Transport capacity detection method and device, storage medium and electronic device

Abstract

The present disclosure belongs to the technical field of computers and relates to a transport capacity detection method and device, a storage medium and an electronic device. The method comprises: obtaining a transport capacity data feature map of multi-source transport capacity data, and performing data alignment processing on the transport capacity data feature map to obtain a multi-source feature map; performing feature extraction processing on the multi-source feature map to obtain a transport capacity feature map, and performing dimension unification processing on the transport capacity feature map to obtain a high-level feature map; and performing transport capacity detection processing on the high-level feature map to obtain a transport capacity detection result of the multi-source transport capacity data. The present disclosure is conducive to training a machine learning model for transport capacity detection, facilitates migration to different cities and regions, improves the self-adaptation degree of transport capacity detection and the richness of application scenarios, solves the problem of size difference of corresponding feature maps caused by different multi-source transport capacity data, guarantees the accuracy and real-time performance of the transport capacity detection result, solves the lag problem of transport capacity detection, and can perform real-time estimation of the impact of transport capacity anomalies.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of computer, and particularly relates to a transport capacity detection method and device, a computer readable storage medium and an electronic device. BACKGROUND

[0002] With the continuous expansion of the city size and the growth of the population, the demand for urban living resources is gradually increasing. Among them, when the demand for urban traffic travel resources is increasing, the demand for transport capacity in a certain area will exceed the supply of current time transport capacity, resulting in traffic paralysis. Especially in places like airports, railway stations and commercial areas, there is a high probability of transport capacity shortage.

[0003] In this case, the vehicle dispatching in the area is mostly concentrated, including taxis, buses and shared bicycles, etc. to alleviate, or through crawling network public opinion data for dispatching. However, this dispatching method cannot detect and perceive the situation of abnormal gathering of people and shortage of vehicle transport capacity in the area, and the detection method has a lag, which is difficult to accurately evaluate the impact of the event. Further, since the number of labels of abnormal gathering of people and shortage of vehicle transport capacity is small and difficult to obtain, a large-scale machine learning model cannot be trained for detection, and it is also impossible to migrate to different areas and cities, so the applicability is poor.

[0004] Therefore, there is an urgent need in the art to develop a new transport capacity detection method and device.

[0005] It should be noted that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present disclosure, and therefore can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0006] The purpose of the present disclosure is to provide a transport capacity detection method, a transport capacity detection device, a computer readable storage medium and an electronic device, thereby at least partially overcoming the technical problems of transport capacity detection lag, inaccuracy and low self-adaptation caused by the limitations of related technologies.

[0007] Other characteristics and advantages of the present disclosure will become apparent from the following detailed description, or will be learned by practice of the present disclosure.

[0008] According to a first aspect of the embodiments of the present disclosure, a transport capacity detection method is provided, the method comprising: obtaining a transport capacity data feature map of multi-source transport capacity data, and performing data alignment processing on the transport capacity data feature map to obtain a multi-source feature map;

[0009] The multi-source feature map is subjected to feature extraction processing to obtain a transport capacity feature map, and the transport capacity feature map is subjected to dimension unification processing to obtain a high-level feature map.

[0010] The high-level feature map is subjected to transport capacity detection processing to obtain a transport capacity detection result of the multi-source transport capacity data.

[0011] In an exemplary embodiment of the present application, the data alignment processing of the transport capacity data feature map to obtain a multi-source feature map comprises:

[0012] The transport capacity data feature map is subjected to address coding processing to obtain a plurality of original feature maps, and a standard feature map in the plurality of original feature maps is determined;

[0013] Based on the standard feature map, the feature map alignment processing is performed on the other feature maps in the plurality of original feature maps except the standard feature map to obtain a plurality of aligned feature maps;

[0014] The standard feature map and the plurality of aligned feature maps are subjected to feature map connection processing to obtain a multi-source feature map.

[0015] In an exemplary embodiment of the present application, the dimension unification processing of the transport capacity feature map to obtain a high-level feature map comprises:

[0016] The transport capacity feature map is subjected to feature map division processing to obtain a plurality of pooling regions;

[0017] The plurality of pooling regions are subjected to target value processing to obtain a plurality of target values, and the plurality of target values are subjected to splicing processing to obtain a high-level feature map.

[0018] In an exemplary embodiment of the present application, the transport capacity data feature map of the multi-source transport capacity data comprises:

[0019] A reference region, a proximal region associated with the reference region and a distal region are determined, and reference region data of the reference region, proximal region data of the proximal region and distal region data of the distal region are respectively acquired;

[0020] The reference region data and the proximal region data are subjected to data combination processing to obtain positive example region data, and the reference region data and the distal region data are subjected to data combination processing to obtain negative example region data, so as to determine a transport capacity data feature map of multi-source transport capacity data according to the positive example region data and the negative example region data.

[0021] In an exemplary embodiment of the present application, the transport capacity data feature map of the multi-source transport capacity data comprises:

[0022] determine a reference time period, a same time period and a different time period associated with the reference time period, and obtain reference traffic data of the reference time period, same traffic data of the same time period and different traffic data of the different time period respectively;

[0023] perform data combination processing on the reference traffic data and the same traffic data to obtain positive example time data, and perform data combination processing on the reference traffic data and the different traffic data to obtain negative example time data, so as to determine traffic data features of multi-source traffic data according to the positive example time data and the negative example time data.

[0024] In an exemplary embodiment of the present application, the traffic detection result of the multi-source traffic data obtained by performing traffic detection processing on the high-level feature map comprises:

[0025] perform feature map conversion processing on the two high-level feature maps corresponding to the traffic data feature map to obtain two feature vectors;

[0026] perform training loss calculation on the two feature vectors to obtain a training loss value, and perform traffic detection processing on the high-level feature map according to the training loss value to obtain the traffic detection result of the multi-source traffic data.

[0027] In an exemplary embodiment of the present application, the traffic detection result of the multi-source traffic data obtained by performing traffic detection processing on the high-level feature map according to the training loss value comprises:

[0028] if the training loss value is less than a training loss threshold corresponding to the training loss value, perform traffic detection processing on the high-level feature map to obtain the traffic detection result of the multi-source traffic data;

[0029] if the training loss value is greater than or equal to a training loss threshold corresponding to the training loss value, perform data alignment processing, feature extraction processing and dimension unification processing on the traffic data feature map again to obtain the traffic detection result of the multi-source traffic data.

[0030] In an exemplary embodiment of the present application, the traffic detection result of the multi-source traffic data obtained by performing traffic detection processing on the high-level feature map comprises:

[0031] obtain a reference detection result corresponding to the multi-source traffic data, and perform traffic detection processing on the high-level feature map to obtain a to-be-determined detection result of the multi-source traffic data;

[0032] The inference loss value is calculated by performing inference loss calculation on the reference detection result and the pending detection result, and the transport capacity detection result of the multi-source transport capacity data is determined according to the inference loss value.

[0033] In an example embodiment of the present application, the determination of the transport capacity detection result of the multi-source transport capacity data according to the inference loss value comprises:

[0034] If the inference loss value is less than the inference loss threshold corresponding to the inference loss value, the pending detection result is determined as the transport capacity detection result of the multi-source transport capacity data.

[0035] If the inference loss value is greater than or equal to the inference loss threshold corresponding to the inference loss value, the transport capacity detection result of the multi-source transport capacity data is obtained by performing transport capacity detection processing on the high-level feature map according to the reference detection result.

[0036] According to a second aspect of the embodiments of the present application, a transport capacity detection device is provided, which comprises: a data alignment module configured to obtain a transport capacity data feature map of multi-source transport capacity data, and perform data alignment processing on the transport capacity data feature map to obtain a multi-source feature map;

[0037] a dimension unification module configured to perform feature extraction processing on the multi-source feature map to obtain a transport capacity feature map, and perform dimension unification processing on the transport capacity feature map to obtain a high-level feature map;

[0038] a transport capacity detection module configured to perform transport capacity detection processing on the high-level feature map to obtain the transport capacity detection result of the multi-source transport capacity data.

[0039] According to a third aspect of the embodiments of the present application, an electronic device is provided, which comprises a processor and a memory; wherein the memory has stored thereon computer readable instructions, which, when executed by the processor, implement the transport capacity detection method of any of the example embodiments described above.

[0040] According to a fourth aspect of the embodiments of the present application, a computer readable storage medium is provided, which has stored thereon a computer program, which, when executed by a processor, implements the transport capacity detection method in any of the example embodiments described above.

[0041] From the above technical solutions, it can be seen that the transport capacity detection method, the transport capacity detection device, the computer storage medium and the electronic device in the example embodiments of the present application at least have the following advantages and positive effects:

[0042] In the method and device provided in the exemplary embodiments of the present disclosure, on the one hand, the data alignment processing is performed on the transport capacity data feature map, and in consideration of the fact that the transport capacity detection result is affected by multiple transport capacity data, the data fusion is performed on the multiple transport capacity data to ensure the accuracy of the transport capacity detection in the data aspect, and the richness of the transport capacity data is improved, which is beneficial to train the machine learning model for the transport capacity detection, facilitates the migration to different cities and regions, and improves the self-adaptation degree of the transport capacity detection and the richness of the application scenarios; on the other hand, the dimension unification processing is performed on the transport capacity feature map, and the differences in the reachable areas of different multi-source transport capacity data are considered, the size difference problem of the corresponding feature maps caused by different multi-source transport capacity data is solved, the integration is good, and the accuracy and real-time performance of the transport capacity detection result are ensured, the lag problem of the transport capacity detection is solved, and the influence caused by the transport capacity anomaly can be estimated in real time.

[0043] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0044] The drawings incorporated into the specification and forming a part thereof illustrate embodiments in accordance with the present disclosure and, together with the description, serve to explain the principles of the disclosure. It is clear that the drawings in the following description are only some embodiments of the present disclosure, and other drawings can be obtained by those skilled in the art without creative labor based on these drawings.

[0045] Figure 1 The flowchart schematically shows a transport capacity detection method in the exemplary embodiments of the present disclosure;

[0046] Figure 2 The flowchart schematically shows a method for obtaining a transport capacity data feature map in the exemplary embodiments of the present disclosure;

[0047] Figure 3 The interface diagram schematically shows each region determined in the exemplary embodiments of the present disclosure;

[0048] Figure 4 The flowchart schematically shows another method for obtaining a transport capacity data feature map in the exemplary embodiments of the present disclosure;

[0049] Figure 5 The flowchart schematically shows a method for data alignment processing in the exemplary embodiments of the present disclosure;

[0050] Figure 6 The flowchart schematically shows a method for dimension unification processing in the exemplary embodiments of the present disclosure;

[0051] Figure 7A schematic diagram of a dimension unification process in an example embodiment of the present disclosure is shown.

[0052] Figure 8 A flowchart of a method of a transport capacity detection process in an example embodiment of the present disclosure is shown.

[0053] Figure 9 A model architecture diagram of training a transport capacity detection model in an example embodiment of the present disclosure is shown.

[0054] Figure 10 A flowchart of a method of determining a transport capacity detection result in an example embodiment of the present disclosure is shown.

[0055] Figure 11 A flowchart of a method of further performing a transport capacity detection process in an example embodiment of the present disclosure is shown.

[0056] Figure 12 A flowchart of a method of further determining a transport capacity detection result in an example embodiment of the present disclosure is shown.

[0057] Figure 13 A module diagram of a transport capacity detection model in an application scenario in an example embodiment of the present disclosure is shown.

[0058] Figure 14 A structural diagram of a transport capacity detection apparatus in an example embodiment of the present disclosure is shown.

[0059] Figure 15 An electronic device for implementing a transport capacity detection method in an example embodiment of the present disclosure is shown.

[0060] Figure 16 A computer readable storage medium for implementing a transport capacity detection method in an example embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0061] Example implementations are now described with reference to the drawings. Example implementations can, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these implementations are provided so that this disclosure will be thorough and complete, and will fully convey the concept of example implementations to those skilled in the art. The described features, structures, or characteristics can be combined in one or more implementations. In the following description, numerous specific details are provided to give a thorough understanding of implementations of the disclosure. One skilled in the relevant art will recognize, however, that the implementations of the disclosure can be practiced without one or more of the

[0062] The terms "one", "a", "an", and "the" as used herein mean "at least one" or "one or more" unless expressly specified otherwise. The term "includes" means "includes but is not limited to". The term "including" means "including without limitation". The term "based on" means "based at least in part on". The term "coupled" means "directly or indirectly coupled".

[0063] In addition, the drawings are merely schematic and are not necessarily drawn to scale. Like reference numerals designate like elements throughout the drawings. Some of the blocks in the diagrams can be functional blocks that represent functions implemented by a processor, software, or other implementation, or a combination thereof. In some embodiments, the functional blocks can be implemented as one or more systems or components of a system.

[0064] With the continuous expansion of the city size, the population growth, the demand for urban life resources is gradually increasing. Among them, when the demand for urban traffic travel resources is increasing, the demand for transport capacity in a certain area will exceed the supply of current time transport capacity, resulting in traffic paralysis. Especially in places like airports, train stations and other hub stations or commercial areas, there is a high probability of insufficient transport capacity.

[0065] In this case, the vehicle dispatching in the concentrated area is mostly used to alleviate, including taxis, buses and shared bicycles, etc. Or through the crawling of network public opinion data for dispatching.

[0066] However, this kind of operation scheduling has many shortcomings. Firstly, the number of regional operation abnormal labels is small or difficult to obtain, and the positive and negative samples are imbalanced, which cannot train a large-scale machine learning model, and it is difficult to migrate to different regions and cities. Secondly, the operation in the detection region is strongly related to the POI (Point of Interest) in the region, but the scale of the POI is not regular. Small-scale POIs can be in one coding grid, and large-scale POIs need multiple grids to cover. For example, movie theaters, shops, and airports, hubs, etc. have large differences in area. Therefore, the model needs to be self-adaptive to different regions, and the range of the detection region is different and changes greatly. Thirdly, the regional passenger flow and operation in different cities have different patterns, and even the functional areas in different regions of the same city also have large differences. Without combining multi-source data to analyze the operation in the region and the distribution of POI, the heterogeneity of multi-source data is considered.

[0067] To solve the problems in the related art, the present disclosure provides an operation detection method. Figure 1 The flowchart of the operation detection method is shown in FIG. 1. Figure 1 As shown in FIG. 1, the operation detection method at least includes the following steps:

[0068] Step S110. Obtain the operation data feature map of the multi-source operation data, and perform data alignment processing on the operation data feature map to obtain a multi-source feature map.

[0069] Step S120. Perform feature extraction processing on the multi-source feature map to obtain an operation feature map, and perform dimension unification processing on the operation feature map to obtain a high-level feature map.

[0070] Step S130. Perform operation detection processing on the high-level feature map to obtain the operation detection result of the multi-source operation data.

[0071] In the exemplary embodiments of the present disclosure, on the one hand, the operation data feature map is processed by data alignment, which considers that the operation detection result is affected by multiple operation data, and ensures the accuracy of operation detection in data by fusing multiple operation data, improves the richness of operation data, is beneficial to train a machine learning model for operation detection, is convenient for migration to different cities and regions, improves the self-adaptation degree of operation detection and the richness of application scenarios; on the other hand, the operation feature map is processed by dimension unification, which considers the differences of reachable regions of different multi-source operation data, solves the size difference problem of corresponding feature maps caused by different multi-source operation data, has good integration, also guarantees the accuracy and real-time of operation detection result, solves the lag problem of operation detection, and can real-time estimate the influence of operation abnormality.

[0072] The steps of the traffic capacity detection method are described in detail below.

[0073] In step S110, a traffic capacity data feature map of multi-source traffic capacity data is obtained, and data alignment processing is performed on the traffic capacity data feature map to obtain a multi-source feature map.

[0074] In the example embodiments of the present disclosure, traffic capacity refers to the total number of people transported by all public passenger transport vehicles that can provide passenger transport services in a selected area at equal intervals of time, for example, with 5-minute equal time intervals. The public passenger transport vehicles can include taxis, online car-hailing vehicles, buses, and the like. For the traffic capacity provided by different public passenger transport vehicles, due to individual differences, the traffic capacity can be evaluated according to the carrying capacity.

[0075] Therefore, the multi-source traffic capacity data can include population flow data, taxi GPS (Global Positioning System) position data, online car-hailing GPS position data, bus GPS trajectory data, and spatial region POI data in the region as multi-source traffic capacity data. In addition to this, the multi-source traffic capacity data can also include other data that affect traffic capacity, which is not particularly limited in the example embodiments.

[0076] In addition, in the training process of the traffic capacity detection model, the multi-source traffic capacity data is the data source of the feature map samples for training the traffic capacity detection model. Since the number of labels for the abnormal gathering of people and the insufficient vehicle traffic capacity in a certain region is small or difficult to obtain in the process of training the traffic capacity detection model, and the positive and negative samples are unbalanced, a machine learning model with a large scale cannot be trained, and migration to different regions and cities is more difficult.

[0077] Therefore, in the process of training the traffic capacity detection model, training samples of traffic capacity data with positive and negative correlations are generated by sampling similar and different samples. At the same time, data augmentation can also be considered from the time and space aspects.

[0078] In an optional embodiment, Figure 2 A flowchart of a method for obtaining a traffic capacity data feature map is shown, as shown in Figure 2 The method at least includes the following steps: in step S210, a reference region and adjacent regions and distant regions associated with the reference region are determined, and reference region data of the reference region, adjacent region data of the adjacent regions, and distant region data of the distant regions are obtained respectively.

[0079] Due to the traffic capacity situation and properties of urban areas, the traffic capacity situation of adjacent regions is similar, and there is almost no sudden change, so samples for training the traffic capacity detection model can be generated by sampling according to the spatial regions of the positions.

[0080] Figure 3 An interface diagram of the determined respective regions is shown as Figure 3 As shown, four samples are sampled in the urban area. Among them, A1 is a reference region, region A2 is a neighboring region adjacent to the reference region A1, region B1 is a remote region far away from the reference region, and region B2 can be a remote region in the case of taking A2 as the reference region.

[0081] After determining the reference region, the neighboring region and the remote region, the in-region population flow data, the taxi GPS position data, the online car-hailing GPS position data, the bus GPS trajectory data and the spatial region POI data of the reference region, the neighboring region and the remote region can be obtained respectively to obtain the corresponding reference region data, the neighboring region data and the remote region data.

[0082] In step S220, the reference region data and the neighboring region data are combined to obtain positive example region data, and the reference region data and the remote region data are combined to obtain negative example region data, so as to obtain the transport capacity data feature map according to the positive example region data and the negative example region data.

[0083] After obtaining the reference region data, the neighboring region data and the remote region data, the reference region data in A1 and the neighboring region data in A2 can be combined to obtain a positive example sample, that is, the positive example region data.

[0084] Further, the reference region data in A1 and the remote region data in B1 can also be combined to obtain a negative example sample, that is, the negative example region data. The negative example sample has randomness, and the probability of generating the negative example sample is higher as the spatial distance of the sampling sample increases.

[0085] Therefore, the positive example region data and the negative example region data can be determined as the multi-source transport capacity data for training the transport capacity detection model.

[0086] Further, the corresponding feature map, that is, the transport capacity data feature map, can be constructed according to the multi-source transport capacity data as the data basis for subsequent processing.

[0087] In the example embodiment, the multi-source transport capacity data including the positive example region data and the negative example region data is obtained by the construction mode of the spatial correlation, which solves the problem of few labels of the regional transport capacity anomaly and has important significance for training the transport capacity detection model.

[0088] In an optional embodiment, Figure 4 Another flowchart of a method for obtaining the transport capacity data feature map is shown as Figure 4As shown, the method at least includes the following steps: in step S410, determining a reference time period and a same time period and a different time period associated with the reference time period, and respectively acquiring reference traffic data of the reference time period, same traffic data of the same time period and different traffic data of the different time period.

[0089] The distribution of traffic capacity also has periodicity in time, which is embodied in the same time period of each day, each week or each month.

[0090] Therefore, a reference time period can be determined, for example, 9 o'clock in the morning every day. Correspondingly, the time period same as this time period and in different days is the same time period, and the time period different from this time period and with a large time interval is the different time period.

[0091] Further, the regional population flow data, taxi GPS position data, online car GPS position data, bus GPS trajectory data and spatial region POI data of the reference time period, the same time period and the different time period are acquired to obtain the corresponding reference traffic data, same traffic data and different traffic data.

[0092] In step S420, the reference traffic data and the same traffic data are combined to obtain positive example time data, and the reference traffic data and the different traffic data are combined to obtain negative example time data, so as to determine the traffic data feature map of the multi-source traffic data according to the positive example time data and the negative example time data.

[0093] After obtaining the reference traffic data, the same traffic data and the different traffic data, the reference traffic data and the same traffic data can be combined to obtain a group of positive example samples, that is, the positive example time data.

[0094] Correspondingly, the reference traffic data and the different traffic data can also be combined to obtain a group of negative example samples, that is, the negative example time data.

[0095] Further, the multi-source traffic data on the time correlation of the training detection model is composed of the positive example time data and the negative example time data.

[0096] Further, the corresponding feature map, that is, the traffic data feature map, can be constructed according to the multi-source traffic data as the data basis for subsequent processing.

[0097] In the present exemplary embodiment, the multi-source traffic data including the positive example time data and the negative example time data is obtained by the construction mode of the spatial correlation, which solves the problem of few labels of regional traffic anomaly and has important significance for training the traffic detection model.

[0098] After obtaining the traffic data feature map of the multi-source traffic data, the traffic data feature map of the multi-source traffic data can also be subjected to data alignment processing.

[0099] In an optional embodiment, Figure 5 A flowchart of a method of data alignment processing is shown, as Figure 5 shown, the method at least includes the following steps: in step S510, address encoding processing is performed on the traffic data feature map to obtain a plurality of original feature maps, and a standard feature map in the plurality of original feature maps is determined.

[0100] In order to align the positions in the multi-source traffic data space, geohash can be used to perform address encoding processing on the traffic data feature map to perform grid division. Geohash is an address encoding method that encodes two-dimensional spatial latitude and longitude grid data into a string.

[0101] The geohash algorithm can first convert latitude and longitude into binary. Then, the latitude and longitude are merged. Among them, the longitude occupies the even bits, and the latitude occupies the odd bits, and 0 is also an even bit. Finally, encoding is performed according to Base32. Geohash is much more efficient than directly using latitude and longitude, and users can publish address codes to indicate their location without exposing their precise coordinates, which helps to protect privacy.

[0102] Among them, the length of the geohash algorithm can be selected as 7, and the space size is approximately a rectangular region of 152*152.

[0103] Therefore, address encoding processing by the geohash algorithm can obtain original feature maps corresponding to regional population flow data, taxi GPS position data, online car-hailing GPS position data, bus GPS trajectory data, and spatial region POI data in the multi-source traffic data, respectively.

[0104] Due to different detection areas, the size of the original feature map is not fixed as a constant, but can be adjusted according to the actual situation of the size of the detection area.

[0105] Because the accessible area of the regional population flow and the public passenger transport vehicle is different, the shape of the original feature map of the multi-source traffic data of different dimensions can be different. Among them, the shape of the original feature map includes the dimension of the tensor and the length of each dimension.

[0106] Further, because the accessible area of the population flow is relatively standard, the region is usually divided according to the population size, so the original feature map corresponding to the population flow data can be determined as the standard feature map for subsequent feature map alignment processing.

[0107] In step S520, based on the standard feature map, the feature map alignment processing is performed on the other feature maps in the plurality of original feature maps to obtain a plurality of aligned feature maps.

[0108] In the feature map alignment processing, the other feature maps are respectively preprocessed by convolution based on the standard feature map.

[0109] Specifically, the size of the standard feature map is X people_rawout *Y people_raw_out The size of the original feature map of the taxi GPS position data is X taxi_raw_out *Y taxi_raw_out Therefore, the preprocessing is performed by a convolution kernel of (X taxi_raw_out -X people_raw_out +1, Y taxi_raw_out -Y people_raw_out +1) and a convolution with a step of 1. The size of X taxi_raw_out *Y taxi_raw_out is greater than the size of X people_raw_out *Y people_raw_out .

[0110] The convolution kernel for preprocessing can be trained, and the convolution kernel can be one convolution or a stack of multiple convolution structures.

[0111] It is worth noting that the size of the original feature map corresponding to the remaining data such as the online car-hailing GPS position data, the bus GPS trajectory data, and the spatial region POI data is also greater than the size of the standard data map.

[0112] Therefore, after the feature map alignment processing is performed on the original feature map corresponding to the online car-hailing GPS position data, the bus GPS trajectory data, and the spatial region POI data by using the standard feature map, a plurality of aligned feature maps can be obtained.

[0113] In step S530, the feature map connection processing is performed on the standard feature map and the plurality of aligned feature maps to obtain a multi-source feature map.

[0114] After the feature map alignment processing, the size of the plurality of aligned feature maps is the same as the size of the standard feature map, and therefore, the feature map connection processing can be performed on the standard feature map and the plurality of aligned feature maps.

[0115] The feature map connection processing can make one standard feature map and four aligned feature maps respectively constitute one channel to form a three-dimensional tensor, i.e., a multi-source feature map.

[0116] In the example embodiment, the data alignment processing is performed on the transport capacity data feature map to obtain a multi-source feature map. The multi-source transport capacity data is fused considering the heterogeneity of the multi-source transport capacity data, and a data basis is provided for subsequent feature extraction processing.

[0117] In step S120, the multi-source feature map is subjected to feature extraction processing to obtain a transport capacity feature map, and the transport capacity feature map is subjected to dimension unification processing to obtain a high-level feature map.

[0118] In an example embodiment of the present disclosure, the multi-source feature map is subjected to feature extraction processing to obtain a transport capacity feature map. The feature extraction processing can be implemented using a ResNet (Residual Network). The ResNet network is composed of a plurality of structures stacked by CNN (Convolutional Neural Networks), BN (Batch Normalization), and relu (Rectified Linear Unit) to implement the feature extraction processing on the multi-source feature map.

[0119] In addition, other CNN structure stacking structures such as VGG, googleNet, transformer, or GNN can also be used to implement the feature extraction processing, and the example embodiment does not make special limitations.

[0120] Since the multi-source feature maps of the multi-source transport capacity data have different sizes, dimension unification is required. Therefore, after obtaining the transport capacity feature map, the dimension unification processing can be performed on the transport capacity feature map.

[0121] In an optional embodiment, Figure 6 A flowchart of a method for dimension unification processing is shown, as shown in Figure 6 In step S610, the transport capacity feature map is subjected to feature map division processing to obtain a plurality of pooling regions.

[0122] When performing the dimension unification processing, the last ROI Pooling (Regions of Interest Pooling) layer of the network for feature extraction processing can be processed.

[0123] The ROI Pooling layer is a kind of Pooling layer, and the feature map size of the input is not fixed, but the feature map size of the output is fixed, which is used to process the case where the dimensions of samples are not unified.

[0124] In addition, the size of the output feature map of the ROI Pooling layer, that is, the high-level feature map, is Xroi_out *Y roi_out The size of the feature map after the feature extraction processing before the ROI Pooling layer is X fea_out *Y fea_out .

[0125] Further, the pooling area is divided, and the maximum value of each pooling area is calculated and

[0126] The lower horizontal and vertical index ranges of the i-th area are calculated as Therefore, the feature map with the size of X fea_out *Y fea_out is divided into X roi_out *Y roi_out areas, i.e., pooling areas.

[0127] In step S620, the target value processing is performed on the plurality of pooling areas to obtain a plurality of target values, and the plurality of target values are spliced to obtain a high-level feature map.

[0128] After the pooling areas are obtained, the target value processing can be performed on each pooling area. For example, the maximum value of each pooling area can be taken to obtain a plurality of target values.

[0129] Further, the plurality of target values are spliced according to the relative positions of the pooling area division to obtain a high-level feature map. The high-level feature map is a feature map with the size of X roi_out *Y roi_out .

[0130] Figure 7 An example diagram of the dimension unification processing is shown, as shown in Figure 7 , the pooling area is divided into four pooling areas, and the maximum values in the four pooling areas, i.e., the target values, are taken. The target values are 0.89, 0.96, 0.90, and 0.86. Then, the plurality of target values are spliced according to the relative positions of the pooling area division to obtain a high-level feature map composed of 0.89, 0.96, 0.90, and 0.86.

[0131] Compared with the feature map, the high-level feature map has a larger thickness, but a smaller length and width. For example, the feature map can be a feature map with the size of 64*64*5, and the high-level feature map can be a feature map with the size of 16*16*512.

[0132] In the example embodiment, the high-level feature map corresponding to the feature map can be obtained through the dimension unification processing, which can adapt to the differences of different sizes of original feature maps and integrate different sizes of original feature maps.

[0133] In step S130, the high-level feature map is subjected to the capacity detection processing to obtain a capacity detection result of the multi-source capacity data.

[0134] In an exemplary embodiment of the present disclosure, after obtaining the high-level feature map, the data alignment processing layer, the feature extraction processing layer, and the dimension unification processing layer in the capacity detection model can be trained by subjecting the high-level feature map to the capacity detection processing.

[0135] In an optional embodiment, Figure 8 A flowchart of a method of capacity detection processing is shown, as shown in Figure 8 The method at least includes the following steps: in step S810, two high-level feature maps corresponding to the capacity data feature map are subjected to feature map conversion processing to obtain two feature vectors.

[0136] Different cities or different regions of a city have differences, and it is difficult to obtain sample labels, so it is difficult to train a large-scale capacity detection model, and therefore a self-supervised training method can be used.

[0137] Figure 9 A model architecture diagram of training the capacity detection model is shown, as shown in Figure 9 In the model architecture of training the capacity detection model, in addition to the input module for inputting the capacity data feature map, the preprocessing module for implementing data alignment processing, and the feature extraction module for implementing feature extraction processing and dimension unification processing, a linear layer can also be included. The linear layer is used to implement feature map conversion processing on the two high-level feature maps, so the linear layer can convert the high-level feature maps into feature vectors.

[0138] It is worth noting that the parameters of the preprocessing module, the feature extraction module, and the linear layer are shared. And in the process of feature map conversion processing, two multi-source capacity data obtained by data augmentation from time and / or space features are simultaneously input. The feature vectors are output after being processed by the preprocessing module, the feature extraction module, and the linear layer.

[0139] In step S820, the two feature vectors are subjected to training loss calculation to obtain a training loss value, and the high-level feature maps are subjected to capacity detection processing according to the training loss value to obtain a capacity detection result of the multi-source capacity data.

[0140] After obtaining the two feature vectors, the backpropagation training is performed through the training loss calculation. When the two multi-source capacity data are positive example samples, the feature representation distance of the two positive example samples, i.e., the training loss value, is as close as possible; and when the two multi-source capacity data are negative example samples, the feature representation distance of the two negative example samples, i.e., the training loss value, is as far as possible.

[0141] wherein the training loss value is determined in a manner as shown in equation (1):

[0142]

[0143] wherein, the value of [·] is in the range of {0, 1}, the function value is 1 when [·] is true, and the function value is 0 when [·] is false. X represents the multi-source traffic data of the sample data, and τ is a constant.

[0144] After the training loss value is calculated, the traffic detection result of the multi-source traffic data can be determined according to the training loss value.

[0145] In an optional embodiment, Figure 10 a flowchart of a method for determining the traffic detection result is shown, as shown in Figure 10 the method at least includes the following steps: in step S1010, if the training loss value is less than the training loss threshold value corresponding to the training loss value, the traffic detection processing is performed on the high-level feature map to obtain the traffic detection result of the multi-source traffic data.

[0146] In order to determine whether the preprocessing module and the feature extraction module in the traffic detection model are successfully trained, the training loss threshold value corresponding to the training loss value can be obtained for comparison and determination.

[0147] When the comparison result is that the training loss value is less than the training loss threshold value, it is determined that the preprocessing module and the feature extraction module in the traffic detection model have been successfully trained, and the traffic detection processing can be performed on the high-level feature map.

[0148] In an optional embodiment, Figure 11 a flowchart of a method for further performing traffic detection processing is shown, as shown in Figure 11 the method at least includes the following steps: in step S1110, the reference detection result corresponding to the multi-source traffic data is obtained, and the traffic detection processing is performed on the high-level feature map to obtain the to-be-determined detection result of the multi-source traffic data.

[0149] In the process of performing traffic detection processing, the inference module in the traffic detection model can be further trained and determined whether it is successfully trained.

[0150] Therefore, the reference detection result corresponding to the multi-source transport capacity data can be obtained. The multi-source transport capacity data used for training the inference module is real sample data, and sample data obtained by using data augmentation in the process of training the preprocessing module and the feature extraction module is not used, so the reference detection result is also a real transport capacity abnormality result corresponding to real sample data. Moreover, since only a small amount of sample data is required for training the inference module, the multi-source transport capacity data and the reference detection result can be manually labeled, or can be automatically labeled according to actual conditions, and the present exemplary embodiment does not make special limitation thereto.

[0151] After obtaining the reference detection result, the high-level feature map can be input to a binary classification network, such as an MLP (Muti-Layer Perception), for transport capacity detection processing to obtain a pending detection result of the multi-source transport capacity data, which includes probabilities of two results of transport capacity abnormality and transport capacity normality.

[0152] The MLP neural network is a neural network with normal signal propagation and error backpropagation. During forward propagation, the input signal is transmitted to the output layer after being processed by the hidden layer. If the output layer neurons fail to achieve the expected output, the output error will be returned to the input layer through the hidden layer in a certain form during the input error backpropagation stage, and will be “shared” by the neurons of the hidden layer and the input layer, so as to obtain the error signals of the neurons of each layer as the basis for modifying the weight values of each unit.

[0153] The modification process of the weight matrix of each layer of such signal normal propagation and error backpropagation is repeated, and the weight is continuously modified until the error of the network output gradually decreases to an acceptable level or a set number of learning times is reached, thereby completing the learning and training of the network.

[0154] In step S1120, the inference loss value is calculated by performing inference loss calculation on the reference detection result and the pending detection result, and the transport capacity detection result of the multi-source transport capacity data is determined according to the inference loss value.

[0155] Further, the log loss (cross-entropy) loss function is selected to calculate the inference loss value by performing inference loss calculation on the reference detection result and the pending detection result. The inference loss calculation method is referred to formula (2):

[0156] Loss = -(ylog(p) + (1-y)log(1-p)) (2)

[0157] Wherein, y is the real label of the sample, that is, the reference detection result, and p is the probability predicted by the model.

[0158] After obtaining the inference loss value, the transport capacity detection result of the multi-source transport capacity data can be determined according to the inference loss value.

[0159] In an optional embodiment, Figure 12 A flowchart of a method for further determining the capacity detection result is shown, as Figure 12 As shown, the method comprises the following steps: in step S1210, if the inference loss value is less than the inference loss threshold value corresponding to the inference loss value, the pending detection result is determined as the capacity detection result of the multi-source capacity data.

[0160] In order to determine the capacity detection result of the multi-source capacity data, the inference loss threshold value corresponding to the inference loss value can be determined, and the inference loss value and the inference loss threshold value are compared and determined.

[0161] When the inference loss value is less than the inference loss threshold value, it indicates that the inference layer for capacity detection processing is completed, so the pending detection result can be determined as the capacity detection result of the multi-source capacity data.

[0162] In step S1220, if the inference loss value is greater than or equal to the inference loss threshold value corresponding to the inference loss value, the capacity detection processing of the high-level feature map is re-performed according to the reference detection result to obtain the capacity detection result of the multi-source capacity data.

[0163] When the inference loss value is greater than or equal to the inference loss threshold value, it indicates that the inference layer for capacity detection processing is not trained, so the weight matrix of the MLP network for capacity detection processing can be modified, and the capacity detection processing of the high-level feature map is re-performed until the inference loss value between the pending detection result of the capacity detection processing and the reference detection result meets the standard of the inference loss threshold value, and the corresponding pending detection result is determined as the capacity detection result of the multi-source capacity data.

[0164] In the present exemplary embodiment, after the pre-processing module and the feature extraction module for data alignment processing, feature extraction processing and dimension unification processing are trained, the inference module of the capacity detection model can also be trained to obtain the capacity detection result using the real label. The training method is simple and accurate, and is closely linked and logically rigorous, which ensures the training integrity and accuracy of the capacity detection model, and the correctness of the subsequent detection result of the multi-source capacity data.

[0165] In step S1020, if the training loss value is greater than or equal to the training loss threshold value corresponding to the training loss value, the data alignment processing, the feature extraction processing and the dimension unification processing of the capacity data feature map are re-performed to obtain the capacity detection result of the multi-source capacity data.

[0166] During the training of the preprocessing and feature extraction modules, which perform data alignment, feature extraction, and dimensionality unification, the training loss value may sometimes be greater than or equal to the training loss threshold. This indicates that the preprocessing and feature extraction modules have not been trained successfully.

[0167] Therefore, the weights in the preprocessing and feature extraction modules can be modified, and the capacity data feature map can be re-processed for data alignment, feature extraction, and dimensionality unification. In other words, the preprocessing and feature extraction modules can be trained again until the training loss value meets the training loss threshold. Figure 11 The method shown is used to obtain the capacity detection results of multi-source capacity data.

[0168] In this exemplary embodiment, the modules involved in data alignment processing, feature extraction processing, and dimensionality unification processing are fully trained by using training loss values ​​and training loss thresholds. This is of great significance for training the capacity detection model and also ensures the accuracy of the capacity detection results output.

[0169] The capacity detection method in this embodiment of the present disclosure will be described in detail below with reference to an application scenario.

[0170] Figure 13 The diagram illustrates the modules of the capacity detection model in an application scenario, such as... Figure 13 As shown, the capacity detection model consists of five parts: data input, preprocessing, feature extraction network, inference, and output.

[0171] In the data input module, various data sources are used as multi-source transportation capacity data, including population flow data within the region, taxi GPS location data, ride-hailing GPS location data, bus GPS trajectory data, and spatial area POI data. A transportation capacity data feature map input network corresponding to the multi-source transportation capacity data is constructed. In addition, the multi-source transportation capacity data may also include other data that affect transportation capacity, and this exemplary embodiment does not impose any special limitations on this.

[0172] In the preprocessing module, since the feature maps input by the input module have different shapes, the preprocessing module needs to preprocess the feature maps to form a multi-channel feature map.

[0173] To align the spatial locations of multi-source capacity data, geohash can be used to encode the capacity data feature map and perform grid partitioning. Geohash is an address encoding method that encodes two-dimensional spatial latitude and longitude grid data into a string.

[0174] The length of the geohash algorithm can be selected as 7, and the space size is approximately a rectangular area of ​​152*152.

[0175] Therefore, the address coding processing by the geohash algorithm can obtain the original feature map corresponding to the population flow data in the region, the taxi GPS position data, the online car-hailing GPS position data, the bus GPS trajectory data and the spatial region POI data in the multi-source transport capacity data, respectively.

[0176] Due to the difference of the detection region, the size of the original feature map is not fixed as a constant, but can be adjusted according to the actual situation of the size of the detection region.

[0177] Due to the difference of the detection region, the size of the original feature map is not fixed as a constant, but can be adjusted according to the actual situation of the size of the detection region.

[0178] Further, since the accessible region of the population flow is relatively standard, the region is usually divided according to the population size, so the original feature map corresponding to the population flow data can be determined as a standard feature map for subsequent feature map alignment processing.

[0179] Among them, the feature map alignment processing needs to take the standard feature map as the basis to respectively perform convolution preprocessing on other feature maps.

[0180] Specifically, the size of the standard feature map is X people_raw_out *Y people_raw_out , the size of the original feature map of the taxi GPS position data is X taxi_raw_out *Y taxi_raw_out , therefore, by using the convolution kernel of (X taxi_raw_out -X people_raw_out +1, Y taxi_raw_out -Y people_raw_out +1) and the convolution with the step length of 1 for preprocessing. Among them, the size of X taxi_raw_out *Y taxi_raw_out is greater than X people_raw_out *Y people_raw_out .

[0181] The convolution kernel for preprocessing can be trained, and the convolution kernel can be one convolution or a stack of multiple similar convolution structures.

[0182] It is worth noting that the size of the original feature map corresponding to the remaining online car-hailing GPS position data, bus GPS trajectory data and spatial region POI data is also greater than the size of the standard data map.

[0183] Therefore, after the standard feature map is used to perform feature map alignment processing on the original feature maps corresponding to the data such as the online car-hailing GPS location data, the bus GPS trajectory data, and the spatial region POI data, a plurality of aligned feature maps can be obtained.

[0184] After the feature map alignment processing, the size of the plurality of aligned feature maps is the same as the size of the standard feature map, and therefore, the feature map connection processing can be performed on the standard feature map and the plurality of aligned feature maps.

[0185] The feature map connection processing can make one standard feature map and four aligned feature maps respectively constitute one channel, form a three-dimensional tensor, and form a multi-source feature map.

[0186] The feature extraction module mainly uses the structure of the convolutional neural network to perform feature extraction. Considering that the sample labels are unlabeled and the number of labels is small, the feature extraction module is pre-trained in a contrast learning manner.

[0187] The multi-source feature map is subjected to feature extraction processing to obtain a transport capacity feature map. The feature extraction processing can be implemented by using a ResNet. The ResNet network is composed of a plurality of structures stacked by CNN, BN, and relu, to implement the feature extraction processing on the multi-source feature map.

[0188] In addition, other CNN structure stacking structures such as VGG, googleNet, transformer, or GNN can also be used to implement the feature extraction processing, and the present example embodiment does not specially limit this.

[0189] Since the sizes of the multi-source feature maps of the multi-source transport capacity data are different, dimension unification is required. Then, after the transport capacity feature map is obtained, the dimension unification processing can be performed on the transport capacity feature map.

[0190] When the dimension unification processing is performed, the ROI Pooling layer can be added at the end of the network of the feature extraction processing.

[0191] The ROI Pooling layer is a kind of Pooling layer, and the feature of the ROI Pooling layer is that the size of the input feature map is not fixed, but the size of the output feature map is fixed, and is used to process the case that the dimensions of samples are not unified.

[0192] In addition, the size of the output feature map, that is, the high-level feature map, of the ROI Pooling layer is X roi_out *Y roi_out The size of the transport capacity feature map after the feature extraction processing before the ROI Pooling layer is X fea_out *Y fea_out .

[0193] Further, the pooling area is divided, and the lower horizontal and vertical index ranges of the i-th area are calculated respectively and

[0194] The lower horizontal and vertical index ranges of the i-th area are calculated respectively Therefore, the size of the capacity feature map is X fea_out *Y fea_out is divided into X roi_out *Y roi_out areas, that is, pooling areas.

[0195] After obtaining the pooling area, target value processing can be performed on each pooling area. For example, the maximum value of each pooling area can be taken to obtain a plurality of target values.

[0196] Further, the plurality of target values are spliced according to the relative positions of the pooling area division to obtain a high-level feature map. The high-level feature map is a feature map with a size of X roi_out *Y roi_out .

[0197] Compared with the capacity feature map, the high-level feature map has a larger thickness, but a smaller length and width. For example, the capacity feature map can be a feature map with a size of 64*64*5, and the high-level feature map can be a feature map with a size of 16*16*512.

[0198] The pre-processing and feature extraction modules trained by contrastive learning are fixed parameters, and the final inference module is fine-tuned through a small amount of labels. A small amount of labels can be manually annotated, and samples can be roughly screened through external events or setting thresholds for various vehicle and population data, and further annotated by artificial means. The final output of the model is a vector with a length of 2, which is the probability of capacity anomaly and normality.

[0199] During the capacity detection process, the inference module in the capacity detection model can be further trained and whether the training is successful can be determined.

[0200] Therefore, the reference detection result corresponding to the multi-source transport capacity data can be obtained. The feature map of the multi-source transport capacity data used for training the inference module is real sample data, and sample data obtained by using data augmentation in the training of the preprocessing module and the feature extraction module is not used. Therefore, the reference detection result is also a real transport capacity abnormality result corresponding to real sample data. Moreover, since only a small amount of sample data is required for training the inference module, the feature map of the multi-source transport capacity data and the reference detection result can be manually labeled, or can be automatically labeled according to actual conditions, and the present exemplary embodiment does not make special limitations on this.

[0201] After obtaining the reference detection result, the high-level feature map can be input to a binary classification network, such as an MLP, for transport capacity detection processing to obtain a pending detection result of the multi-source transport capacity data, which includes probabilities of two results of transport capacity abnormality and transport capacity normality.

[0202] Further, a log loss (cross-entropy) loss function is selected to calculate an inference loss value by performing inference loss calculation on the reference detection result and the pending detection result. The inference loss calculation method is referred to formula (2).

[0203] After obtaining the inference loss value, the transport capacity detection result of the multi-source transport capacity data can be determined according to the inference loss value.

[0204] To determine the transport capacity detection result of the multi-source transport capacity data, an inference loss threshold value corresponding to the inference loss value can be determined, and the inference loss value and the inference loss threshold value can be compared and determined.

[0205] When the inference loss value is less than the inference loss threshold value, it indicates that the inference layer for transport capacity detection processing is trained, and therefore the pending detection result can be determined as the transport capacity detection result of the multi-source transport capacity data.

[0206] When the inference loss value is greater than or equal to the inference loss threshold value, it indicates that the inference layer for transport capacity detection processing is not trained, and therefore the weight matrix of the MLP network for transport capacity detection processing can be modified, and the high-level feature map can be reprocessed for transport capacity detection processing until the inference loss value between the pending detection result of the transport capacity detection processing and the reference detection result meets the standard of the inference loss threshold value, and the corresponding pending detection result is determined as the transport capacity detection result of the multi-source transport capacity data.

[0207] In the application scenario of the present disclosure, on the one hand, the data alignment processing is performed on the transport capacity data feature map, and in the case that the transport capacity detection result is affected by multiple transport capacity data, the data fusion of the multiple transport capacity data is performed to ensure the accuracy of the transport capacity detection in the data aspect, improve the richness of the transport capacity data, facilitate the training of the machine learning model for the transport capacity detection, facilitate the migration to different cities and regions, and improve the self-adaptation degree of the transport capacity detection and the richness of the application scenario; on the other hand, the dimension unification processing is performed on the transport capacity feature map, and the differences in the reachable areas of different multi-source transport capacity data are considered, the size difference problems of the corresponding feature maps caused by different multi-source transport capacity data are solved, the integration is good, and the accuracy and real-time performance of the transport capacity detection result are ensured, the lag problem of the transport capacity detection is solved, and the influence caused by the transport capacity anomaly can be estimated in real time.

[0208] In addition, in the exemplary embodiments of the present disclosure, a transport capacity detection device is also provided. Figure 14 The structure diagram of the transport capacity detection device is shown as Figure 14 As shown, the transport capacity detection device 1400 can include a data alignment module 1410, a dimension unification module 1420, and a transport capacity detection module 1430. Among them:

[0209] The data alignment module 1410 is configured to obtain a transport capacity data feature map of multi-source transport capacity data, and perform data alignment processing on the transport capacity data feature map to obtain a multi-source feature map; the dimension unification module 1420 is configured to perform feature extraction processing on the multi-source feature map to obtain a transport capacity feature map, and perform dimension unification processing on the transport capacity feature map to obtain a high-level feature map; and the transport capacity detection module 1430 is configured to perform transport capacity detection processing on the high-level feature map to obtain a transport capacity detection result of the multi-source transport capacity data.

[0210] In an exemplary embodiment of the present disclosure, the data alignment processing on the transport capacity data feature map obtains a multi-source feature map, which includes:

[0211] Address encoding processing is performed on the transport capacity data feature map to obtain a plurality of original feature maps, and a standard feature map in the plurality of original feature maps is determined;

[0212] Based on the standard feature map, feature map alignment processing is performed on the feature maps other than the standard feature map in the plurality of original feature maps to obtain a plurality of aligned feature maps;

[0213] Feature map connection processing is performed on the standard feature map and the plurality of aligned feature maps to obtain the multi-source feature map.

[0214] In an exemplary embodiment of the present disclosure, the dimension unification processing on the transport capacity feature map obtains a high-level feature map, which includes:

[0215] The feature map is divided to obtain a plurality of pooling regions;

[0216] The target values of the plurality of pooling regions are obtained, and the plurality of target values are spliced to obtain a high-level feature map.

[0217] In an exemplary embodiment of the present application, the feature map of the multi-source transport capacity data is obtained, comprising:

[0218] The reference region, the adjacent region and the distant region associated with the reference region are determined, and the reference region data of the reference region, the adjacent region data of the adjacent region and the distant region data of the distant region are obtained respectively;

[0219] The reference region data and the adjacent region data are combined to obtain positive example region data, and the reference region data and the distant region data are combined to obtain negative example region data, so as to determine the feature map of the multi-source transport capacity data according to the positive example region data and the negative example region data.

[0220] In an exemplary embodiment of the present application, the feature map of the multi-source transport capacity data is obtained, comprising:

[0221] The reference time period, the same time period and the different time period associated with the reference time period are determined, and the reference transport capacity data of the reference time period, the same transport capacity data of the same time period and the different transport capacity data of the different time period are obtained respectively;

[0222] The reference transport capacity data and the same transport capacity data are combined to obtain positive example time data, and the reference transport capacity data and the different transport capacity data are combined to obtain negative example time data, so as to determine the feature map of the multi-source transport capacity data according to the positive example time data and the negative example time data.

[0223] In an exemplary embodiment of the present application, the high-level feature map is detected to obtain the transport capacity detection result of the multi-source transport capacity data, comprising:

[0224] The two high-level feature maps corresponding to the feature map of the transport capacity data are converted to obtain two feature vectors;

[0225] The training loss value is calculated according to the two feature vectors, and the high-level feature map is detected according to the training loss value to obtain the transport capacity detection result of the multi-source transport capacity data.

[0226] In an exemplary embodiment of the present application, the high-level feature map is detected according to the training loss value to obtain the transport capacity detection result of the multi-source transport capacity data, comprising:

[0227] If the training loss value is less than the training loss threshold corresponding to the training loss value, the high-level feature map is subjected to the operation detection processing to obtain the operation detection result of the multi-source operation data.

[0228] If the training loss value is greater than or equal to the training loss threshold corresponding to the training loss value, the operation data feature map is subjected to the data alignment processing, the feature extraction processing and the dimension unification processing again to obtain the operation detection result of the multi-source operation data.

[0229] In an exemplary embodiment of the present application, the operation detection result of the multi-source operation data obtained by subjecting the high-level feature map to the operation detection processing comprises:

[0230] The reference detection result corresponding to the multi-source operation data is obtained, and the high-level feature map is subjected to the operation detection processing to obtain the to-be-determined detection result of the multi-source operation data.

[0231] The inference loss value is calculated by inferring the reference detection result and the to-be-determined detection result, and the operation detection result of the multi-source operation data is determined according to the inference loss value.

[0232] In an exemplary embodiment of the present application, the operation detection result of the multi-source operation data is determined according to the inference loss value, comprising:

[0233] If the inference loss value is less than the inference loss threshold corresponding to the inference loss value, the to-be-determined detection result is determined as the operation detection result of the multi-source operation data.

[0234] If the inference loss value is greater than or equal to the inference loss threshold corresponding to the inference loss value, the high-level feature map is subjected to the operation detection processing again according to the reference detection result to obtain the operation detection result of the multi-source operation data.

[0235] The specific details of the operation detection device 1400 have been described in detail in the corresponding operation detection method, and therefore will not be described here again.

[0236] It should be noted that although several modules or units of the operation detection device 1400 are mentioned in the foregoing detailed description, such division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units.

[0237] In addition, in an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided.

[0238] The electronic device 1500 according to this embodiment of the present application will be described below with reference to Figure 15 .Figure 15 The electronic device 1500 shown is merely one example and should not be taken as limiting the scope of the present embodiments.

[0239] As shown in Figure 15 The electronic device 1500 is in the form of a general computing device. Components of the electronic device 1500 can include, but are not limited to, the at least one processing unit 1510 described above, the at least one storage unit 1520 described above, a bus 1530 that connects different system components, including the storage unit 1520 and the processing unit 1510, a display unit 1540.

[0240] The storage unit stores program code that can be executed by the processing unit 1510 such that the processing unit 1510 performs the steps described above in the "Exemplary Methods" section of this specification according to various exemplary embodiments of the present application.

[0241] The storage unit 1520 can include a readable medium in the form of volatile storage such as random access memory (RAM) 1521 and / or cache memory 1522, and can further include non-volatile storage such as read-only memory (ROM) 1523.

[0242] The storage unit 1520 can further include program / utility 1524 having a set of programs / modules 1525, each of which performs one or more functions in accordance with the present embodiments. These programs / modules include, but are not limited to, an operating system, one or more applications, other programs / modules, and program / data, each or a combination of which can include implementation of a network environment.

[0243] The bus 1530 can represent one or more of several types of bus structures, including a storage bus or bus controller, a peripheral bus, a graphics bus, a processor or local bus using any of a variety of bus structures, and the like.

[0244] The electronic device 1500 can also communicate with one or more external devices 1700 such as a keyboard, a pointing device, a Bluetooth device, etc.; and can communicate with one or more devices that enable a user to interact with the electronic device 1500 and / or one or more devices (e.g. routers, modems, etc.) that enable the electronic device 1500 to communicate with one or more other computing devices. Such communication can occur via Input / Output (I / O) interface 1550. Still yet, the electronic device 1500 can communicate with one or more networks such as a local area network (LAN), a wide area network (WAN), and / or the Internet through network adapter 1560. As depicted, network adapter 1560 communicates with the other components of the electronic device 1500 via bus 1530. It should be appreciated that although not shown, other hardware and / or software modules could be used in connection with the electronic device 1500. Such modules include, but are not limited to, microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc.

[0245] From the above description of the embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software, or by software in combination with necessary hardware. Therefore, the technical solutions according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash disk, a mobile hard disk, etc.) or a network, and includes a number of instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to perform the methods according to the embodiments of the present disclosure.

[0246] In the example embodiments of the present disclosure, a computer readable storage medium is also provided, on which a program product capable of implementing the above-mentioned method of the present disclosure is stored. In some possible embodiments, various aspects of the present disclosure can also be implemented in the form of a program product, which includes program codes for causing a terminal device to perform the steps described in the above-mentioned "example method" section according to various example embodiments of the present disclosure when the program product is run on the terminal device.

[0247] Reference Figure 16 As shown, a program product 1600 for implementing the above-mentioned method according to the embodiments of the present disclosure is described, which can adopt a portable compact disc read-only memory (CD-ROM) and include program codes, and can be run on a terminal device such as a personal computer. However, the program product of the present disclosure is not limited to this, and in this document, the readable storage medium can be any tangible medium containing or storing a program, which can be used or combined with an instruction execution system, device or apparatus.

[0248] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0249] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0250] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0251] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0252] Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the features disclosed herein. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

Claims

1. A method for detecting transport capacity, characterized in that, The method includes: A capacity data feature map is obtained from multi-source capacity data, and the capacity data feature map is then processed by a preprocessing module to obtain a multi-source feature map. The multi-source feature map is processed by the feature extraction module to obtain the capacity feature map, and the capacity feature map is then processed by dimensional unification to obtain the high-level feature map. The capacity detection results of the multi-source capacity data are obtained by performing capacity detection processing on the high-level feature map; The method further includes: Obtain positive and negative samples. Each positive sample includes the capacity data feature map of two similar multi-source capacity data, and each negative sample includes the capacity data feature map of two dissimilar multi-source capacity data. The positive and negative samples are processed by the preprocessing module, the feature extraction module, and the linear layer, and the feature representation distances of the positive and negative samples are determined. The training loss value is determined based on the feature representation distance between the positive sample and the feature representation distance between the negative sample; the smaller the feature representation distance of the positive sample, the smaller the training loss value; the larger the feature representation distance of the negative sample, the smaller the training loss value. The preprocessing module and the feature extraction module are trained based on the training loss value.

2. The capacity detection method according to claim 1, characterized in that, The process of aligning the capacity data feature map to obtain a multi-source feature map includes: The capacity data feature map is processed by address encoding to obtain multiple original feature maps, and the standard feature map in the multiple original feature maps is determined. Based on the standard feature map, feature map alignment processing is performed on the other feature maps in the multiple original feature maps, excluding the standard feature map, to obtain multiple aligned feature maps; The standard feature map and the plurality of aligned feature maps are subjected to feature map concatenation processing to obtain a multi-source feature map.

3. The capacity detection method according to claim 1, characterized in that, The process of unifying the dimensions of the capacity feature map to obtain a high-level feature map includes: The capacity feature map is divided into multiple pooled regions. Multiple target values ​​are obtained by performing target value selection on the multiple pooling regions, and the multiple target values ​​are then concatenated to obtain a high-level feature map.

4. The capacity detection method according to claim 1, characterized in that, The acquisition of positive and negative samples includes: Determine a reference region and its neighboring and distant regions associated with the reference region, and obtain the reference region data of the reference region, the neighboring region data of the neighboring region, and the distant region data of the distant region, respectively; Positive examples are obtained by combining the data from the reference area and the data from the neighboring area, and negative examples are obtained by combining the data from the reference area and the data from the distant area.

5. The capacity detection method according to claim 1, characterized in that, The acquisition of positive and negative samples includes: Determine a baseline time period and concurrent time periods and different time periods associated with the baseline time period, and obtain the baseline capacity data for the baseline time period, the concurrent capacity data for the concurrent time periods, and the different capacity data for the different time periods, respectively. Positive examples are obtained by combining the baseline capacity data and the simultaneous capacity data, and negative examples are obtained by combining the baseline capacity data and the different capacity data.

6. The capacity detection method according to claim 4 or 5, characterized in that, The process of performing capacity detection processing on the high-level feature map to obtain the capacity detection result of the multi-source capacity data includes: Two feature vectors are obtained by performing feature map transformation on the two high-level feature maps corresponding to the capacity data feature map; The training loss value is obtained by calculating the training loss of the two feature vectors, and the capacity detection result of the multi-source capacity data is obtained by performing capacity detection processing on the high-level feature map based on the training loss value.

7. The capacity detection method according to claim 6, characterized in that, The step of performing capacity detection processing on the high-level feature map based on the training loss value to obtain the capacity detection result of the multi-source capacity data includes: If the training loss value is less than the training loss threshold corresponding to the training loss value, the high-level feature map is processed for capacity detection to obtain the capacity detection result of the multi-source capacity data. If the training loss value is greater than or equal to the training loss threshold corresponding to the training loss value, the capacity data feature map is re-processed with data alignment, feature extraction, and dimension unification to obtain the capacity detection result of the multi-source capacity data.

8. The capacity detection method according to claim 7, characterized in that, The process of performing capacity detection processing on the high-level feature map to obtain the capacity detection result of the multi-source capacity data includes: Obtain reference detection results corresponding to the multi-source capacity data, and perform capacity detection processing on the high-level feature map to obtain the undetermined detection results of the multi-source capacity data; The inferred loss value is obtained by performing inference loss calculation on the reference detection result and the pending detection result, and the capacity detection result of the multi-source capacity data is determined based on the inferred loss value.

9. The capacity detection method according to claim 8, characterized in that, Determining the capacity detection result of the multi-source capacity data based on the inferred loss value includes: If the inferred loss value is less than the inferred loss threshold corresponding to the inferred loss value, the pending detection result is determined to be the capacity detection result of the multi-source capacity data; If the inferred loss value is greater than or equal to the inferred loss threshold corresponding to the inferred loss value, the high-level feature map is reprocessed for capacity detection based on the reference detection result to obtain the capacity detection result of the multi-source capacity data.

10. A capacity detection device, characterized in that, include: The data alignment module is configured to acquire a capacity data feature map of multi-source capacity data, and to perform data alignment processing on the capacity data feature map through the preprocessing module to obtain a multi-source feature map. The dimension unification module is configured to perform feature extraction processing on the multi-source feature map through the feature extraction module to obtain a capacity feature map, and perform dimension unification processing on the capacity feature map to obtain an advanced feature map; The capacity detection module is configured to perform capacity detection processing on the high-level feature map to obtain the capacity detection result of the multi-source capacity data; The device is further configured to: Obtain positive and negative samples. Each positive sample includes the capacity data feature map of two similar multi-source capacity data, and each negative sample includes the capacity data feature map of two dissimilar multi-source capacity data. The positive and negative samples are processed by the preprocessing module, the feature extraction module, and the linear layer, and the feature representation distances of the positive and negative samples are determined. The training loss value is determined based on the feature representation distance between the positive sample and the feature representation distance between the negative sample; the smaller the feature representation distance of the positive sample, the smaller the training loss value; the larger the feature representation distance of the negative sample, the smaller the training loss value. The preprocessing module and the feature extraction module are trained based on the training loss value.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the capacity detection method according to any one of claims 1-9.

12. An electronic device, characterized in that, include: processor; Memory for storing the executable instructions of the processor; The processor is configured to execute the capacity detection method according to any one of claims 1-9 by executing the executable instructions.

Images