AMC concentration information extraction method and system based on multi-channel spatiotemporal slice fusion

Abstract

The present invention is applicable to the field of air data monitoring technology, and in particular relates to an AMC concentration information extraction method and system with multi-channel spatiotemporal slice fusion. The method comprises receiving gas concentrations collected by an AMC monitor and constructing a spatiotemporal data cube; selecting a time point in the spatiotemporal data cube, reading a time complementarity cube in the spatiotemporal data cube based on the selected time point, and determining spatial features based on the time complementarity cube; selecting a spatial position in the spatiotemporal data cube based on the spatial features, reading a position complementarity cube based on the selected spatial position, and determining concentration change information of various types of gases using the position complementarity cube; marking gases according to their concentration change information, querying preset characteristic convolution kernels of the gases, traversing the spatiotemporal data cube, matching abnormal times and abnormal positions, and feeding back to a management end. The present invention can help staff quickly locate key data, has high data integration, and fast response speed.

Description

Technical Field

[0001] The present invention relates to the field of air data monitoring technology, and in particular to a method and system for extracting AMC concentration information by fusion of multi-channel spatiotemporal slices. Background Art

[0002] AMC (Airborne Molecular Contaminants) refers to air pollutants that exist in the air in gaseous or molecular form and have adverse effects on the environment, human health, equipment, or industrial processes. AMC concentration monitoring is essential in scenarios where extremely high precision is required (such as clean rooms and chip processing workshops).

[0003] With the development of AMC monitoring equipment, there are many types of AMC monitoring equipment in the existing technology, which have multi-channel and multi-position monitoring functions. Accordingly, when the number of AMC monitoring devices is large, the data obtained is very complicated and the information extraction process is very difficult. Therefore, how to process the AMC concentration information of multiple channels, multiple positions and multiple times to improve the efficiency of information acquisition and reduce the data processing pressure of personnel is the technical problem that the technical solution of the present invention aims to solve. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for extracting AMC concentration information by fusion of multi-channel spatiotemporal slices to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for extracting AMC concentration information by fusion of multi-channel spatiotemporal slices, the method comprising:

[0007] Receive the gas concentration with time, location and type collected by the AMC monitor and construct a spatiotemporal data cube; the total number of elements in the spatiotemporal data cube is: Where, is the total number of elements, is the total number of gas types to be monitored, is the number of time steps, is the spatial grid size;

[0008] Selecting a time point in the spatiotemporal data cube, reading a time complement subcube in the spatiotemporal data cube based on the selected time point, and determining a spatial feature based on the time complement subcube;

[0009] Selecting a spatial position in the spatiotemporal data cube based on the spatial characteristics, reading a position complement subcube based on the selected spatial position, and determining concentration change information of each type of gas from the position complement subcube;

[0010] The gas is marked according to the concentration change information of the gas, the preset characteristic convolution kernel of the gas is queried, the spatiotemporal data cube is traversed, the abnormal time and abnormal location are matched, and the feedback is sent to the management end.

[0011] Furthermore, the step of receiving the gas concentration data including time, location and type collected by the AMC monitor and constructing a spatiotemporal data cube includes:

[0012] Obtain the position of the monitor at each moment according to the position collector built into the AMC monitor;

[0013] Regularly receive the gas concentration including time and type obtained by AMC monitor;

[0014] Construct a spatiotemporal data cube based on the statistics of the concentration of each type of gas at each location over time;

[0015] Among them, when obtaining the position, the position change rate is calculated in real time, and the position acquisition frequency is adjusted according to the change rate; when obtaining the gas concentration, the concentration change rate is calculated in real time, and the acquisition frequency of each type of gas concentration is adjusted based on the concentration change rate and the benchmark change rate determined by the type.

[0016] Furthermore, the step of calculating the concentration of each type of gas at each location according to time and constructing a spatiotemporal data cube includes:

[0017] Receive the time step set by the administrator and determine the time dimension within the preset time range;

[0018] Receive a location calibration grid set by an administrator and determine the coordinate dimensions of each coordinate; the grid cell size in the location calibration grid is a preset value; and the coordinates do not exceed three parameters;

[0019] Receive the total number of types set by the administrator as the type dimension;

[0020] Count the time dimension, coordinate dimension and type dimension to build a cube structure;

[0021] Based on the cube structure, the gas concentration containing time, location and type is counted to obtain the transit data cube;

[0022] Based on the preset interpolation algorithm, interpolation is performed in the transit data cube to fill in the empty data positions and obtain the spatiotemporal data cube.

[0023] Furthermore, the steps of selecting a time point in the spatiotemporal data cube, reading a time complement subcube in the spatiotemporal data cube based on the selected time point, and determining the spatial feature based on the time complement subcube include:

[0024] Extract the time parameter and query the sub-cube corresponding to each time point in the time parameter;

[0025] Input the sub-cube into the preset numerical normalization model and convert it into eigenvalues;

[0026] Arrange the eigenvalues ​​in time sequence, fit the eigenvalues, and obtain the time variation function;

[0027] Calculate the derivative of a preset order of a time-varying function, and query the function value and derivative value at a time point in the time-varying function and all its derivatives;

[0028] Clustering the time points based on the function value and the derivative value, and selecting a class of time points with the largest number of time points as the selected time points;

[0029] Read the sub-cube corresponding to the selected time point as the time complementary sub-cube;

[0030] Gradient identification is performed on each temporal subcube to determine the spatial characteristics.

[0031] Furthermore, the step of performing gradient identification on each temporal subcube to determine spatial features includes:

[0032] Extract each type of data in each time sub-cube in turn to obtain the data distribution matrix;

[0033] Perform gradient identification on the data distribution matrix to determine the spatial contour;

[0034] Count all types of spatial profiles at all time points, merge the spatial profiles to obtain a comprehensive profile; the merging process determines the retention degree based on the number of spatial profiles;

[0035] The comprehensive profile whose retention reaches the preset retention threshold is selected as the spatial feature.

[0036] Furthermore, the steps of selecting a spatial position in the spatiotemporal data cube based on the spatial feature, reading a position complement cube based on the selected spatial position, and determining concentration change information of each type of gas from the position complement cube include:

[0037] Read the comprehensive outline of the spatial features and divide the area into sub-areas;

[0038] The number of sub-regions is determined based on the area of ​​each sub-region, and the number of spatial locations is randomly selected in the sub-region; the number is proportional to the area;

[0039] Query the concentration of each type of gas at each time at each spatial position as a position co-cube; the position co-cube is a two-dimensional matrix;

[0040] Extract the gas concentration of each type of gas at all times in the position co-cube, input it into the preset fluctuation recognition model, and obtain the fluctuation degree of this type;

[0041] All types of fluctuations are counted as concentration change information at that spatial location.

[0042] Furthermore, the steps of marking the gas according to the gas concentration change information, querying the preset characteristic convolution kernel of the gas, traversing the spatiotemporal data cube, matching the abnormal time and abnormal location, and feeding back to the management end include:

[0043] Query the concentration change information of the selected spatial location and mark the type whose fluctuation reaches the preset fluctuation threshold;

[0044] Record the total number of tags of each type. When the total number of tags reaches the preset number, query the feature convolution kernel corresponding to the type.

[0045] Traversing the spatiotemporal data cube based on the feature convolution kernel, matching abnormal time and abnormal location, and feeding back to the management end;

[0046] Among them, the characteristic convolution kernel is the gas concentration of this type of gas over a period of time under preset dangerous conditions.

[0047] The technical solution of the present invention also provides an AMC concentration information extraction system based on multi-channel spatiotemporal slice fusion, the system comprising:

[0048] The cube building module is used to receive the gas concentration data including time, location and type collected by the AMC monitor and build a spatiotemporal data cube; the total number of elements in the spatiotemporal data cube is: Where, is the total number of elements, is the total number of gas types to be monitored, is the number of time steps, is the spatial grid size;

[0049] A spatial feature determination module is used to select a time point in the spatiotemporal data cube, read a time complement subcube in the spatiotemporal data cube based on the selected time point, and determine the spatial feature based on the time complement subcube;

[0050] The gas concentration analysis module is used to select a spatial position in the spatiotemporal data cube based on spatial characteristics, read the position complement cube based on the selected spatial position, and determine the concentration change information of each type of gas from the position complement cube;

[0051] The anomaly matching module is used to mark the gas according to the concentration change information of the gas, query the preset characteristic convolution kernel of the gas, traverse the spatiotemporal data cube, match the abnormal time and abnormal location, and feed back to the management end.

[0052] Furthermore, the cube building module includes:

[0053] A position acquisition unit, used to obtain the position of the monitor at each moment according to a position collector built into the AMC monitor;

[0054] A gas concentration receiving unit is used to periodically receive the gas concentration including time and type obtained by the AMC monitor;

[0055] A data statistics unit is used to count the concentrations of various types of gases at various locations according to time and construct a spatiotemporal data cube;

[0056] Among them, when obtaining the position, the position change rate is calculated in real time, and the position acquisition frequency is adjusted according to the change rate; when obtaining the gas concentration, the concentration change rate is calculated in real time, and the acquisition frequency of each type of gas concentration is adjusted based on the concentration change rate and the benchmark change rate determined by the type.

[0057] Furthermore, the spatial feature determination module includes:

[0058] The data query unit is used to extract the time parameter and query the sub-cube corresponding to each time point in the time parameter;

[0059] The data normalization unit is used to input the sub-cube into a preset numerical normalization model and convert it into a eigenvalue;

[0060] A feature fitting unit is used to arrange the feature values ​​according to the time sequence, fit the feature values, and obtain the time variation function;

[0061] A numerical calculation unit, used to calculate a derivative function of a preset order of a time-varying function, and query a function value and a derivative value at a time point in the time-varying function and all its derivative functions;

[0062] a time point classification unit, configured to cluster the time points based on the function value and the derivative value, and select a class of time points with the largest number of time points as the selected time points;

[0063] A data output unit is used to read the sub-cube corresponding to the selected time point as a time complementary sub-cube;

[0064] The gradient recognition unit is used to perform gradient recognition on each time sub-cube and determine the spatial characteristics.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] The present invention collects monitoring data from AMC monitors, constructs a spatiotemporal data cube, extracts data segments under different parameters from the spatiotemporal data cube, analyzes the data segments, selects important locations, and simultaneously determines gas changes at the important locations to further determine the risk gas types. The convolutional recognition model is then used to match abnormal locations and times corresponding to the risk gas types, helping staff to quickly locate key data. The method has extremely high data integration and extremely fast response speed, which helps to improve data acquisition efficiency and reduce the difficulty for staff to obtain data. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention.

[0068] Figure 1 This is a flowchart of the overall process of the AMC concentration information extraction method based on multi-channel spatiotemporal slice fusion provided by an embodiment of the present invention.

[0069] Figure 2 This is a structural diagram of the AMC concentration information extraction system with multi-channel spatiotemporal slice fusion provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0070] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below with reference to 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.

[0071] Figure 1 The overall flow chart of the AMC concentration information extraction method based on multi-channel spatiotemporal slice fusion is shown. In an embodiment of the present invention, a method for extracting AMC concentration information based on multi-channel spatiotemporal slice fusion is provided, the method comprising:

[0072] Step S100: receiving the gas concentration including time, location and type collected by the AMC monitor, and constructing a spatiotemporal data cube;

[0073] The AMC monitor is a device used to collect gas pollutant concentrations. It can be used to collect gas concentrations of various gas pollutants. Its installation location is not fixed and can be installed on some mobile devices or some fixed facilities. When the AMC monitor obtains gas concentrations, it records the time, location, and type, indicating the gas concentration of what type of gas at what time and place. The gas concentrations obtained within a preset time range (such as a day or a week) are counted and inserted into a multidimensional array called a spatiotemporal data cube. The total number of elements in the spatiotemporal data cube is: Where, is the total number of elements, is the total number of gas types to be monitored, is the number of time steps, is the spatial grid size.

[0074] Step S200: selecting a time point in the spatiotemporal data cube, reading a time complement sub-cube in the spatiotemporal data cube based on the selected time point, and determining a spatial feature based on the time complement sub-cube;

[0075] In the spatiotemporal data cube, time is used as a parameter. After removing the time parameter, the remaining data is still a multidimensional data, called a remainder cube. Since it is the remainder cube after removing the time parameter, it is called a time remainder cube. There are many time points in the time parameter. The technical solution of the present invention does not need to analyze each time point. It only needs to select the time remainder cubes of some time points, identify the time remainder cubes, and then determine the spatial characteristics.

[0076] Specifically, regarding the concept of cube in the above content, the following examples are given:

[0077] Assume there are three types of gas data, and the collection location of the gas data is three-dimensional. Then the data of the spatiotemporal data cube is A[t][x][y][z][a][b][c], where t represents time, x, y, and z represent three coordinates, and a, b, and c correspond to three types respectively. This is a seven-dimensional array. After extracting the time parameter, A[x][y][z][a][b][c] is obtained, which is a six-dimensional data and a co-subcube of the seven-dimensional array. For details, please refer to the concept of co-sub-formula.

[0078] By analyzing A[x][y][z][a][b][c] and extracting the state of the gas at different coordinates, we can obtain spatial features. In layman's terms, spatial features are some boundaries, such as the location where the gas concentration suddenly changes.

[0079] Step S300: selecting a spatial position in the spatiotemporal data cube based on the spatial feature, reading a position complement subcube based on the selected spatial position, and determining concentration change information of each type of gas from the position complement subcube;

[0080] After the spatial features are determined, some relatively important locations are selected in the entire area. Then, based on these locations, the remainder cubes are extracted. Taking the above example as an example, A[t][a][b][c] is extracted. By analyzing these data, the changes in various types of gases can be obtained, that is, the concentration change information of various types of gases.

[0081] Step S400: Marking the gas according to the concentration change information of the gas, querying the preset characteristic convolution kernel of the gas, traversing the spatiotemporal data cube, matching the abnormal time and abnormal location, and feeding back to the management end;

[0082] After obtaining the gas changes at important locations, the gas with more abnormal changes is selected and its type is queried. If it is considered that this type requires further detection, the preset feature convolution kernel of the gas is queried and the entire spatiotemporal data cube is traversed. If a match is successful, the abnormal time and abnormal location are recorded. The abnormal time is a time period, and the abnormal location is generally a range. The abnormal time and abnormal location are fed back to the management end.

[0083] Among them, the characteristic convolution kernel is preset. Before the application of the technical solution of the present invention, the staff will pre-simulate some dangerous conditions (the time range and spatial range are generally small), and record the gas concentration in the same way to obtain the spatiotemporal data cube under the dangerous state. Since the time range and spatial range are generally small, the range of each parameter of the simulated spatiotemporal data cube is very small, which is called the characteristic convolution kernel, which can be compared to the image convolution kernel. For example, in a 1920*1080 image, the convolution kernel can use a 16*16 convolution kernel. The image is two-dimensional, and the spatiotemporal data cube in the technical solution of the present invention is multi-dimensional.

[0084] Regarding step S100, the step of receiving the gas concentration including time, location and type collected by the AMC monitor and constructing a spatiotemporal data cube includes:

[0085] Obtain the position of the monitor at each moment according to the position collector built into the AMC monitor;

[0086] Regularly receive the gas concentration including time and type obtained by AMC monitor;

[0087] The concentration of each type of gas at each location is statistically analyzed over time to construct a spatiotemporal data cube.

[0088] The AMC monitor has a built-in locator (position collector) for obtaining the position at each moment. The AMC monitor has a built-in acquisition module for obtaining the gas concentration of different types of gases. The AMC monitor also has a clock module (basic module) that can record time. Thus, the position at different moments and the concentration of different types of gases can be obtained. The concentration of each type of gas at each position is statistically calculated based on time to construct a spatiotemporal data cube.

[0089] It is worth mentioning that when obtaining the position, the position change rate is calculated in real time, and the position acquisition frequency is adjusted according to the change rate; the position acquisition frequency is proportional to the position change rate. When the AMC monitor is installed on a fixed facility, the position change rate is zero, and the position acquisition frequency takes the minimum value; similar to the position, when obtaining the gas concentration, the concentration change rate is calculated in real time, and the gas acquisition frequency is also proportional to the concentration change rate. When the gas is stable, the acquisition frequency is low, but the difference is that the importance of different gases is different, and the benchmark change rate is used to represent it. The benchmark change rate is related to the importance of the gas. The more important the gas, the higher the benchmark change rate, and accordingly, the acquisition frequency will also be higher.

[0090] Specifically, the step of statistically analyzing the concentration of each type of gas at each location over time to construct a spatiotemporal data cube includes:

[0091] Receive the time step set by the administrator and determine the time dimension within the preset time range;

[0092] Receive a location calibration grid set by an administrator and determine the coordinate dimensions of each coordinate; the grid cell size in the location calibration grid is a preset value; and the coordinates do not exceed three parameters;

[0093] Receive the total number of types set by the administrator as the type dimension;

[0094] Count the time dimension, coordinate dimension and type dimension to build a cube structure;

[0095] Based on the cube structure, the gas concentration containing time, location and type is counted to obtain the transit data cube;

[0096] Based on the preset interpolation algorithm, interpolation is performed in the transit data cube to fill in the empty data positions and obtain the spatiotemporal data cube.

[0097] In an example of the technical solution of the present invention, the data statistics process is explained, and the time step set by the receiving manager, that is, how long is the statistics interval, such as 10 seconds, is used to determine the time dimension within the preset time range, such as how many time points are obtained in a day; the position calibration grid set by the manager is received and inserted into the area to be detected. The position calibration grid indicates how many positions are selected in the area to be detected. The grid unit size in the position calibration grid is a preset value. The smaller the grid unit size, the more positions are selected. In actual applications, the position calibration grid can be two-dimensional or three-dimensional. If the gas concentration is obtained only in a plane, it is two-dimensional. If the gas concentration is obtained in space, it is three-dimensional.

[0098] The time dimension, coordinate dimension and type dimension are counted to construct a cube structure. The gas concentration containing time, position and type is counted based on the cube structure. Since the position and acquisition frequency of the AMC detector are changing, the acquisition of gas concentration is discontinuous both in time and space. This makes only certain data in the cube structure real data, which is called the transit data cube. Interpolation is performed in the transit data cube based on the preset interpolation algorithm to fill in the empty data positions and obtain the spatiotemporal data cube.

[0099] Regarding the method of filling in the empty data positions, it can use the conventional interpolation process. For a certain type of gas, the gas concentration at each moment can be spatially interpolated, and each empty data can be set as the nearest known data. Of course, it is also possible to introduce a diffusion model to predict it; the gas concentration at each position can also be temporally interpolated, the gas concentration at a known moment is queried, the gas concentration function is fitted, and then the gas concentration at an unknown moment is determined.

[0100] Regarding step S200, the steps of selecting a time point in the spatiotemporal data cube, reading a time complement sub-cube in the spatiotemporal data cube based on the selected time point, and determining spatial features based on the time complement sub-cube include:

[0101] Extract the time parameter and query the sub-cube corresponding to each time point in the time parameter;

[0102] Input the sub-cube into the preset numerical normalization model and convert it into eigenvalues;

[0103] Arrange the eigenvalues ​​in time sequence, fit the eigenvalues, and obtain the time variation function;

[0104] Calculate the derivative of a preset order of a time-varying function, and query the function value and derivative value at a time point in the time-varying function and all its derivatives;

[0105] Clustering the time points based on the function value and the derivative value, and selecting a class of time points with the largest number of time points as the selected time points;

[0106] Read the sub-cube corresponding to the selected time point as the time complementary sub-cube;

[0107] Gradient identification is performed on each temporal subcube to determine the spatial characteristics.

[0108] In one embodiment of the technical solution of the present invention, a specific data screening solution is provided, which focuses on selecting important time points from all time points within a period. The specific operations are as follows:

[0109] Extract the time parameter, query the sub-cube corresponding to each time point in the time parameter, input the sub-cube into the preset numerical normalization model, and convert it into an eigenvalue. The numerical normalization model is used to convert a multidimensional array into a value. First, all data are dimensionlessly processed and then summed. It has no practical significance, and the specific normalization method is not limited here. Since each time point corresponds to a value, the eigenvalues ​​are arranged in time order, and the eigenvalues ​​are fitted to obtain a time-varying function. The derivative function of the preset order of the time-varying function is calculated, and the function value and derivative value of the time point are queried in the time-varying function and all its derivative functions. The order of the derivative function is at least one and at most three. More is not very meaningful and will make the subsequent clustering results more diverse. Based on the function value and derivative value, the time points with sufficiently similar states can be classified into one category. The time points in the category with the largest number of time points are selected as the selected time points, and the sub-cube corresponding to the selected time points are read as the time complement sub-cube.

[0110] The time influence of the sub-cube selected through the above content is extremely low. When used to identify spatial features, the influence of time is very small; gradient recognition is performed on each time sub-cube to obtain spatial features.

[0111] As an example of the technical solution of the present invention, the step of performing gradient identification on each temporal subcube and determining spatial features includes:

[0112] Extract each type of data in each time sub-cube in turn to obtain the data distribution matrix;

[0113] Perform gradient identification on the data distribution matrix to determine the spatial contour;

[0114] Count all types of spatial profiles at all time points, merge the spatial profiles to obtain a comprehensive profile; the merging process determines the retention degree based on the number of spatial profiles;

[0115] The comprehensive profile whose retention reaches the preset retention threshold is selected as the spatial feature.

[0116] Taking the above example as the standard, each type of data is extracted in turn in each time subcube, and the data obtained is A[t][x][y][z][a], where t and a are known and are constants. It is actually a[x][y][z], that is, the gas concentration of type a gas at different positions, which is called the data distribution matrix. A data distribution matrix can be obtained for each type at each time point.

[0117] Perform gradient identification on the data distribution matrix to determine the spatial contour; count all types of spatial contours at all time points, merge the spatial contours, and obtain a comprehensive contour; the merging process is to merge sufficiently close contours into one contour, record how many data distribution matrices each comprehensive contour is obtained by merging, and determine the retention degree. The larger the number, the higher the retention degree. Eliminate comprehensive contours with less retention (such as spatial contours that only appear in a few data distribution matrices), and select comprehensive contours with a retention degree that reaches a preset retention threshold as spatial features.

[0118] Regarding step S300, the steps of selecting a spatial position in the spatiotemporal data cube based on the spatial feature, reading a position complement cube based on the selected spatial position, and determining concentration change information of each type of gas from the position complement cube include:

[0119] Read the comprehensive outline of the spatial features and divide the area into sub-areas;

[0120] The number of sub-regions is determined based on the area of ​​each sub-region, and the number of spatial locations is randomly selected in the sub-region; the number is proportional to the area;

[0121] Query the concentration of each type of gas at each time at each spatial position as a position co-cube; the position co-cube is a two-dimensional matrix;

[0122] Extract the gas concentration of each type of gas at all times in the position co-cube, input it into the preset fluctuation recognition model, and obtain the fluctuation degree of this type;

[0123] All types of fluctuations are counted as concentration change information at that spatial location.

[0124] In an example of the technical solution of the present invention, the comprehensive outline in the spatial feature is read, the area is divided into sub-areas, and the number of sub-areas is determined according to the area of ​​each sub-area. The larger the area, the greater the number. The same number of spatial positions is randomly selected in the sub-area, and the concentration of each type of gas at each spatial position at each time is queried as the position residual cube; based on the above example, the concentration of each type of gas at each spatial position at each time is A[t][a], A[t][b] and A[t][c], which is actually to obtain a[t], b[t] and c[t], which are some time series. They are input into a preset fluctuation recognition model to obtain the fluctuation degree of this type. The fluctuation recognition model can use an existing fluctuation recognition model, which is generally related to variance and periodicity. All types of fluctuations are counted as the concentration change information of the spatial position; in actual applications, there are far more than three types of gases to be monitored, so there will be many fluctuations.

[0125] Regarding step S400, the steps of marking the gas according to the gas concentration change information, querying the preset characteristic convolution kernel of the gas, traversing the spatiotemporal data cube, matching the abnormal time and abnormal location, and feeding back to the management end include:

[0126] Query the concentration change information of the selected spatial location and mark the type whose fluctuation reaches the preset fluctuation threshold;

[0127] Record the total number of tags of each type. When the total number of tags reaches the preset number, query the feature convolution kernel corresponding to the type.

[0128] The spatiotemporal data cube is traversed based on the feature convolution kernel, abnormal time and abnormal location are matched, and feedback is given to the management end.

[0129] In an example of the technical solution of the present invention, each spatial position is analyzed separately, all types of fluctuations at each spatial position are queried, and the types whose fluctuations reach a preset fluctuation threshold are marked. Each spatial position is analyzed in turn, and the total number of marks of each type is recorded. When the total number of marks is large enough, it means that this type of gas is abnormal at multiple locations in the entire area. At this time, the feature convolution kernel corresponding to this type is queried; based on the feature convolution kernel, the spatiotemporal data cube is traversed, the abnormal time and abnormal position are matched, and the feedback is fed back to the management end; combined with the description of the feature convolution kernel in the above content, the feature convolution kernel is the gas concentration of this type of gas over a period of time under a preset dangerous situation.

[0130] Combine Figure 2 The technical solution of the present invention also provides an AMC concentration information extraction system based on multi-channel spatiotemporal slice fusion, wherein the system 10 comprises:

[0131] The cube construction module 11 is used to receive the gas concentration data including time, location and type collected by the AMC monitor and construct a spatiotemporal data cube; the total number of elements in the spatiotemporal data cube is: Where, is the total number of elements, is the total number of gas types to be monitored, is the number of time steps, is the spatial grid size;

[0132] A spatial feature determination module 12 is configured to select a time point in the spatiotemporal data cube, read a time complement subcube in the spatiotemporal data cube based on the selected time point, and determine a spatial feature based on the time complement subcube;

[0133] The gas concentration analysis module 13 is used to select a spatial position in the spatiotemporal data cube based on the spatial characteristics, read the position complement subcube based on the selected spatial position, and determine the concentration change information of each type of gas from the position complement subcube;

[0134] The abnormal matching module 14 is used to mark the gas according to the concentration change information of the gas, query the preset characteristic convolution kernel of the gas, traverse the spatiotemporal data cube, match the abnormal time and abnormal location, and feed back to the management end.

[0135] Specifically, the cube building module 11 includes:

[0136] A position acquisition unit, used to obtain the position of the monitor at each moment according to a position collector built into the AMC monitor;

[0137] A gas concentration receiving unit is used to periodically receive the gas concentration including time and type obtained by the AMC monitor;

[0138] A data statistics unit is used to count the concentrations of various types of gases at various locations according to time and construct a spatiotemporal data cube;

[0139] Among them, when obtaining the position, the position change rate is calculated in real time, and the position acquisition frequency is adjusted according to the change rate; when obtaining the gas concentration, the concentration change rate is calculated in real time, and the acquisition frequency of each type of gas concentration is adjusted based on the concentration change rate and the benchmark change rate determined by the type.

[0140] Specifically, the spatial feature determination module 12 includes:

[0141] The data query unit is used to extract the time parameter and query the sub-cube corresponding to each time point in the time parameter;

[0142] The data normalization unit is used to input the sub-cube into a preset numerical normalization model and convert it into a eigenvalue;

[0143] A feature fitting unit is used to arrange the feature values ​​according to the time sequence, fit the feature values, and obtain the time variation function;

[0144] A numerical calculation unit, used to calculate a derivative function of a preset order of a time-varying function, and query a function value and a derivative value at a time point in the time-varying function and all its derivative functions;

[0145] a time point classification unit, configured to cluster the time points based on the function value and the derivative value, and select a class of time points with the largest number of time points as the selected time points;

[0146] A data output unit is used to read the sub-cube corresponding to the selected time point as a time complementary sub-cube;

[0147] The gradient recognition unit is used to perform gradient recognition on each time sub-cube and determine the spatial characteristics.

[0148] The above are only preferred embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A multi-channel spatiotemporal slice fusion AMC concentration information extraction method, characterized in that: The method comprises: Receive the gas concentration with time, location and type collected by the AMC monitor and construct a spatiotemporal data cube; the total number of elements in the spatiotemporal data cube is: Where, is the total number of elements, is the total number of gas types to be monitored, is the number of time steps, is the spatial grid size; Selecting a time point in the spatiotemporal data cube, reading a time complement subcube in the spatiotemporal data cube based on the selected time point, and determining a spatial feature based on the time complement subcube; Selecting a spatial position in the spatiotemporal data cube based on the spatial characteristics, reading a position complement subcube based on the selected spatial position, and determining concentration change information of each type of gas from the position complement subcube; Mark the gas according to its concentration change information, query the preset characteristic convolution kernel of the gas, traverse the spatiotemporal data cube, match the abnormal time and abnormal location, and feedback to the management end; The steps of selecting a time point in the spatiotemporal data cube, reading a time complement subcube in the spatiotemporal data cube based on the selected time point, and determining the spatial feature based on the time complement subcube include: Extract the time parameter and query the sub-cube corresponding to each time point in the time parameter; Input the sub-cube into the preset numerical normalization model and convert it into eigenvalues; Arrange the eigenvalues ​​in time sequence, fit the eigenvalues, and obtain the time variation function; Calculate the derivative of a preset order of a time-varying function, and query the function value and derivative value at a time point in the time-varying function and all its derivatives; Clustering the time points based on the function value and the derivative value, and selecting a class of time points with the largest number of time points as the selected time points; Read the sub-cube corresponding to the selected time point as the time complementary sub-cube; Perform gradient identification on each temporal subcube to determine spatial features; The steps of selecting a spatial position in the spatiotemporal data cube based on the spatial feature, reading a position complement cube based on the selected spatial position, and determining concentration change information of each type of gas from the position complement cube include: Read the comprehensive outline of the spatial features and divide the area into sub-areas; The number of sub-regions is determined based on the area of ​​each sub-region, and the number of spatial locations is randomly selected in the sub-region; the number is proportional to the area; Query the concentration of each type of gas at each time at each spatial position as a position co-cube; the position co-cube is a two-dimensional matrix; Extract the gas concentration of each type of gas at all times in the position co-cube, input it into the preset fluctuation recognition model, and obtain the fluctuation degree of this type; All types of fluctuations are counted as concentration change information at that spatial location.

2. The AMC concentration information extraction method based on multi-channel spatiotemporal slice fusion according to claim 1 is characterized in that: The step of receiving the gas concentration including time, location and type collected by the AMC monitor and constructing a spatiotemporal data cube includes: Obtain the position of the monitor at each moment according to the position collector built into the AMC monitor; Regularly receive the gas concentration including time and type obtained by AMC monitor; Construct a spatiotemporal data cube based on the statistics of the concentration of each type of gas at each location over time; Among them, when obtaining the position, the position change rate is calculated in real time, and the position acquisition frequency is adjusted according to the change rate; when obtaining the gas concentration, the concentration change rate is calculated in real time, and the acquisition frequency of each type of gas concentration is adjusted based on the concentration change rate and the benchmark change rate determined by the type.

3. The AMC concentration information extraction method based on multi-channel spatiotemporal slice fusion according to claim 2 is characterized in that: The step of constructing a spatiotemporal data cube by counting the concentrations of various types of gases at various locations according to time includes: Receive the time step set by the administrator and determine the time dimension within the preset time range; Receive a location calibration grid set by an administrator and determine the coordinate dimensions of each coordinate; the grid cell size in the location calibration grid is a preset value; and the coordinates do not exceed three parameters; Receive the total number of types set by the administrator as the type dimension; Count the time dimension, coordinate dimension and type dimension to build a cube structure; Based on the cube structure, the gas concentration containing time, location and type is counted to obtain the transit data cube; Based on the preset interpolation algorithm, interpolation is performed in the transit data cube to fill in the empty data positions and obtain the spatiotemporal data cube.

4. The AMC concentration information extraction method based on multi-channel spatiotemporal slice fusion according to claim 1 is characterized in that: The step of performing gradient identification on each temporal sub-cube to determine spatial features includes: Extract each type of data in each time sub-cube in turn to obtain the data distribution matrix; Perform gradient identification on the data distribution matrix to determine the spatial contour; Count all types of spatial profiles at all time points, merge the spatial profiles to obtain a comprehensive profile; the merging process determines the retention degree based on the number of spatial profiles; The comprehensive profile whose retention reaches the preset retention threshold is selected as the spatial feature.

5. The AMC concentration information extraction method based on multi-channel spatiotemporal slice fusion according to claim 1 is characterized in that: The steps of marking gas according to gas concentration change information, querying a preset characteristic convolution kernel of the gas, traversing the spatiotemporal data cube, matching abnormal time and abnormal location, and feeding back to the management end include: Query the concentration change information of the selected spatial location and mark the type whose fluctuation reaches the preset fluctuation threshold; Record the total number of tags of each type. When the total number of tags reaches the preset number, query the feature convolution kernel corresponding to the type. Traversing the spatiotemporal data cube based on the feature convolution kernel, matching abnormal time and abnormal location, and feeding back to the management end; Among them, the characteristic convolution kernel is the gas concentration of this type of gas over a period of time under preset dangerous conditions.

6. A multi-channel spatiotemporal slice fusion AMC concentration information extraction system, characterized by: The system comprises: The cube building module is used to receive the gas concentration data including time, location and type collected by the AMC monitor and build a spatiotemporal data cube; the total number of elements in the spatiotemporal data cube is: Where, is the total number of elements, is the total number of gas types to be monitored, is the number of time steps, is the spatial grid size; A spatial feature determination module is used to select a time point in the spatiotemporal data cube, read a time complement subcube in the spatiotemporal data cube based on the selected time point, and determine the spatial feature based on the time complement subcube; The gas concentration analysis module is used to select a spatial position in the spatiotemporal data cube based on spatial characteristics, read the position complement cube based on the selected spatial position, and determine the concentration change information of each type of gas from the position complement cube; An anomaly matching module is used to mark the gas according to the gas concentration change information, query the preset characteristic convolution kernel of the gas, traverse the spatiotemporal data cube, match the abnormal time and abnormal location, and feedback to the management end; The spatial feature determination module includes: The data query unit is used to extract the time parameter and query the sub-cube corresponding to each time point in the time parameter; The data normalization unit is used to input the sub-cube into a preset numerical normalization model and convert it into a eigenvalue; A feature fitting unit is used to arrange the feature values ​​according to the time sequence, fit the feature values, and obtain the time variation function; A numerical calculation unit, used to calculate a derivative function of a preset order of a time-varying function, and query a function value and a derivative value at a time point in the time-varying function and all its derivative functions; a time point classification unit, configured to cluster the time points based on the function value and the derivative value, and select a class of time points with the largest number of time points as the selected time points; A data output unit is used to read the sub-cube corresponding to the selected time point as a time complementary sub-cube; A gradient recognition unit is used to perform gradient recognition on each temporal subcube to determine spatial features; The selecting of a spatial position in the spatiotemporal data cube based on the spatial feature, reading of a position complement cube based on the selected spatial position, and determination of concentration change information of each type of gas in the position complement cube include: Read the comprehensive outline of the spatial features and divide the area into sub-areas; The number of sub-regions is determined based on the area of ​​each sub-region, and the number of spatial locations is randomly selected in the sub-region; the number is proportional to the area; Query the concentration of each type of gas at each time at each spatial position as a position co-cube; the position co-cube is a two-dimensional matrix; Extract the gas concentration of each type of gas at all times in the position co-cube, input it into the preset fluctuation recognition model, and obtain the fluctuation degree of this type; All types of fluctuations are counted as concentration change information at that spatial location.

7. The AMC concentration information extraction system based on multi-channel spatiotemporal slice fusion according to claim 6 is characterized in that: The cube building blocks include: A position acquisition unit, used to obtain the position of the monitor at each moment according to a position collector built into the AMC monitor; A gas concentration receiving unit is used to periodically receive the gas concentration including time and type obtained by the AMC monitor; A data statistics unit is used to count the concentrations of various types of gases at various locations according to time and construct a spatiotemporal data cube; Among them, when obtaining the position, the position change rate is calculated in real time, and the position acquisition frequency is adjusted according to the change rate; when obtaining the gas concentration, the concentration change rate is calculated in real time, and the acquisition frequency of each type of gas concentration is adjusted based on the concentration change rate and the benchmark change rate determined by the type.

Images