors are set up indoors to collect real-time data on the pollutant concentration distribution of the current mobile pollution source throughout the indoor area. The collected pollutant concentration distribution sequence is sequentially input into a trained dynamic positioning prediction model to output the dynamic position coordinates of the mobile pollution source. The training method of the dynamic positioning prediction model includes: Construct an indoor space model, establish indoor space coordinates, and use fluid dynamics real-time simulation to generate the pollutant concentration field distribution sequence of mobile pollution sources in a steady state; Encoding the pollutant concentration field distribution sequence into grayscale images in sequence, dividing the grayscale images into multiple sub-images and screening them to obtain multiple sub-images covering key feature areas of mobile pollution sources; Perform data augmentation on each filtered sub-image to generate multiple derivative images, and generate the derivative position coordinates of each derivative image in indoor space coordinates; Based on the multiple derived images and the corresponding derived position coordinates, multiple sets of training sample data sets are generated, and the multiple sets of training sample data sets are input into the neural network model for training to obtain a dynamic positioning prediction model.

2. The method for dynamic positioning of an indoor mobile pollution source according to claim 1, wherein: The convection-diffusion equation simulated in real time using fluid dynamics is: ； in, express time simulated concentration at the location; Indicates the flow field where the pollution source is located; represents the concentration diffusion coefficient of the pollutant; is the source term, indicating that the pollutant Location, in Instantaneous emission at any moment.

3. The method for dynamic positioning of indoor mobile pollution sources according to claim 2, wherein: When the pollutant release rate is constant and the pollution source is a point source: ； in, Dirac Function, which means that only when the coordinates are The release rate at position source of pollution.

4. The method for dynamic positioning of indoor mobile pollution sources according to claim 1, wherein: The method of generating multiple derivative images is to perform rotation step processing on each sub-image.

5. The method for dynamic positioning of indoor mobile pollution sources according to claim 1, wherein: The pollutant concentration field distribution sequence collected at set time intervals is input into the dynamic positioning prediction model to obtain the dynamic position coordinates of multiple mobile pollution sources in sequence; the dynamic position coordinates of the multiple mobile pollution sources obtained in sequence are combined to obtain the dynamic moving path of the mobile pollution source.

6. The method for dynamic positioning of indoor mobile pollution sources according to claim 1, characterized in that: The performance of the dynamic positioning prediction model is improved by using the coefficient of determination, the weighted accuracy based on probability density, and the accuracy of the Euclidean distance between the coordinate positions predicted by the dynamic positioning prediction model and the actual coordinate positions.

7. The method for dynamic positioning of indoor mobile pollution sources according to claim 1, characterized in that: The grayscale image is divided into a plurality of sub-images based on a grid cutting formula.

8. A method for dynamic positioning of indoor mobile pollution sources according to claim 7, characterized in that: The grid cutting formula is: ； in, Indicates the spatial size, used to dynamically adjust the cropping ratio .

9. A dynamic positioning system for indoor mobile pollution sources, comprising a memory, a processor, and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

Images