Vehicle Internet of Things Data Hierarchical Classification Method Based on Probabilistic Neural Network and Reinforcement Learning

Abstract

The present invention discloses a method for hierarchical classification of Internet of Vehicles (IoV) data based on probabilistic neural network and reinforcement learning, belonging to the field of hierarchical classification of IoV data, which includes the following steps: S1, obtaining data samples in the IoV scenario and establishing a sample data set; S2, calculating the road condition vectors of each roadside according to the established sample data set; S3, deploying agents in each roadside, and each agent makes decisions based on its own and the road condition vectors of adjacent road sides; S4, using a probabilistic neural network to sort the data in the sample data set; S5, uploading the sorted data according to the decisions of the agents to achieve hierarchical classification of IoV data. By adopting the above method for hierarchical classification of IoV data based on probabilistic neural network and reinforcement learning, the present invention can not only classify the IoV data in combination with the actual situation, but also better allocate broadband resources and ensure that emergency data is preferentially uploaded to the cloud.

Description

Technical Field

[0001] The present invention relates to the field of vehicle networking data classification, and in particular to a method for classifying vehicle networking data based on a probabilistic neural network and reinforcement learning. Background Art

[0002] With the development of communication technology and mobile networks, cars are no longer just simple means of transportation. Vehicles equipped with intelligent devices such as wireless sensors, in-vehicle computers, GPS antennas, and radars can collect and process a large amount of data, and at the same time realize information interaction between vehicles. However, in a vehicle networking system, the upload bandwidth is limited. Uploading all information to the cloud for analysis takes a lot of time and is difficult to handle emergencies. Therefore, it is necessary to classify the uploaded data and give priority to uploading emergency data.

[0003] In a vehicle networking system, information such as traffic flow, vehicle speed, and traffic accidents changes over time. It is difficult for a fixed model to adjust itself according to real-time road conditions. Therefore, the model needs to adjust the classification criteria in combination with current road condition information to achieve better bandwidth utilization. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for classifying vehicle networking data based on a probabilistic neural network and reinforcement learning, which can not only classify vehicle networking data according to actual situations, but also better allocate broadband resources and ensure that emergency data is preferentially uploaded to the cloud.

[0005] To achieve the above purpose, the present invention provides a method for classifying vehicle networking data based on a probabilistic neural network and reinforcement learning, including the following steps:

[0006] S1. Obtain data samples in a vehicle networking scenario and establish a sample data set;

[0007] S2. Calculate the road condition vector of each roadside according to the established sample data set;

[0008] S3. Deploy agents in each roadside, and each agent makes a decision based on its own and the road condition vectors of adjacent road sides:

[0009] S4. Use a probabilistic neural network to sort the data in the sample data set;

[0010] S5. The sorted data is uploaded according to the decision of the agent to achieve classification of vehicle networking data.

[0011] Preferably, step S2 is specifically:

[0012] Obtain the traffic flow, the number of vehicles, and the accident situation in the sample dataset, and process and merge these data into a road condition vector. Among them, the traffic flow is a normalized real number between 0 and 1, which means that the average traffic flow within 1 / 2 sampling time before and after the current moment is divided by the maximum traffic flow; the number of vehicles is a real number between 0 and 1, and is calculated using the following formula:

[0013] N = sigmoid(n - n m )

[0014] where, N represents the calculated number of vehicles; n represents the current number of vehicles; n m represents the number of vehicles in the sample dataset; the accident situation is a real number between 0 and 1, which means the proportion of normal traffic lanes in all lanes. Specifically, when the accident situation is 0, it means that all lanes are passing normally; when the accident situation is 1, it means that all lanes cannot pass due to a serious traffic accident or a sudden situation.

[0015] Preferably, in step S3, the agent calculates the decision value according to its own and the road condition vectors of adjacent roadside, and uses the following formula:

[0016] p = π(a|s)

[0017] where, p is a percentage between 0 and 1, indicating that the data in the sample dataset needs to be uploaded to the cloud, and the remaining data is processed locally; π represents the decision network of the agent; is a 9-dimensional vector, which contains three 3-dimensional vectors, namely the road condition vectors of the previous roadside, the current roadside, and the subsequent roadside; a represents the action space taken by the agent when the observation is s.

[0018] Preferably, in step S4, first, the semi-parametric approximation method in GRNN is used to estimate the model parameters:

[0019]

[0020] Thus, the overall calculation process of the model is represented by the following formula:

[0021]

[0022] where, x represents the input vector; x i represents the other vectors except the current calculation vector; c i represents the center vector of each class; c k represents the center vector of the k-th class; f k represents corresponding to c k the kernel learning function; Z k represents corresponding to the center c k the number of data points; fi (x,σ) represents a kernel learning function with center x and smoothing parameter σ; y i is the output value, representing the importance degree of the current data; represents the predicted output of the model; Z i represents the center as c i the number of data points; ∑ i Z i = NV represents the total number of data points in the dataset;

[0023] After calculating the importance degree of each data, all the data in the sample dataset are sorted in descending order.

[0024] Preferably, according to the percentage p obtained in step S3, the first p data among the data obtained in step S4 are uploaded to the cloud of the vehicle network, and the remaining data are processed locally, realizing the hierarchical classification of vehicle network data during data uploading.

[0025] Therefore, the present invention adopts the above-mentioned method for hierarchical classification of vehicle network data based on a probabilistic neural network and reinforcement learning, and has the following beneficial effects:

[0026] By selecting traffic flow, vehicle quantity, and accident situation as evaluation indicators of the current road condition and using a road condition vector to describe the current road condition information, it is possible to better combine the actual situation during subsequent data hierarchical classification.

[0027] By sorting the data through a probabilistic neural network, this network can effectively evaluate the urgency of the data, and thus sort the data in the order from urgent to non-urgent to ensure that urgent data can be preferentially uploaded to the cloud.

[0028] At the same time, using an intelligent agent to select the proportion of data uploaded to the cloud according to the current road condition can better allocate broadband resources, ensure that urgent data can be preferentially uploaded to the cloud, and at the same time can better utilize broadband resources and upload more data under non-congested conditions.

[0029] Next, through embodiments, the technical solution of the present invention will be further described in detail. Specific embodiments

[0030] The following further illustrates the technical solution of the present invention through embodiments.

[0031] Unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the ordinary meaning understood by those of ordinary skill in the field to which the present invention belongs.

[0032] Embodiment 1

[0033] The present invention provides a method for hierarchical classification of Internet of Vehicles data based on probabilistic neural network and reinforcement learning, comprising the following steps:

[0034] S1. Obtain data samples in the Internet of Vehicles scenario and establish a sample data set;

[0035] S2. Calculate the road condition vector of each roadside in the sample data set, and obtain the traffic flow, the number of vehicles, and the accident situation in the sample data set. After processing these data, they are merged into the road condition vector. Among them, the traffic flow is a normalized real number between 0 and 1, which means that the average traffic flow within 1 / 2 sampling moments before and after the current moment is divided by the maximum traffic flow; the number of vehicles is a real number between 0 and 1, and is calculated using the following formula:

[0036] N = sigmoid(n - n m )

[0037] where N represents the calculated number of vehicles; n represents the current number of vehicles; n m represents the number of vehicles in the sample data set; the accident situation is a real number between 0 and 1, which means the proportion of normal traffic lanes in all lanes. Specifically, when the accident situation is 0, all lanes are in normal traffic; when the accident situation is 1, all lanes are unable to pass due to serious traffic accidents or emergencies.

[0038] S3. Deploy agents in each roadside. Each agent makes a decision based on its own and the road condition vectors of adjacent road sides, and uses the following formula:

[0039] p = π(a|s)

[0040] where p is a percentage between 0 and 1, indicating that the data in the sample data set needs to be uploaded to the cloud, and the remaining data is processed locally; π represents the decision-making network of the agent; is a 9-dimensional vector, which contains three 3-dimensional vectors, namely the road condition vectors of the previous roadside, the current roadside, and the next roadside; a represents the action space taken by the agent when the observation is s.

[0041] S4. All the data in the sample data set is used as the information to be uploaded, and is sorted using a probabilistic neural network based on boost. First, the semi-parametric approximation method in GRNN is used to estimate the model parameters:

[0042]

[0043] Thus, the overall calculation process of the model is represented by the following formula:

[0044]

[0045] wherex represents the input vector; x i represents vectors other than the current calculated vector; c i represents the central vector of each class; c k represents the central vector of the k-th class; f k represents corresponding to c k the kernel learning function; Z k represents corresponding to the center c k the number of data points; f i (x,σ) represents the kernel learning function with center x and smoothing parameter σ; y i is the output value, representing the importance degree of the current data; represents the predicted output of the model; Z i represents the center as c i the number of data points; ∑ i Z i Z = NV represents the total number of data points in the dataset.

[0046] After calculating the importance degree of each data, all the data in the dataset are sorted in descending order.

[0047] S5. The sorted data are uploaded according to the decision of the agent to achieve hierarchical classification of vehicle network data. Specifically, according to the percentage p obtained in step S3, the first p data among the data obtained in step S4 are uploaded to the cloud of the vehicle network, and the remaining data are processed locally, so as to achieve hierarchical classification of vehicle network data during data uploading.

[0048] The present invention adopts the above-mentioned method for hierarchical classification of vehicle network data based on probabilistic neural network and reinforcement learning, which can not only classify vehicle network data in combination with the actual situation, but also better allocate broadband resources and ensure that emergency data is preferentially uploaded to the cloud.

[0049] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for grading and classifying Internet of Vehicles data based on probabilistic neural network and reinforcement learning, characterized in that: The following steps are involved: S1. Obtain data samples in the Internet of Vehicles scenario and establish a sample data set; S2. Calculate the road condition vector of each road side according to the established sample data set; S3. An agent is deployed on each roadside. Each agent makes a decision based on the road condition vectors of itself and the adjacent roadside. The following formula is used: p=π(a|s) Among them, p is a percentage between 0 and 1, indicating that the data in the sample data set needs to be uploaded to the cloud, and the rest of the data is processed locally; π represents the decision network of the intelligent agent; is a 9-dimensional vector, which contains three 3D vectors, which are the road condition vectors of the previous road side, the current road side and the next road side; a represents the action space taken by the agent when the observation is s; S4. Use probabilistic neural network to sort the data in the sample data set; first use the semi-parametric approximation method in GRNN to estimate the model parameters: Therefore, the overall calculation process of the model is expressed by the following formula: in, x represents the input vector; x i Represents other vectors besides the current calculation vector; c i Represents the center vector of each class; c k represents the center vector of the kth class; f k Indicates that it corresponds to c k The kernel learning function of Z k Represents the center c k The number of data points; f i (x,σ) represents the kernel learning function with center x and smoothing parameter σ; y i is the output value, indicating the importance of the current data; Represents the predicted output of the model; Z i Indicates the center is c i The number of data points; i Z i =NV represents the total number of data points in the data set; After calculating the importance of each data, sort all the data in the sample data set from large to small; S5. The sorted data is uploaded according to the decision of the intelligent agent to achieve hierarchical classification of Internet of Vehicles data.

2. The method for classifying Internet of Vehicles data based on probabilistic neural network and reinforcement learning according to claim 1, characterized in that: Step S2 is specifically as follows: The traffic volume, number of vehicles, and accident conditions are obtained from the sample data set, and these data are processed and merged into a road condition vector, where the traffic volume is a normalized real number between 0 and 1, which means the average traffic volume in the 1 / 2 sampling moments before and after the current moment divided by the maximum traffic volume; the number of vehicles is a real number between 0 and 1, calculated using the following formula: N=sigmoid(n-n m ) Where N represents the number of vehicles after calculation; n represents the current number of vehicles; n m Represents the number of vehicles in the sample data set; the accident situation is a real number between 0 and 1, which means the proportion of normal traffic lanes to all lanes.

3. The method for grading and classifying Internet of Vehicles data based on probabilistic neural network and reinforcement learning according to claim 2 is characterized in that: According to step S3, the percentage p is obtained, and the first p data of the data obtained in step S4 are uploaded to the cloud of the Internet of Vehicles, and the remaining data are processed locally, so as to realize the hierarchical classification of the Internet of Vehicles data during data upload.

Images