Traffic situation prediction methods, devices, computer program products and storage media

Abstract

This invention provides a traffic situation prediction method, apparatus, computer program product, and storage medium. The method includes: determining a first tensor set of a deep cellular automaton corresponding to the first geographical region based on road network data of a first geographical region and a traffic situation map of a second geographical region; inputting the first tensor set into a traffic situation prediction network to obtain a target tensor output by the traffic situation prediction network; and determining a traffic situation prediction map of the first geographical region based on the target tensor. This invention can quickly predict the road network traffic situation of the entire area of ​​a first geographical region based on a deep cellular automaton and a traffic situation map of a portion of the first geographical region, allowing users to promptly obtain information on traffic flow.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a traffic situation prediction method, apparatus, computer program product, and storage medium. Background Technology

[0002] Traffic congestion is a frequent occurrence on city roads, resulting in low vehicle traffic efficiency.

[0003] In existing technologies, road network traffic conditions can intuitively reflect traffic operation, thus they can be used for urban planning, road planning, population planning, etc. They can also be used to analyze road traffic congestion at a specific moment or over a certain period, and can generate historical traffic data to facilitate analysis of urban road congestion, thereby enabling the adoption of effective measures to alleviate traffic congestion and ensure travel. However, how to quickly predict road network traffic conditions is a problem that urgently needs to be solved. Summary of the Invention

[0004] This invention provides a traffic situation prediction method, device, computer program product, and storage medium to solve the problem of how to quickly predict the traffic situation of a road network.

[0005] This invention provides a traffic situation prediction method, comprising:

[0006] Based on road network data of the first geographic region and traffic situation map of the second geographic region, a first tensor set of the deep cellular automaton corresponding to the first geographic region is determined; wherein, the first tensor set includes initial weight values ​​of at least one channel; each channel includes at least one traffic situation type and survival status; the survival status is used to indicate whether the cell is alive; the second geographic region is a part of the first geographic region;

[0007] The first tensor set is input into the traffic situation prediction network to obtain the target tensor output by the traffic situation prediction network; wherein, the target tensor includes the weight prediction value of each channel; the traffic situation prediction network is obtained by training with road network data of the sample geographical area and traffic situation map of the sample geographical area as training samples.

[0008] Based on the target tensor, a traffic situation prediction map for the first geographical region is determined.

[0009] According to a traffic situation prediction method provided by the present invention, the step of inputting the first tensor set into a traffic situation prediction network to obtain the target tensor output by the traffic situation prediction network includes:

[0010] Based on at least one Sobel filter of the traffic situation prediction network, a depth convolution operation is performed on the first tensor set to obtain the output results of each Sobel filter.

[0011] The first tensor set is input into the maximum value filter of the traffic situation prediction network to obtain the maximum weight of each channel within the target range of each cell in the deep cellular automaton output by the maximum value filter.

[0012] The outputs of each Sobel filter, the maximum weight of each channel within the target range of each cell in the deep cellular automaton output by the maximum value filter, and the first tensor set are concatenated to obtain the second tensor.

[0013] The second tensor is input into at least one layer of a fully connected deep learning network of the traffic situation prediction network for deep convolution operation to obtain a third tensor; wherein, the third tensor includes the weight difference of each channel of each cell in the deep cellular automaton;

[0014] Based on at least one layer of the obstruction layer of the traffic situation prediction network, the weight difference of each channel of the target cell that meets the target condition in the third tensor is set to 0 to obtain the fourth tensor.

[0015] Gradient descent is applied to the weight differences of each channel in the fourth tensor corresponding to the second geographical region to obtain the fifth tensor;

[0016] Based on the first tensor set and the fifth tensor, the target tensor output by the traffic situation prediction network is determined.

[0017] According to a traffic situation prediction method provided by the present invention, the target condition includes at least one of the following:

[0018] The weight of the survival status of each cell within the second range of the target cell does not exceed the first target threshold;

[0019] The random value of the target cell exceeds the second target threshold;

[0020] The geographical region corresponding to the target cell is not a road.

[0021] According to a traffic situation prediction method provided by the present invention, the step of performing gradient descent on the weight differences of each channel of the cell corresponding to the second geographical region in the fourth tensor to obtain the fifth tensor includes:

[0022] Gradient descent is performed on the weight differences of each channel of the cell corresponding to the second geographic region in the fourth tensor using formulas (1) and (2);

[0023] x i,t+1 =x i,t +β*δ update,i,t -C*δ pre,t (1)

[0024] δ pre,t =-p i (1-h i,t (2)

[0025] Wherein, the x i,t+1 x represents the weight of the i-th cell at iteration step t+1. i,t δ represents the weight of the i-th cell at iteration step t. update,i,t δ represents the weight difference obtained by the i-th cell in the t-th iteration. pre,t p represents the weight difference obtained by the cell corresponding to the second geographical region in the t-th iteration. i h represents the weight corresponding to the i-th cell. i,t This represents the output value of the i-th cell after the t-th iteration; β and C are both constants.

[0026] According to a traffic situation prediction method provided by the present invention, determining a traffic situation prediction map of the first geographical region based on the target tensor includes:

[0027] Based on the weighted predicted values ​​of each traffic situation type in the target tensor, the probability of each traffic situation type corresponding to each cell in the deep cellular automaton is determined.

[0028] Based on the probability of each traffic situation type corresponding to each cell in the deep cellular automaton, a traffic situation prediction map for the first geographic region is determined.

[0029] According to a traffic situation prediction method provided by the present invention, the traffic situation type includes at least one of the following: smooth traffic, light congestion, moderate congestion, and heavy congestion.

[0030] The present invention also provides a traffic situation prediction device, comprising:

[0031] The first determining module is used to determine the first tensor set of the deep cellular automaton corresponding to the first geographical region based on the road network data of the first geographical region and the traffic situation map of the second geographical region; wherein, the first tensor set includes the initial weight values ​​of at least one channel; each channel includes at least one traffic situation type and survival status; the survival status is used to indicate whether the cell is alive; the second geographical region is a part of the first geographical region;

[0032] The prediction module is used to input the first tensor set into the traffic situation prediction network to obtain the target tensor output by the traffic situation prediction network; wherein, the target tensor includes the weight prediction value of each channel; the traffic situation prediction network is obtained by training with road network data of the sample geographical area and traffic situation map of the sample geographical area as training samples.

[0033] The second determining module is used to determine a traffic situation prediction map of the first geographical region based on the target tensor.

[0034] The present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein when the program instructions are executed by a computer, the computer is able to execute the program instructions to implement the steps of any of the traffic situation prediction methods described above.

[0035] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the traffic situation prediction methods described above.

[0036] The traffic situation prediction method provided by this invention obtains road network data of a first geographical region and traffic situation map of a second geographical region in advance. It then inputs the first tensor set of the deep cellular automaton corresponding to the first geographical region into the traffic situation prediction network for iterative prediction to obtain the target tensor corresponding to the first geographical region. This allows for the determination of the traffic situation prediction map of the first geographical region, enabling rapid prediction of the road network traffic situation of the entire first geographical region based on the deep cellular automaton and the traffic situation map of a portion of the first geographical region. This facilitates timely information on traffic flow for users. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0038] Figure 1 This is one of the flowcharts illustrating the traffic situation prediction method provided by the present invention;

[0039] Figure 2 This is the second flowchart of the traffic situation prediction method provided by the present invention;

[0040] Figure 3 This is the third flowchart of the traffic situation prediction method provided by the present invention;

[0041] Figure 4 This is a schematic diagram of the iterative prediction process of the traffic situation prediction method provided by the present invention;

[0042] Figure 5 This is a schematic diagram of the traffic situation prediction device provided by the present invention;

[0043] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0045] The following combination Figures 1-4 The traffic situation prediction method of the present invention is described.

[0046] Figure 1 As shown in Figure 1, one of the flowcharts of the traffic situation prediction method provided by the present invention includes:

[0047] Step 110: Based on the road network data of the first geographic region and the traffic situation map of the second geographic region, determine the first tensor set of the deep cellular automaton corresponding to the first geographic region; wherein, the first tensor set includes the initial weight values ​​of at least one channel; each channel includes at least one traffic situation type and survival status; the survival status is used to indicate whether the cell is alive; the second geographic region is a part of the first geographic region.

[0048] It should be noted that the traffic situation prediction method provided by this invention can be applied to scenarios where road network traffic situations are predicted. The executing entity of the traffic situation prediction method provided by this invention can be the traffic situation prediction device provided by this invention, such as an electronic device, or the control module in the traffic situation prediction device for executing the traffic situation prediction method.

[0049] Optionally, the first geographic region is the entire area where traffic conditions are to be predicted. The road network data for the first geographic region refers to the distribution of roads, lakes, parks, and buildings within the first geographic region; that is, the location of each cell in the deep cellular automaton corresponding to the first geographic region is pre-set. The second geographic region is a partial geographic area within the first geographic region, and the traffic situation map for the second geographic region refers to the traffic conditions of roads at different locations within the second geographic region.

[0050] It's important to note that deep cellular automata refer to cellular automata applied in two-layer fully connected deep learning networks. A cellular automaton consists of cells, cell states, neighborhoods, and state update rules. A cell is the most basic unit of a cellular automaton; a cell state refers to the state a cell can only have at any given time—either alive or not alive; a neighborhood is the region of cells spatially adjacent to the cell; and a state update rule is the state transition function that determines the next state of the cell based on its current state and the states of its neighboring cells. Each cell in a deep cellular automaton can only know the survival status of other cells within a certain neighborhood.

[0051] Optionally, the first tensor set refers to the set of tensors corresponding to each cell in the deep cellular automaton corresponding to the first geographic region. A tensor represents the data set of each cell in the deep cellular automaton; each tensor includes weight values ​​for at least one channel; a channel represents the parameters in each tensor, and the weight value of a channel refers to the parameter value of each channel.

[0052] For example, a tensor might have 16 parameters, meaning it comprises 16 channels. The weights of the first four channels represent four different traffic situation types, such as smooth traffic, light congestion, moderate congestion, and heavy congestion. The weight of the fifth channel is represented by alpha, which indicates the survival status of the corresponding cell within the tensor. If the alpha value is greater than 0.1, it means the cell is alive. Cell survival indicates that the cell is located within a road area in the first geographic region, while cell inactivity indicates that the cell is located outside a road area, such as within a building, lake, or park area. The remaining 11 channels of this tensor, besides the aforementioned five channels, refer to parameters iteratively calculated in the deep learning network.

[0053] In practice, a deep cellular automaton can contain multiple cells, the number of which is determined by the number of pixels in the traffic situation map to be predicted corresponding to the first geographic region. Assuming the traffic situation map to be predicted for the first geographic region has a length and width of 80 pixels (represented as (80, 80)), then the deep cellular automaton contains 6400 cells. The first tensor set has the same shape as the deep cellular automaton, which is (80, 80, 16), where 80 represents the number of pixels in the traffic situation map to be predicted within the first geographic region, and 16 represents the number of channels. There is a one-to-one correspondence between the deep cellular automaton and the first tensor set corresponding to the first geographic region.

[0054] Optionally, the first tensor set of the deep cellular automaton corresponding to the first geographic region is determined based on the road network data of the first geographic region and the traffic situation map of the second geographic region. Specifically, the weight values ​​of the first four channels in the first tensor set of the deep cellular automaton corresponding to the first geographic region are obtained from the traffic situation map of the second geographic region, the weight value of the fifth channel is obtained from the survival status of each cell in the first geographic region based on the road network data of the first geographic region, and the remaining channel values ​​are obtained from the road network data of the second geographic region, thereby determining the first tensor set of the deep cellular automaton corresponding to the first geographic region.

[0055] Optionally, traffic conditions may include, but are not limited to, the following four types: smooth traffic, light congestion, moderate congestion, and heavy congestion.

[0056] Step 120: Input the first tensor set into the traffic situation prediction network to obtain the target tensor output by the traffic situation prediction network; wherein, the target tensor includes the weight prediction value of each channel; the traffic situation prediction network is obtained by training with road network data of the sample geographical area and traffic situation map of the sample geographical area as training samples.

[0057] Optionally, the first tensor set is input into the traffic situation prediction network, which iteratively predicts and calculates the first tensor set to obtain the target tensor output by the traffic situation prediction network. The weight prediction value of the target tensor is the weight value of each channel obtained after the first tensor set has been iteratively calculated by the traffic situation prediction network.

[0058] Optionally, the shape of the target tensor output by the traffic situation prediction network is consistent with the shape of the deep cellular automaton. For example, the shape of the target tensor can be (80, 80, 16).

[0059] Step 130: Based on the target tensor, determine the traffic situation prediction map of the first geographical region.

[0060] Optionally, based on the weighted predicted values ​​of the target tensor, the first four weighted predicted values ​​of the tensor corresponding to each cell in the target tensor are calculated using a logistic regression (softmax) function to obtain the probabilities of the first four weighted predicted values. The highest probability is selected as the traffic situation type of the location of each cell in the target tensor, thereby obtaining the traffic situation prediction map corresponding to the first geographic region. This traffic situation prediction map includes at least one of the four traffic situation types: smooth traffic, light congestion, moderate congestion, and heavy congestion.

[0061] The traffic situation prediction method provided by this invention obtains road network data of a first geographical region and traffic situation map of a second geographical region in advance. It then inputs the first tensor set of the deep cellular automaton corresponding to the first geographical region into the traffic situation prediction network for iterative prediction to obtain the target tensor corresponding to the first geographical region. This allows for the determination of the traffic situation prediction map of the first geographical region, enabling rapid prediction of the road network traffic situation of the entire first geographical region based on the deep cellular automaton and the traffic situation map of a portion of the first geographical region. This facilitates timely information on traffic flow for users.

[0062] Optionally, Figure 2 This is the second flowchart illustrating the traffic situation prediction method provided by the present invention, as shown below. Figure 2 As shown, the method includes:

[0063] Step 210: Based on the road network data of the first geographic region and the traffic situation map of the second geographic region, determine the first tensor set of the deep cellular automaton corresponding to the first geographic region; wherein, the first tensor set includes the initial weight values ​​of at least one channel; each channel includes at least one traffic situation type and survival status; the survival status is used to indicate whether the cell is alive; the second geographic region is a part of the first geographic region.

[0064] Optionally, the explanation and description of step 210 can refer to the explanation and description of step 110 above, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0065] Step 220: Based on at least one Sobel filter of the traffic situation prediction network, perform a depthwise convolution operation on the first tensor set to obtain the output results of each Sobel filter.

[0066] Optionally, the Sobel filter may include at least one of three different Sobel filters: 3x3, 5x5, and 7x7. The Sobel filter operates only on the convolution object of the current pixel and its neighboring pixels; each pixel can only know information about other pixels in its neighborhood. The output of each Sobel filter includes the results of the horizontal depthwise convolution operation and the results of the vertical depthwise convolution operation.

[0067] Step 230: Input the first tensor set into the maximum value filter of the traffic situation prediction network to obtain the maximum weight of each channel within the target range of each cell in the deep cellular automaton output by the maximum value filter.

[0068] Optionally, the maximum value filter refers to the maximum weight of each channel of each cell in a deep cellular automaton within a certain neighborhood. Therefore, the output of the maximum value filter is the maximum weight of each channel within the target range of each cell in the deep cellular automaton.

[0069] Step 240: The output results of each Sobel filter, the maximum weight of each channel within the target range of each cell in the deep cellular automaton output by the maximum value filter, and the first tensor set are concatenated to obtain the second tensor.

[0070] Optionally, the target range for each cell refers to the location of cells within a certain neighborhood around each cell, for example, the target range is a 3*3 neighborhood around each cell.

[0071] Optionally, the outputs of the Sobel filter, the maximum value filter, and the first tensor set are concatenated in the following order to obtain the second tensor. For example, the shape of the second tensor is (80, 80, 16*8). It should be noted that the output of a Sobel filter includes the weight values ​​of 16 channels output horizontally and 16 channels output vertically, i.e., 16*2 channel weight values. Therefore, the outputs of the 3*3, 5*5, and 7*7 Sobel filters are 16*2*3 channel weight values. The maximum value filter outputs the maximum weight values ​​of the 16 channels in each cell within a 3*3 neighborhood, thus the maximum value filter outputs 16 channel weight values. The first tensor set corresponds to the 16 channel weight values. Therefore, the shape of the second tensor obtained by concatenating the outputs of the Sobel filter, the maximum value filter, and the first tensor set is (80, 80, 16*8).

[0072] Step 250: Input the second tensor into at least one layer of the fully connected deep learning network of the traffic situation prediction network to perform a deep convolution operation to obtain a third tensor; wherein, the third tensor includes the weight difference of each channel of each cell in the deep cellular automaton.

[0073] Optionally, the shape of the output of the fully connected deep learning network is consistent with the shape of the deep cellular automaton, i.e., (80, 80, 16). The output of the fully connected deep learning network is the weight difference between the channels of each cell.

[0074] Step 260: Based on at least one layer of the obstruction layer of the traffic situation prediction network, set the weight difference of each channel of the target cell that meets the target condition in the third tensor to 0 to obtain the fourth tensor.

[0075] Optionally, at least one masking layer is used to indicate that the target cells meeting the target conditions have undergone three treatments, wherein the target conditions include at least one of the following:

[0076] Within the second range of the target cell, the weights representing the survival status of each cell do not exceed the first target threshold. The second range includes the target cell and the range within a certain neighborhood around it.

[0077] The random value of the target cell exceeds the second target threshold;

[0078] The geographical region corresponding to the target cell is not a road.

[0079] For example, a cell is considered alive when the weight value of the 5th channel is greater than the first target threshold. Taking a second target threshold of 50% as an example, in this case, the target cell has a 50% probability of not being updated; that is, when the random value of the target cell is greater than 0.5, the target cell will not be updated; while when the random value of the target cell does not exceed 0.5, the cell will be updated.

[0080] Specifically, the third tensor output by the fully connected deep learning network passes through at least one occlusion layer of the traffic situation prediction network, and the at least one occlusion layer processes the third tensor sequentially. For example, the first target threshold is set to 0.1, and the second target threshold is set to 50%.

[0081] 1) When the third tensor passes through the first shielding layer, if there are no living cells in the second range including itself and a certain neighborhood, that is, the weights representing the survival status of each cell in the second range do not exceed 0.1, then the weight difference of each channel of each cell in the second range is set to 0.

[0082] 2) The third tensor passes through the second occlusion layer, which is a random occlusion layer. Each cell has a 50% probability of not being updated. If the random value of each cell is greater than 0.5, the weight difference of the 16 channels of the third tensor corresponding to each cell is multiplied by 1; otherwise, it is multiplied by 0. If the random value of each cell does not exceed 0.5, the weight difference of the 16 channels of the third tensor corresponding to each cell is multiplied by 0, that is, the weight difference of each channel in the third tensor is set to 0.

[0083] 3) The third tensor passes through the third occlusion layer, which is a validity occlusion layer. When each cell is located outside the road area in the geographic region, the weight difference of the 16 channels of the third tensor processed by the second occlusion layer is multiplied by 0, that is, the weight difference of each channel in the third tensor is set to 0. In other words, cells located outside the road area, such as buildings, parks, or lakes, are not updated. When each cell is located within the road area in the geographic region, the weight difference of the 16 channels of the third tensor processed by the second occlusion layer is multiplied by 1, that is, cells located within the road area are updated. Thus, after passing through three occlusion layers in sequence, the third tensor becomes the fourth tensor.

[0084] It should be noted that in the first iteration, the third tensor does not pass through the first occlusion layer, but is processed directly through the second occlusion layer.

[0085] Step 270: Perform gradient descent on the weight differences of each channel of the cell corresponding to the second geographical region in the fourth tensor to obtain the fifth tensor.

[0086] Optionally, the traffic conditions of roads at different locations in the second geographic region are known in advance. An additional gradient descent operation is performed based on the weight differences of each channel of the cell corresponding to the second geographic region in the fourth tensor to obtain a set of weight differences. The obtained set of weight differences is used as the fifth tensor.

[0087] Step 280: Based on the first tensor set and the fifth tensor, determine the target tensor output by the traffic situation prediction network.

[0088] Optionally, after each iteration, the weight values ​​of each channel in the first tensor set are added to the weight differences of each channel in the fifth tensor set, and this sum is used as the initial value for the next iteration. This process is repeated sequentially according to steps 210-270 above. After the iteration is complete, the result of the last iteration is used as the target tensor, which is the output of the traffic situation prediction network. The shape of this target tensor is (80, 80, 16).

[0089] Step 290: Based on the target tensor, determine the traffic situation prediction map of the first geographical region.

[0090] Optionally, the explanation and description of step 290 can refer to the explanation and description of step 130 above, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0091] The traffic situation prediction method provided by this invention adds the result of each iteration to the weight value of each channel of each cell before the iteration as the initial value for the next iteration. Moreover, each iteration updates the cells in a certain neighborhood around each cell, thereby determining the traffic situation prediction map of the first geographical region. This enables the rapid prediction of the road network traffic situation of the entire area of ​​the first geographical region based on the traffic situation map of a part of the first geographical region and the deep cellular automata, making it convenient for users to know the traffic flow level in a timely manner.

[0092] Optionally, gradient descent is performed on the weight differences of each channel of the cell corresponding to the second geographic region in the fourth tensor using formulas (1) and (2).

[0093] x i,t+1 =x i,t +β*δ update,i,t -C*δ pre,t (1)

[0094] δ pre,t =-p i (1-h i,t (2)

[0095] Wherein, the x i,t+1 Let x represent the weight value of the i-th cell at iteration step t+1. i,t δ represents the weight value of the i-th cell at iteration t. update,i,t δ represents the weight difference obtained by the i-th cell in the t-th iteration. pre,t p represents the weight difference obtained by the cell corresponding to the second geographical region in the t-th iteration. i h represents the weight value corresponding to the i-th cell. i,t This represents the output value of the i-th cell after the t-th iteration; β and C are both constants.

[0096] Specifically, the second geographic region is a pre-defined region whose road traffic conditions are known in advance. During the iterative calculation of the first tensor set through the traffic situation prediction network, the obtained fourth tensor is processed through three layers of shading. Then, the weight difference of each channel of the cell corresponding to the second geographic region in the fourth tensor is subjected to a gradient descent operation using formula (2) to obtain an additional set of difference values ​​δ. pre,t For the cells corresponding to the second geographic region, the difference δ is obtained by performing gradient descent operation on them using formula (1). pre,t Subtract it. If the cell corresponding to the second geographic region is located within the road's range, then the difference δ obtained after performing gradient descent on it... pre,tIt must not be 0; if the cell corresponding to the second geographic region is located within a building, park, or lake area, then the difference δ obtained after performing gradient descent on it will be... pre,t is 0.

[0097] Optionally, Figure 3 This is the third flowchart illustrating the traffic situation prediction method provided by the present invention, as shown below. Figure 3 As shown, the method includes:

[0098] Step 310: Based on the road network data of the first geographic region and the traffic situation map of the second geographic region, determine the first tensor set of the deep cellular automaton corresponding to the first geographic region; wherein, the first tensor set includes the initial weight values ​​of at least one channel; each channel includes at least one traffic situation type and survival status; the survival status is used to indicate whether the cell is alive; the second geographic region is a part of the first geographic region.

[0099] Step 320: Input the first tensor set into the traffic situation prediction network to obtain the target tensor output by the traffic situation prediction network; wherein, the target tensor includes the weight prediction value of each channel; the traffic situation prediction network is obtained by training with road network data of the sample geographical area and traffic situation map of the sample geographical area as training samples.

[0100] Optionally, the explanations and descriptions of steps 310-320 can refer to the explanations and descriptions of steps 110-120 above, and the same technical effect can be achieved. To avoid repetition, they will not be repeated here.

[0101] Step 330: Based on the weight prediction values ​​of each traffic situation type in the target tensor, determine the probability of each traffic situation type corresponding to each cell in the deep cellular automaton.

[0102] Optionally, the weight values ​​corresponding to the first four tensors in the target tensor are selected as the predicted weight values ​​for each traffic situation type. The weight values ​​corresponding to the first four tensors are then normalized using the softmax function to obtain the probability of each traffic situation type corresponding to each cell in the deep cellular automaton.

[0103] Step 340: Based on the probability of each traffic situation type corresponding to each cell in the deep cellular automaton, determine the traffic situation prediction map of the first geographical area.

[0104] Optionally, based on the probabilities of each traffic situation type corresponding to each cell in the obtained deep cellular automaton, the traffic situation type corresponding to the highest probability among the probabilities of each traffic situation type corresponding to each cell is selected as the traffic situation of the cell's location, thereby obtaining a traffic situation prediction map of the first geographic region.

[0105] The traffic situation prediction method provided by this invention obtains the probability of each weight prediction value by normalizing the predicted weight value through the softmax function based on the weight prediction value of each traffic situation type in the target tensor. This enables the determination of the traffic situation prediction map of the first geographic region. Based on the traffic situation map of a part of the first geographic region and the deep cellular automata, the method can quickly predict the road network traffic situation of the entire area of ​​the first geographic region, making it convenient for users to know the traffic flow level in a timely manner.

[0106] Optionally, Figure 4 This is a schematic diagram of the iterative prediction process of the traffic situation prediction method provided by the present invention, as shown below. Figure 4 As shown, the specific implementation steps are as follows:

[0107] Step 1: Determine the initial values ​​for iteration. In this invention, the length and width of the first geographic region are both 80 pixels. Each pixel corresponds to a length of 20 meters in a real-world application scenario, meaning the size of the first geographic region is 1.6 * 1.6 square kilometers. For example, the deep cellular automaton corresponding to the first geographic region in this invention comprises 3 * 3 cells, and the shape of the first tensor set of the deep cellular automaton corresponding to the first geographic region is (3, 3, 16).

[0108] To clearly illustrate the iterative process of each cell in a deep cellular automaton, we will focus on one cell. The tensor corresponding to this cell can be referred to as the first tensor set, which has the shape (1, 1, 16). This cell contains 16 channels, each corresponding to a weight value. The weight values ​​of the first four channels represent four different traffic conditions: smooth, lightly congested, moderately congested, and heavily congested. The weight value of the fifth channel is represented by alpha, indicating the cell's survival status; if the alpha value is greater than 0.1, the cell is considered alive. The remaining 11 channels represent the parameters iteratively calculated in the deep learning network. It should be noted that in the first iteration, all weight values ​​in the first tensor set are initialized to 0.

[0109] Step 2: The first tensor set undergoes a depthwise convolution operation with filters. The first tensor set corresponding to the cell is first passed through three different Sobel filters: 3x3, 5x5, and 7x7. When the first tensor set passes through a 3x3 Sobel filter 401, it obtains 16 weight values ​​for each channel corresponding to the horizontal and vertical outputs of the 3x3 Sobel filter. When the first tensor set passes through a 5x5 Sobel filter 402, it obtains 16 weight values ​​for each channel corresponding to the horizontal and vertical outputs of the 5x5 Sobel filter. When the first tensor set passes through a 7x7 Sobel filter 403, it obtains 16 weight values ​​for each channel corresponding to the horizontal and vertical outputs of the 7x7 Sobel filter. Then, the first tensor set passes through a 3x3 maximum value filter 404 to obtain the maximum value of the 16 channel weight values ​​for each cell within a 3x3 neighborhood of the cell. The output of the maximum value filter is 16 weight values; the first tensor set corresponds to the 16 channel weight values.

[0110] Step 3: The output results of each filter are concatenated with the first tensor set. The output results of the Sobel filter, the maximum weight of each channel within the 3*3 neighborhood of the maximum value filter, and the first tensor set 405 corresponding to the cell are concatenated by the concatenation module 406 in the order of the Sobel filter output results, the maximum value filter output results, and the first tensor set corresponding to the cell, to obtain the second tensor corresponding to the cell. The shape of the second tensor is (1, 1, 16*8).

[0111] It should be noted that the output of a single Sobel filter includes the weight values ​​of 16 channels output horizontally and 16 channels output vertically, i.e., the weight values ​​of 16*2 channels. Therefore, the outputs of the three different Sobel filters mentioned above correspond to the weight values ​​of 16*2*3 channels. The maximum value filter outputs the maximum weight values ​​of the 16 channels in each cell within a 3*3 neighborhood, and the first tensor set corresponding to this cell contains the weight values ​​of the 16 channels. Therefore, the shape of the second tensor obtained by concatenating the output of the Sobel filter, the maximum weight values ​​of each channel within a 3*3 neighborhood of the maximum value filter, and the first tensor set corresponding to this cell is (1, 1, 16*8).

[0112] Step 4: The second tensor undergoes a deep convolution operation through a fully connected deep learning network. The second tensor is input into a two-layer fully connected deep learning network. The second tensor passes through the first deep learning network layer 407, where it is transformed from 128 weight values ​​to 256 weight values. Then, it passes through the second deep learning network layer 408, where it is transformed from 256 weight values ​​to 16 weight difference values. That is, after the second tensor passes through the fully connected deep learning network, it obtains 16 weight difference values, which serve as the third tensor of the deep cellular automaton.

[0113] Step 5: The third tensor is processed by the occlusion layers. The third tensor output by the fully connected deep learning network passes through three occlusion layers in sequence, and each of these three layers processes the third tensor sequentially.

[0114] First, when the third tensor passes through the first occlusion layer 409, if there are no live cells within the second range including itself and its surrounding 3*3 neighborhood, meaning the weights representing the survival status of each cell within the second range do not exceed the first target threshold, then the weight differences of the 16 channels for each cell within the second range are set to 0. It should be noted that in the first iteration, the third tensor does not pass through the first occlusion layer but is processed directly through the second occlusion layer.

[0115] Next, the third tensor passes through the second occlusion layer 410, which is a random occlusion layer. There is a 50% probability that the cell will not be updated. If the random value of the cell is greater than the second target threshold, the weight difference of the 16 channels of the third tensor corresponding to the cell is multiplied by 1; otherwise, it is multiplied by 0. If the random value of the cell does not exceed the second target threshold of 0.5, the weight difference of the 16 channels of the third tensor corresponding to the cell is multiplied by 0, that is, the weight difference of each channel in the third tensor is set to 0.

[0116] Then, the third tensor passes through the third occlusion layer 411, which is a legality occlusion layer. When the cell's location in the geographic region is outside the road area, the 16 weight differences of the third tensor processed by the second occlusion layer are multiplied by 0, that is, the weight differences of each channel of the cell in the third tensor are set to 0. In other words, cells located in non-road areas such as buildings, parks, or lakes are not updated. When the cell's location in the geographic region is within the road area, the 16 weight differences of the third tensor processed by the second occlusion layer are multiplied by 1, that is, cells located within the road area are updated. After the third tensor passes through three occlusion layers in sequence, the fourth tensor corresponding to the cell is obtained.

[0117] Step 6: Perform gradient descent on the cells in the second geographic region. If the cell's location in the first geographic region is also in the second geographic region, meaning the traffic situation type of the cell's location is known, then perform an additional gradient descent operation on the 16 weight differences of the fourth tensor corresponding to the cell through the gradient descent module 412 to obtain the fifth tensor corresponding to the cell.

[0118] Step 7: Determine the initial value for the next iteration. The weight differences between channels in the first tensor set corresponding to the cell and the weight differences between channels in the fifth tensor are summed using the summing module 413. The result is used as the initial value for the next iteration. Steps 2-6 are then performed sequentially for a total of 128 iterations. The result of the 128th iteration is used as the target tensor, which is the output of the traffic situation prediction network. The shape of this target tensor is (3, 3, 16).

[0119] Step 8: Determine the traffic situation prediction map for the first region. Select the first four weight values ​​of the tensor corresponding to each cell in the target tensor, i.e. (3, 3, 4). Normalize these four weight values ​​using the softmax function module 414 to obtain the probabilities corresponding to the four weight values. Output the traffic situation type corresponding to the highest probability among the four probabilities as the final result through the final result module 415, thereby obtaining the traffic situation prediction map for the first geographical region.

[0120] It should be noted that the parameter training in this invention uses a loss function, which is calculated based on the output of the traffic situation prediction network after 128 iterations, the KL divergence of the target result corresponding to the first geographic region, and the alpha value of each cell in the deep cellular automaton. The target result is the actual traffic situation map of the first geographic region. The loss function can be calculated using the following formula (3):

[0121]

[0122] Among them, h i,T p is the probability value obtained by processing the first four weights of the tensor corresponding to the i-th cell using the softmax function. i The target probability of the traffic situation type corresponding to the i-th cell in the first geographic region; alive i The survival status of the i-th cell is defined as follows: cells located within buildings, parks, etc., have an survival value of 0, while cells located within roads have an survival value of 1; α i,T Let represent the survival status of the i-th cell in the T-th iteration.

[0123] Finally, the calculated loss function is input into the adaptive moment estimator (Adam) optimizer for training.

[0124] The traffic situation prediction method provided by this invention iteratively predicts and calculates a first tensor set through a traffic situation prediction network. During the iteration process, the first tensor set undergoes gradient descent operations via a Sobel filter and a maximum value filter, followed by depthwise convolution calculations via a two-layer fully connected deep learning network, and then processing through three occlusion layers. This updates the cell survival status within a certain neighborhood of each cell in the deep cellular automaton, thereby predicting the traffic situation map of a first geographical region. This method enables rapid prediction of the road network traffic situation for the entire first geographical region based on a deep cellular automaton and a partial geographical area of ​​the first geographical region, facilitating timely information on traffic flow for users.

[0125] The traffic situation prediction device provided by the present invention is described below. The traffic situation prediction device described below can be referred to in correspondence with the traffic situation prediction method described above.

[0126] This invention provides a traffic situation prediction device. Figure 5 This is a schematic diagram of the traffic situation prediction device provided by the present invention, as shown below. Figure 5 As shown, the traffic situation prediction device 500 includes: a first determining module 501, a prediction module 502, and a second determining module 503; wherein,

[0127] The first determining module 501 is used to determine the first tensor set of the deep cellular automaton corresponding to the first geographical region based on the road network data of the first geographical region and the traffic situation map of the second geographical region; wherein, the first tensor set includes the initial weight values ​​of at least one channel; each channel includes at least one traffic situation type and survival status; the survival status is used to indicate whether the cell is alive; the second geographical region is a part of the first geographical region;

[0128] The prediction module 502 is used to input the first tensor set into the traffic situation prediction network to obtain the target tensor output by the traffic situation prediction network; wherein, the target tensor includes the weight prediction value of each channel; the traffic situation prediction network is obtained by training with road network data of the sample geographical area and traffic situation map of the sample geographical area as training samples.

[0129] The second determining module 503 is used to determine a traffic situation prediction map of the first geographical region based on the target tensor.

[0130] The traffic situation prediction device provided by this invention determines the first tensor set of the deep cellular automaton corresponding to the first geographical region by pre-setting road network data of the first geographical region and traffic situation map of the second geographical region. The first tensor set is iteratively calculated through the traffic situation prediction network to obtain the target tensor corresponding to the first geographical region, thereby determining the traffic situation prediction map of the first geographical region. This enables the device to quickly predict the road network traffic situation of the entire area of ​​the first geographical region based on the deep cellular automaton and the traffic situation map of a part of the first geographical region, making it convenient for users to know the traffic flow level in a timely manner.

[0131] Optionally, the prediction module 502 is specifically used for:

[0132] Based on at least one Sobel filter of the traffic situation prediction network, a depth convolution operation is performed on the first tensor set to obtain the output results of each Sobel filter.

[0133] The first tensor set is input into the maximum value filter of the traffic situation prediction network to obtain the maximum weight of each channel within the target range of each cell in the deep cellular automaton output by the maximum value filter.

[0134] The outputs of each Sobel filter, the maximum weight of each channel within the target range of each cell in the deep cellular automaton output by the maximum value filter, and the first tensor set are concatenated to obtain the second tensor;

[0135] The second tensor is input into at least one layer of a fully connected deep learning network of the traffic situation prediction network for deep convolution operation to obtain a third tensor; wherein, the third tensor includes the weight difference of each channel of each cell in the deep cellular automaton;

[0136] Based on at least one layer of the obstruction layer of the traffic situation prediction network, the weight difference of each channel of the target cell that meets the target condition in the third tensor is set to 0 to obtain the fourth tensor.

[0137] Gradient descent is applied to the weight differences of each channel in the fourth tensor corresponding to the second geographical region to obtain the fifth tensor;

[0138] Based on the first tensor set and the fifth tensor, the target tensor output by the traffic situation prediction network is determined.

[0139] Optionally, the target condition includes at least one of the following:

[0140] The weight of the survival status of each cell within the second range of the target cell does not exceed the first target threshold;

[0141] The random value of the target cell exceeds the second target threshold;

[0142] The geographical region corresponding to the target cell is not a road.

[0143] Optionally, the prediction module 502 is specifically used for:

[0144] Gradient descent is performed on the weight differences of each channel of the cell corresponding to the second geographic region in the fourth tensor using formulas (1) and (2);

[0145] x i,t+1 =x i,t +β*δ update,i,t -C*δ pre,t (1)

[0146] δ pre,t =-p i (1-h i,t (2)

[0147] Wherein, the x i,t+1 x represents the weight of the i-th cell at iteration step t+1. i,t δ represents the weight of the i-th cell at iteration step t. update,i,t δ represents the weight difference obtained by the i-th cell in the t-th iteration. pre,t p represents the weight difference obtained by the cell corresponding to the second geographical region in the t-th iteration. i h represents the weight corresponding to the i-th cell. i,t This represents the output value of the i-th cell after the t-th iteration; β and C are both constants.

[0148] Optionally, the second determining module 503 is specifically used for:

[0149] Based on the weighted predicted values ​​of each traffic situation type in the target tensor, the probability of each traffic situation type corresponding to each cell in the deep cellular automaton is determined.

[0150] Based on the probability of each traffic situation type corresponding to each cell in the deep cellular automaton, a traffic situation prediction map for the first geographic region is determined.

[0151] Optionally, the traffic situation type includes at least one of the following: smooth traffic, light congestion, moderate congestion, and heavy congestion.

[0152] Figure 6 This is a schematic diagram of the physical structure of the electronic device provided by the present invention, such as... Figure 6As shown, the electronic device 600 may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other via the communication bus 640. The processor 610 can call logical instructions in the memory 630 to execute a traffic situation prediction method, which includes:

[0153] Based on road network data of the first geographic region and traffic situation map of the second geographic region, a first tensor set of the deep cellular automaton corresponding to the first geographic region is determined; wherein, the first tensor set includes initial weight values ​​of at least one channel; each channel includes at least one traffic situation type and survival status; the survival status is used to indicate whether the cell is alive; the second geographic region is a part of the first geographic region;

[0154] The first tensor set is input into the traffic situation prediction network to obtain the target tensor output by the traffic situation prediction network; wherein, the target tensor includes the weight prediction value of each channel; the traffic situation prediction network is obtained by training with road network data of the sample geographical area and traffic situation map of the sample geographical area as training samples.

[0155] Based on the target tensor, a traffic situation prediction map for the first geographical region is determined.

[0156] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0157] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein when the program instructions are executed by a computer, the computer is able to execute the traffic situation prediction method provided by the above methods, the method comprising:

[0158] Based on road network data of the first geographic region and traffic situation map of the second geographic region, a first tensor set of the deep cellular automaton corresponding to the first geographic region is determined; wherein, the first tensor set includes initial weight values ​​of at least one channel; each channel includes at least one traffic situation type and survival status; the survival status is used to indicate whether the cell is alive; the second geographic region is a part of the first geographic region;

[0159] The first tensor set is input into the traffic situation prediction network to obtain the target tensor output by the traffic situation prediction network; wherein, the target tensor includes the weight prediction value of each channel; the traffic situation prediction network is obtained by training with road network data of the sample geographical area and traffic situation map of the sample geographical area as training samples.

[0160] Based on the target tensor, a traffic situation prediction map for the first geographical region is determined.

[0161] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the traffic situation prediction methods provided above, the method comprising:

[0162] Based on road network data of the first geographic region and traffic situation map of the second geographic region, a first tensor set of the deep cellular automaton corresponding to the first geographic region is determined; wherein, the first tensor set includes initial weight values ​​of at least one channel; each channel includes at least one traffic situation type and survival status; the survival status is used to indicate whether the cell is alive; the second geographic region is a part of the first geographic region;

[0163] The first tensor set is input into the traffic situation prediction network to obtain the target tensor output by the traffic situation prediction network; wherein, the target tensor includes the weight prediction value of each channel; the traffic situation prediction network is obtained by training with road network data of the sample geographical area and traffic situation map of the sample geographical area as training samples.

[0164] Based on the target tensor, a traffic situation prediction map for the first geographical region is determined.

[0165] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0166] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0167] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A traffic situation prediction method, characterized in that, include: Based on road network data of the first geographic region and traffic situation map of the second geographic region, a first tensor set of the deep cellular automaton corresponding to the first geographic region is determined; wherein, the first tensor set includes initial weight values ​​of at least one channel; each channel includes at least one traffic situation type and survival status; the survival status is used to indicate whether the cell is alive; the second geographic region is a part of the first geographic region; The first tensor set is input into the traffic situation prediction network to obtain the target tensor output by the traffic situation prediction network; wherein, the target tensor includes the weight prediction value of each channel; the traffic situation prediction network is obtained by training with road network data of the sample geographical area and traffic situation map of the sample geographical area as training samples. Based on the target tensor, a traffic situation prediction map of the first geographical region is determined; The step of inputting the first tensor set into the traffic situation prediction network to obtain the target tensor output by the traffic situation prediction network includes: Based on at least one Sobel filter of the traffic situation prediction network, a depth convolution operation is performed on the first tensor set to obtain the output results of each Sobel filter. The first tensor set is input into the maximum value filter of the traffic situation prediction network to obtain the maximum weight of each channel within the target range of each cell in the deep cellular automaton output by the maximum value filter. The outputs of each Sobel filter, the maximum weight of each channel within the target range of each cell in the deep cellular automaton output by the maximum value filter, and the first tensor set are concatenated to obtain the second tensor; The second tensor is input into at least one layer of a fully connected deep learning network of the traffic situation prediction network for deep convolution operation to obtain a third tensor; wherein, the third tensor includes the weight difference of each channel of each cell in the deep cellular automaton; Based on at least one layer of the obstruction layer of the traffic situation prediction network, the weight difference of each channel of the target cell that meets the target condition in the third tensor is set to 0 to obtain the fourth tensor. Gradient descent is applied to the weight differences of each channel in the fourth tensor corresponding to the second geographical region to obtain the fifth tensor; Based on the first tensor set and the fifth tensor, the target tensor output by the traffic situation prediction network is determined.

2. The traffic situation prediction method according to claim 1, characterized in that, The target condition includes at least one of the following: The weight of the survival status of each cell within the second range of the target cell does not exceed the first target threshold; The random value of the target cell exceeds the second target threshold; The geographical region corresponding to the target cell is not a road.

3. The traffic situation prediction method according to claim 1, characterized in that, The step of performing gradient descent on the weight differences of each channel in the fourth tensor corresponding to the second geographical region to obtain the fifth tensor includes: Gradient descent is performed on the weight differences of each channel of the cell corresponding to the second geographic region in the fourth tensor using formulas (1) and (2); x i,t+1 =x i,t +b*d update,i,t -C*d pre,t (1) d pre,t =-p i (1-h i,t ) (2) Wherein, the x i,t+1 x represents the weight of the i-th cell at iteration step t+1. i,t δ represents the weight of the i-th cell at iteration t. update,i,t δ represents the weight difference obtained by the i-th cell in the t-th iteration. pre,t p represents the weight difference obtained by the cell corresponding to the second geographical region in the t-th iteration. i h represents the weight corresponding to the i-th cell. i,t This represents the output value of the i-th cell after the t-th iteration; β and C are both constants.

4. The traffic situation prediction method according to claim 1, characterized in that, The step of determining the traffic situation prediction map of the first geographical region based on the target tensor includes: Based on the weighted predicted values ​​of each traffic situation type in the target tensor, the probability of each traffic situation type corresponding to each cell in the deep cellular automaton is determined. Based on the probability of each traffic situation type corresponding to each cell in the deep cellular automaton, a traffic situation prediction map for the first geographic region is determined.

5. The traffic situation prediction method according to claim 1, characterized in that, The traffic situation types include at least one of the following: smooth traffic, light congestion, moderate congestion, and heavy congestion.

6. A traffic situation prediction device, characterized in that, include: The first determining module is used to determine the first tensor set of the deep cellular automaton corresponding to the first geographical region based on the road network data of the first geographical region and the traffic situation map of the second geographical region; wherein, the first tensor set includes the initial weight values ​​of at least one channel; each channel includes at least one traffic situation type and survival status; the survival status is used to indicate whether the cell is alive; the second geographical region is a part of the first geographical region; The prediction module is used to input the first tensor set into the traffic situation prediction network to obtain the target tensor output by the traffic situation prediction network; wherein, the target tensor includes the weight prediction value of each channel; the traffic situation prediction network is obtained by training with road network data of the sample geographical area and traffic situation map of the sample geographical area as training samples. The second determining module is used to determine a traffic situation prediction map of the first geographical region based on the target tensor. The prediction module is specifically used for: Based on at least one Sobel filter of the traffic situation prediction network, a depth convolution operation is performed on the first tensor set to obtain the output results of each Sobel filter. The first tensor set is input into the maximum value filter of the traffic situation prediction network to obtain the maximum weight of each channel within the target range of each cell in the deep cellular automaton output by the maximum value filter. The outputs of each Sobel filter, the maximum weight of each channel within the target range of each cell in the deep cellular automaton output by the maximum value filter, and the first tensor set are concatenated to obtain the second tensor; The second tensor is input into at least one layer of a fully connected deep learning network of the traffic situation prediction network for deep convolution operation to obtain a third tensor; wherein, the third tensor includes the weight difference of each channel of each cell in the deep cellular automaton; Based on at least one layer of the obstruction layer of the traffic situation prediction network, the weight difference of each channel of the target cell that meets the target condition in the third tensor is set to 0 to obtain the fourth tensor. Gradient descent is applied to the weight differences of each channel in the fourth tensor corresponding to the second geographical region to obtain the fifth tensor; Based on the first tensor set and the fifth tensor, the target tensor output by the traffic situation prediction network is determined.

7. A computer program product, characterized in that, The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, enable the computer to perform the steps of the traffic situation prediction method as described in any one of claims 1 to 5.

8. A non-transitory 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 steps of the traffic situation prediction method as described in any one of claims 1 to 5.

Images