Traffic road condition prediction method, classification model training method and medium

Abstract

The invention discloses a traffic road condition prediction method, a classification model training method and a computer readable storage medium, and the method comprises the steps: obtaining the cell load data and user terminal switching data of a target cell in a target time period, and enabling the target cell to be a cell with a signal covering a target road section; constructing a target feature vector according to the cell load data and the user terminal switching data; and inputting the target feature vector into a preset classification model for identification, and obtaining the congestion level of the target road section in the target time period, the preset classification model is obtained by training by taking a feature vector constructed by load data of a cell and user terminal switching data as model input data and taking label data constructed by an actual congestion level of a road section covered by the cell as a model training label.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a traffic condition prediction method, a classification model training method, and a computer-readable storage medium. Background Art

[0002] At present, common traffic condition perception technologies mainly include camera technology, microwave radar technology, GPS technology, etc. By using these technical means, accurate, real-time, and comprehensive monitoring and analysis of traffic conditions can be achieved. However, the above technical means also have many disadvantages, such as dependence on the number and location of cameras, and being affected by weather conditions. In particular, due to economic costs, the limited number of cameras deployed in the traffic network cannot achieve comprehensive monitoring and perception of road conditions, especially on highways. It is even more difficult to achieve comprehensive monitoring and perception of road conditions. Summary of the invention

[0003] The main purpose of this application is to provide a traffic condition prediction method, a classification model training method, a device and a computer-readable storage medium, aiming to provide a method for combining wireless communication network data such as cell load data and user terminal switching data with a machine learning model to predict the congestion level of a road section, so as to achieve accurate and real-time monitoring and perception of road conditions even in sections without cameras.

[0004] To achieve the above objectives, the present application provides a traffic condition prediction method, the method comprising:

[0005] Acquire cell load data and user terminal switching data of a target cell in a target period, wherein the target cell is a cell whose signal covers a target road section;

[0006] Constructing a target feature vector according to the cell load data and the user terminal switching data;

[0007] The target feature vector is input into a preset classification model for identification to obtain the congestion level of the target road section within the target time period, wherein the preset classification model is trained using a feature vector constructed using cell load data and user terminal switching data as model input data and label data constructed using actual congestion levels of the road section covered by the cell as model training labels.

[0008] To achieve the above purpose, the present application also provides a classification model training method, the classification model training method comprising:

[0009] Acquire a sample data set, wherein the sample data set includes a plurality of sample data, each sample data including cell load data and user terminal switching data of a cell in a time period, and an actual congestion level of a road section covered by the cell signal;

[0010] Constructing feature vectors according to the cell load data and user terminal switching data in each of the sample data, and constructing label data according to the actual congestion level in each of the sample data;

[0011] The feature vector corresponding to each piece of sample data is used as model input data, and the label data corresponding to each piece of sample data is used as model training label, and the classification model to be trained is trained to obtain a target classification model for predicting the road congestion level.

[0012] To achieve the above objectives, the present application also proposes a computer-readable storage medium, on which a traffic condition prediction program is stored. When the traffic condition prediction program is executed by a processor, the steps of the traffic condition prediction method as described above are implemented.

[0013] In an embodiment of the present application, a traffic condition prediction method, a classification model training method, a device and a computer-readable storage medium are provided, wherein the method comprises: obtaining cell load data and user terminal switching data of a target cell in a target time period, wherein the target cell is a cell whose signal covers a target road section; constructing a target feature vector according to the cell load data and user terminal switching data in the target time period; inputting the target feature vector into a preset classification model for classification to obtain the congestion level of the target road section in the target time period, wherein the preset classification model is trained by using a feature vector constructed in advance with the cell load data and user terminal switching data as model input data, and label data constructed with the actual congestion level of the road section covered by the cell as model training labels.

[0014] Since the load data of a cell can reflect the number of users in the cell, and further reflect the number of users on the road section covered by the cell signal, and the user terminal switching data of the cell can reflect the switching in and out of the users between cells, and further reflect the flow speed of users on the road section covered by the cell signal, in the embodiment of the present application, the congestion level of the road section is predicted by combining wireless communication network data such as the cell load data and user terminal switching data with a machine learning model, which can achieve accurate and real-time monitoring and perception of the road condition, and there is no need to set up cameras on the road section, and it can be achieved by using the cells deployed in the mobile communication network, thereby achieving accurate and real-time monitoring and perception of the road condition even on sections without cameras. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic diagram of the flow chart of the first embodiment of the traffic condition prediction method of the present application;

[0016] Figure 2 A schematic diagram of the flow chart of the second embodiment of the traffic condition prediction method of the present application;

[0017] Figure 3 A schematic diagram of the training and storage system architecture of a classification model involved in an embodiment of a classification model training method of the present application;

[0018] Figure 4 A schematic diagram of the structure of a classification model involved in an embodiment of a classification model training method of the present application;

[0019] Figure 5 A schematic diagram of the structure of a traffic condition prediction device involved in an embodiment of the present application;

[0020] Figure 6 A schematic diagram of the structure of a training device for a classification model involved in an embodiment of the present application;

[0021] Figure 7 A schematic diagram of a computer-readable storage medium involved in an embodiment of the present application;

[0022] Figure 8 A schematic diagram of another computer-readable storage medium involved in an embodiment of the present application.

[0023] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0024] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0025] Based on the above structure, various embodiments of the traffic condition prediction method are proposed.

[0026] Reference Figure 1 , Figure 1 This is a flowchart of the first embodiment of the traffic condition prediction method of the present application. The embodiment of the present application provides an embodiment of the traffic condition prediction method. It should be noted that although the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in an order different from that here. In this embodiment, there is no restriction on the execution subject of the traffic condition prediction method. For example, it can be a network element device, a personal computer, a server, a smart phone and other devices. For the convenience of description, the following description of each embodiment is based on the prediction device as the execution subject. In this embodiment, the traffic condition prediction method includes:

[0027] Step S10, obtaining cell load data and user terminal switching data of a target cell in a target time period, wherein the target cell is a cell whose signal covers a target road section.

[0028] The road section for which traffic conditions need to be predicted is referred to as a target road section for distinction. It should be understood that the traffic condition prediction method of this embodiment is applicable to any road section for which traffic conditions need to be predicted, and the following is an explanation using a target road section as an example. There are many ways to determine the target road section, which are not limited in this embodiment. For example, it can be manually pre-selected and configured in the prediction device, or it can be obtained by dividing each road according to the road network data by the prediction device. In a feasible implementation, the road section divided by the highway can be used as the target road section, and the traffic condition prediction of the target road section is performed to specifically solve the problem that it is difficult for cameras on highways to achieve comprehensive monitoring and perception of road conditions.

[0029] A cell refers to a communication cell, such as a cellular cell. The application environment of this embodiment may be, but is not limited to, an LTE / NR network environment, that is, a cell may refer to a cell in an LTE / NR network environment. A cell whose signal covers the target road section is taken as a target cell. It should be noted that the signal of the target cell may be an area covering part or all of the target road section, that is, it is not necessarily a cell whose signal covers the entire area of ​​the target road section as a target cell. In a specific implementation, if the signals of two or more cells cover the target road section, each cell may be taken as a target cell, or the cell with the largest area of ​​the target road section covered by the signal may be taken as a target cell. There are many ways to determine the target cell, which are not limited in this embodiment. For example, it may be manually determined in advance according to the signal coverage range of each cell and the selected target road section and configured in the prediction device, or it may be determined by the prediction device according to the signal coverage range of each cell in the configured cell information and the selected target road section. For example, in one feasible implementation, the prediction device may obtain cell information including the signal coverage of each cell, and the prediction device uses the cell whose signal coverage covers the target road section as the target cell corresponding to the target road section.

[0030] If it is necessary to predict the congestion of the target road section in a certain period of time, the period is referred to as the target period of time for distinction. It should be understood that the traffic condition prediction method of this embodiment is applicable to predicting the congestion of the target road section in any period of time, and the following is an explanation using a target period of time as an example. There are many ways to determine the target period of time, which are not limited in this embodiment. For example, it can be manually pre-selected and configured in the prediction device, or it can be divided by the period division rule set by the prediction device to obtain the target period of time.

[0031] The prediction device may obtain the cell load data and the user equipment (UE) switching data of the target cell within the target period.

[0032] Among them, the cell load data is data that can reflect the load level of the target cell within the target time period, but in this embodiment, it is not limited which data items the cell load data specifically includes. In a specific implementation method, the prediction device can be set as needed to obtain which data that can reflect the load level of the target cell within the target time period. It should be noted that the more congested the target road section is, the more users will use the communication services provided by the target cell in the target road section, and the higher the load level of the target cell will be. Conversely, the smoother the target road section is, the fewer users will use the communication services provided by the target cell in the target road section, and the lower the load level of the target cell will be. Therefore, the cell load data of the target cell is associated with the congestion of the target road section.

[0033] For example, in a feasible implementation, the cell load data may include at least one of the following data: the average number of connected users of the cell radio resource control (RRC), the number of activated users, the uplink medium access control (MAC) layer traffic, the downlink medium access control layer traffic, and the physical resource block (PRB) utilization.

[0034] The user terminal switching data is data that can reflect the number of users who cut in or out of the target cell during the target period, but in this embodiment, it is not limited which data items the user terminal switching data specifically includes. In a specific implementation method, the prediction device can be set as needed to obtain which data reflecting the number of users who cut in or out of the target cell during the target period. It should be noted that the more congested the target road section is, the more users use the communication services provided by the target cell in the target road section, and the slower the user's moving speed will be. Then, the longer the user stays in the signal coverage of the target cell, the fewer users will cut in or out of the target cell during the target period; conversely, the smoother the target road section is, the fewer users use the communication services provided by the target cell in the target road section, and the faster the user's moving speed will be, then, the shorter the user stays in the signal coverage of the target cell, and the more users will cut in or out of the target cell during the target period; so the user terminal switching data of the target cell is associated with the congestion of the target road section.

[0035] For example, in a feasible implementation, the user terminal switching data may include at least one of the following data: the number of times the user terminal successfully switches into the target cell, the number of times the user terminal successfully switches out of the target cell, the number of times the user terminal requests to switch into the target cell, and the number of times the user terminal requests to switch out of the target cell.

[0036] In this embodiment, the prediction device does not limit the way to obtain the cell load data and the user terminal handover data, which can be set as needed. For example, in a feasible implementation, it can be obtained from the network element module, the network management module or the data service module.

[0037] Step S20: constructing a target feature vector according to the cell load data and the user terminal switching data.

[0038] After obtaining the cell load data and user terminal switching data of the target cell in the target time period, the prediction device can use the cell load data and the user terminal switching data to construct a feature vector (hereinafter referred to as a target feature vector for distinction). In a specific implementation, there are many ways to construct the target feature vector, which are not limited in this embodiment and can be set as needed. For example, in a feasible implementation, the step S20 includes:

[0039] Step S201 : pre-processing each item of the cell load data and the user terminal switching data to obtain a characteristic quantity corresponding to each item of the data.

[0040] The cell load data includes at least one item of data, and the user terminal switching data includes at least one item of data. The prediction device performs data preprocessing on each item of data in the cell load data and the user terminal switching data to obtain the feature quantity corresponding to each item of data. The data preprocessing method is not limited in this embodiment. For example, in a feasible embodiment, the preprocessing may include normalization processing.

[0041] Step S202: Combine the feature quantities corresponding to each item of data to obtain a target feature vector.

[0042] The prediction device combines the feature quantities corresponding to each data item to obtain a target feature vector. For example, when M items of data are included, the target feature vector can be expressed as X = [x1, x2, ..., x M ], where xi, i∈{1,…M} represents the feature quantity corresponding to each data item.

[0043] Step S30, inputting the target feature vector into a preset classification model for identification, and obtaining the congestion level of the target road section within the target time period, wherein the preset classification model is trained using a feature vector constructed using cell load data and user terminal switching data as model input data, and label data constructed using actual congestion levels of the road sections covered by the cell as model training labels.

[0044] The preset classification model may be pre-trained and configured in the prediction device. In this embodiment, the model structure and training method of the preset classification model are not limited. For example, the model structure may be a model structure of a commonly used classification model, such as a logistic regression model, a decision tree model, an SVM support vector machine model, etc. The training method of the preset classification model is also not limited. For example, a supervised training method may be used. By using the cell load data and the feature vector constructed by the user terminal switching data related to the congestion situation of the road section as the model input data, and by using the label data constructed by the actual congestion level of the road section covered by the cell as the model training label for training, the preset classification model can learn the mapping relationship between the feature vector constructed by the cell load data and the user terminal switching data and the actual congestion level of the road section, so that the mapping relationship learned in the preset classification model can be used to accurately identify the congestion level of the target road section in the target time period based on the input target feature vector. Multiple congestion levels (hereinafter referred to as preset congestion levels) can be set in advance to represent different degrees of congestion on the road section. The number of preset congestion levels can be set as needed and is not limited here. For example, two preset congestion levels may be set, namely, traffic congestion and traffic flow. It is understandable that the actual congestion level of the road section is one of the preset congestion levels.

[0045] Since the load data of a cell can reflect the number of users in the cell, and further reflect the number of users on the road section covered by the cell signal, and the user terminal switching data of the cell can reflect the switching in and out of the users between cells, and further reflect the flow speed of users on the road section covered by the cell signal, in the embodiment of the present application, the congestion level of the road section is predicted by combining wireless communication network data such as the cell load data and user terminal switching data with a machine learning model, which can achieve accurate and real-time monitoring and perception of the road condition, and there is no need to set up cameras on the road section, and it can be achieved by using the cells deployed in the mobile communication network, thereby achieving accurate and real-time monitoring and perception of the road condition even on sections without cameras.

[0046] Based on the above first embodiment, a second embodiment of the traffic condition prediction method of the present application is proposed. Figure 2 In this embodiment, after step S30, the following steps are further included:

[0047] Step S40: in response to the congestion level being greater than a preset level, adjusting the signal coverage of a cell adjacent to the target cell, wherein the adjusted signal coverage of the cell adjacent to the target cell includes a portion or all of the target road section.

[0048] After identifying the congestion level of the target road section within the target time period, the prediction device can compare the congestion level with a preset level. The preset level is a level set in advance as needed, so that when the congestion level is greater than the preset level, it can indicate that the target road section is relatively congested, and the number of users in the target road section has exceeded the number of users that the target cell can handle. In response to the congestion level being greater than the preset level, the prediction device can adjust the signal coverage of the neighboring cells of the target cell so that the adjusted signal coverage of the neighboring cells includes part or all of the target road section. There are many ways to adjust the signal coverage of the neighboring cells, which are not limited in this embodiment. For example, in a feasible implementation, the signal coverage weight of the neighboring cells can be adjusted toward the target cell to adjust the signal coverage of the neighboring cells to include part or all of the target road section.

[0049] In this embodiment, when the target road section in the target cell is relatively congested and the number of users in the target road section has exceeded the number of users that the target cell can load, the signal coverage range of the adjacent cells of the target cell is adjusted to include part or all of the target road section, so that the adjacent cells of the target cell can share the load of the target cell, thereby achieving "network follows business", adaptively balancing the business load between cells, and improving the perceived experience of users in congested cells.

[0050] Based on the above first and / or second embodiments, a third embodiment of the traffic condition prediction method of the present application is proposed. In this embodiment, the preset classification model includes an input layer, at least one hidden layer and an output layer, and the step S30 includes:

[0051] Step S301: input the target feature vector to the hidden layer through the input layer.

[0052] In this embodiment, a feasible model structure of a preset classification model is proposed, including an input layer, at least one hidden layer and an output layer. Among them, the input layer is the first layer of the neural network, which is used to receive the input signal (value) and pass it to the next layer, that is, the hidden layer in this embodiment, but does not perform any operation on the input signal (value). The neurons (nodes) of the hidden layer convert the input data in different ways, and a hidden layer is a vertically stacked set of neurons. The output layer is the last layer of the neural network, which receives the input from the last hidden layer and outputs a value within a reasonable range.

[0053] After constructing the target feature vector, the prediction device inputs the target feature vector into the hidden layer through the input layer.

[0054] Step S302: After the input vector is processed by the hidden layer, the processed vector is input into the output layer.

[0055] After the target feature vector is input into the hidden layer, the prediction device processes the input vector through the hidden layer. The specific processing method may be different due to the different combinations of neurons in the hidden layer, which is not limited in this embodiment. After the input vector is processed by the hidden layer, a vector is also obtained, and the prediction device inputs the processed vector into the output layer.

[0056] Step S303: obtaining the congestion level of the target road section within the target time period by identifying the input vector through the output layer.

[0057] The prediction device identifies the vector input to the output layer through the output layer, and can obtain the congestion level of the target road section during the target time period.

[0058] In one feasible implementation, in the process of training the preset classification model, it is necessary to update and optimize the model parameters of each layer in the preset classification model so that the preset classification model can accurately predict the congestion level of the road section covered by the cell based on the input feature vector (the feature vector constructed based on the cell's load data and user terminal switching data).

[0059] In one feasible implementation, step S302 includes:

[0060] Step S3021, after performing linear processing on the input vector through the weight matrix of the neurons in the hidden layer, activation processing is performed on the linearly processed vector through the activation function in the hidden layer.

[0061] In this embodiment, a feasible processing method for the hidden layer to process the input vector is proposed. When there are multiple hidden layers, the prediction device inputs the target feature vector into the first hidden layer through the input layer. After the first hidden layer processes the vector, the processed vector is input into the second hidden layer. After the second hidden layer processes the vector, the processed vector is input into the third hidden layer, and so on. The processing method of each hidden layer for the input vector is the same, except that the number of neurons and weight values ​​of each hidden layer are different. Taking a hidden layer as an example, the prediction device can perform linear processing on the input vector through the weight matrix of the neurons in the hidden layer, and then activate the linearly processed vector through the activation function in the hidden layer, and then input the activated vector into the next layer.

[0062] The activation function can be set as needed. In the present embodiment, there is no limitation. For example, sigmoid, tanh, Relu, etc. can be used.

[0063] Step S3022, inputting the vector obtained through the activation process of the last hidden layer into the output layer.

[0064] The prediction device inputs the vector obtained after the activation processing of the last hidden layer into the output layer.

[0065] It should be noted that when there is only one hidden layer in the preset classification model, the target feature vector is input into the hidden layer through the input layer, and then the input vector is processed by the hidden layer, and the processed vector is input into the output layer.

[0066] In one feasible implementation, step S303 includes:

[0067] Step S3031, linearly processing the input vector through the weight matrix of the neurons in the output layer to obtain output values ​​corresponding to each preset congestion level.

[0068] In this embodiment, a feasible identification method is proposed to obtain the congestion level by identifying the input vector through the output layer. The output layer includes neurons and normalization functions, and the number of neurons is configured to be the same as the number of preset congestion levels, so that the output value corresponds to each preset congestion level. The prediction device can perform linear processing on the input vector through the weight matrix of the neurons in the output layer to obtain the output values ​​corresponding to each preset congestion level.

[0069] Step S3032, normalizing the output values ​​corresponding to the respective preset congestion levels by means of the normalization function in the output layer, and obtaining the probability that the target road section belongs to each preset congestion level within the target time period.

[0070] The prediction device normalizes the output values ​​corresponding to each preset congestion level through the normalization function in the output layer, and obtains the probability that the target road section belongs to each preset congestion level during the target time period. The normalization function can be a function set as needed. In this embodiment, there is no limitation on which normalization function is used. For example, a softmax function can be used.

[0071]

[0072] y1,y2,…y C The values ​​corresponding to the output of the C neuron nodes in the output layer represent the probability that the target road section belongs to the C preset congestion levels during the target period. Since the vector Y of the output layer does not necessarily satisfy the mathematical characteristics of the probability distribution, that is, the sum of the dimensions of the Y vector is not necessarily equal to 1, in order to convert Y into a probability distribution vector, a normalization operation is performed.

[0073] Step S3033: determining the congestion level of the target road section within the target time period from various preset congestion levels according to the probabilities.

[0074] The prediction device determines the congestion level of the target road section within the target time period from the preset congestion levels according to the probabilities of the preset congestion levels. In this embodiment, the method of determining the congestion level according to the probability is not limited. For example, in a feasible embodiment, the preset congestion level with the highest probability can be used as the congestion level of the target road section within the target time period.

[0075] Based on the above-mentioned first, second and / or third embodiments, a fourth embodiment of the training method for the classification model of the present application is proposed. In this embodiment, there is no limitation on the execution subject of the classification model training method, for example, it can be a personal computer, a server, a smart phone and other devices. For the convenience of description, the following description will be based on the training device as the execution subject for each embodiment. It should be noted that the training device and the prediction device can be the same device or different devices, and can be set as needed, which is not limited in this embodiment; when the prediction device and the training device are different devices, after the target classification model is trained in the training device, the target classification model can be deployed in the prediction device. Refer to Figure 3 In this embodiment, the training method of the classification model includes:

[0076] Step A10, obtaining a sample data set, wherein the sample data set includes multiple sample data, each sample data includes cell load data and user terminal switching data of a cell in a time period, and an actual congestion level of a road section covered by the cell signal.

[0077] In this embodiment, a feasible training method for a target classification model is proposed. In this embodiment, there is no restriction on the way in which the training device obtains the sample data set. For example, in a feasible implementation, the user can upload the sample data set in the training device.

[0078] The sample data set includes multiple sample data, each of which includes the cell load data and user terminal switching data of a cell in a time period, as well as the actual congestion level of the road section covered by the cell signal. The duration of the time period can be set as needed, representing the time granularity of the sample data. For example, with the granularity of minutes, for a cell, the cell load data and user terminal switching data and the actual congestion level in a time period of one minute can be obtained to form a sample data. Multiple sample data can be obtained for the cell, and multiple sample data can also be obtained for different cells. The sample data of each cell are combined to obtain the sample data set.

[0079] The actual congestion level of the road section covered by the cell signal may be manually marked or obtained by other means, which is not limited in this embodiment.

[0080] Step A20: constructing feature vectors according to the cell load data and user terminal switching data in each piece of the sample data, and constructing label data according to the actual congestion level in each piece of the sample data.

[0081] For each sample data in the sample data set, the training device can construct a feature vector (also referred to as a "cell congestion feature vector") based on the cell load data and user terminal switching data in the sample data, and construct label data (also referred to as a "congestion level label") based on the actual congestion level in the sample data. For example, in one feasible implementation, constructing label data based on the actual congestion level can be: presetting labels corresponding to each preset congestion level, for example, using C integers from 0 to (C-1) to represent each preset congestion level, respectively, the number of C is determined by the number of preset congestion levels, and using the label corresponding to the actual congestion level as the label data corresponding to the sample data; without loss of generality, when two preset congestion levels are set, namely, traffic is unimpeded and traffic is congested, 0 can be used as the label for traffic is unimpeded, and 1 can be used as the label for traffic is congested. For example, in a feasible implementation, the label data can be constructed by using a one-hot encoding method; the number of preset congestion levels is C, and the label data is represented by a vector of length C, in which each element in the vector corresponds to each preset congestion level, and the element corresponding to the actual congestion level is set to 1, and the elements corresponding to other preset congestion levels are set to 0, indicating that the probability that the congestion level of the road section covered by the cell signal is the actual congestion level is 1, and the probability of being other preset congestion levels is 0.

[0082] The specific implementation method of the training device constructing the feature vector based on the cell load data and user terminal switching data in the sample data can refer to the specific implementation method of the prediction device constructing the target feature vector based on the cell load data and user terminal switching data of the target cell in the above embodiment, which will not be repeated here.

[0083] Step A30, using the feature vector corresponding to each piece of sample data as model input data, and using the label data corresponding to each piece of sample data as model training labels, to train the classification model to be trained, and obtain a target classification model for predicting road congestion levels.

[0084] A classification model to be trained can be pre-set in the training device, and the model structure of the classification model can be set as needed, and there is no limitation here. The model parameters in the classification model can be initialized based on experience or randomly, and the training process is the process of optimizing the model parameters. The training device can use the feature vector corresponding to each sample as the model input data, and the label data corresponding to each sample data as the model training label, to train the classification model to be trained, and obtain a target classification model for predicting the road congestion level.

[0085] In one possible implementation, if Figure 4As shown, the model structure of the classification model can be set to include an input layer, at least one hidden layer and an output layer. The input layer includes M nodes, M is the size of the feature vector, and the input layer is used to input the feature vector into the hidden layer. In a feasible implementation, the number of hidden layers can be 1, and the number of neurons in the hidden layer is K. The number of neurons in the output layer is C, and the probability values ​​of C categories are output. The weight matrix of the neurons in the hidden layer can be expressed as W1, with a dimension of (K, M), and the weight matrix of the neurons in the output layer can be expressed as W2, with a dimension of (C, K). The input feature vector is expressed as X=[x1, x2,…, x M ], the mathematical formula of the classification model is expressed as follows:

[0086] Y=W2*W1*X+b

[0087] Where b is the bias vector. The physical meaning of the W1 and W2 matrices is that they represent the weight influence of different feature vectors in the classification of road congestion levels. W1, W2 and b are the model parameters that need to be optimized during the training process.

[0088] In one feasible implementation, cross entropy can be used as the loss function of the model:

[0089]

[0090] N represents the total number of samples input during training; C represents the number of preset congestion levels, that is, the number of model classifications; y ij is a one-dimensional one-hot encoding vector, indicating whether the i-th sample belongs to the j-th category; p ij Represents the predicted probability of the classification model that the i-th sample belongs to the j-th category. On the training sample set, by using the gradient descent algorithm to optimize the model parameters for the loss function L, the model parameters with the best congestion level classification effect for the road section can be finally obtained. The optimizer (Optimizer) determines the convergence mode of the gradient descent of the model loss function. In one feasible implementation, the optimizer can use Adadelta, which is an adaptive learning rate method. During the training process, the number of samples in each batch of training (BatchSize) can be set as needed, for example, it can be set to 10. The number of training rounds can be set as needed, for example, it can be set to 100 times.

[0091] In a feasible implementation, in order to evaluate the effect of model application, the sample data set can be divided into a training set and a test set, where the division ratio of the training set is K and the corresponding division ratio of the test set is (1-K). On the training set, multiple iterative training processes are completed, and then the effect of the model is evaluated on the test set. Finally, the best model with the smallest loss function is obtained through training, and the trained model is saved for use by subsequent modules.

[0092] In one feasible implementation, the step A10 includes:

[0093] Step A101, collecting cell load data and user terminal switching data of at least one cell in each hour period from a network management module, a network element module or a data service module, wherein each hour period is obtained by dividing a time period of a preset duration according to a preset time granularity.

[0094] In this embodiment, a feasible way to obtain a sample data set is proposed. The network element module is a functional module deployed in the network element device, which stores the communication data in the network element device; the network management module refers to the network management module, which manages multiple network element devices and stores the communication data in multiple network element devices; the data service module is deployed in an external server, provides data calculation and analysis services for the cell, and stores the communication data in the cell. Therefore, in this embodiment, the training device can collect the cell load data and user terminal switching data of at least one cell in each hour period from the network management module, the network element module or the data service model. Each hour period is obtained by dividing the time period of the preset time length according to the preset time granularity. The preset time granularity is the time granularity set in advance according to the needs, for example, it can be set to the minute granularity. The preset duration can be set as needed, for example, it can be set to one month. Then, the training device can obtain the cell load data and user terminal switching data of at least one cell every minute within one month.

[0095] In one possible implementation mode, the training device can collect cell load data and user terminal switching data of the cell within a preset time period. The collected data can carry timestamp information, including but not limited to timestamp information with year / month / day / hour / minute granularity. Then, according to the preset time granularity, the collected data is divided into cell load data and user terminal switching data within each hour period according to the timestamp information.

[0096] Step A102, collecting the actual congestion level of the road section covered by the cell signal in each hour period.

[0097] The training device can also collect the actual congestion level of the road section covered by the cell signal in each hour period. The collection method is not limited in this embodiment, for example, it can be obtained through manual marking.

[0098] Step A103, obtaining one piece of sample data based on the cell load data, user terminal switching data and actual congestion level of the same cell collected in the same hour period, and obtaining the sample data set based on multiple pieces of sample data.

[0099] The training device combines the cell load data, user terminal switching data and actual congestion level of the same cell collected within the same hour period into one sample data. Then, based on the acquired data, multiple sample data can be combined, and multiple sample data can constitute a sample data set.

[0100] In one feasible implementation, the step A102 includes:

[0101] Step A1021, collecting surveillance videos of surveillance cameras installed on the road section covered by the cell signal in each of the hourly periods;

[0102] In this embodiment, a feasible method for a training device to collect the actual congestion level of a road section covered by a cell signal is proposed. When collecting sample data, a road section covered by a surveillance camera can be selected, and then the cell load data and user terminal switching data of the cell with signal coverage on the road section can be collected, and the surveillance video of the surveillance camera set on the road section in each hour period can be collected.

[0103] Step A1022, performing image analysis on the surveillance video of each hour period respectively, and obtaining the actual congestion level of the road section covered by the cell signal in each hour period.

[0104] The training device can perform image analysis on the surveillance videos of each hour period, and obtain the actual congestion level of the road section in each hour period based on the image analysis results. The image analysis method is not limited in this embodiment, for example, an image classification algorithm can be used to classify the image to obtain the actual congestion level.

[0105] In addition, the embodiment of the present application further provides a traffic condition prediction device, the traffic condition prediction device comprising:

[0106] A first acquisition module is used to acquire cell load data and user terminal switching data of a target cell in a target period, wherein the target cell is a cell whose signal covers a target road section;

[0107] A first construction module, configured to construct a target feature vector according to the cell load data and the user terminal switching data;

[0108] An identification module is used to input the target feature vector into a preset classification model for identification, so as to obtain the congestion level of the target road section within the target time period, wherein the preset classification model is trained by using a feature vector constructed with cell load data and user terminal switching data as model input data and label data constructed with actual congestion levels of the road sections covered by the cell as model training labels.

[0109] In one feasible implementation, the traffic condition prediction device further includes:

[0110] The adjustment module is used to adjust the signal coverage of the neighboring cells of the target cell in response to the congestion level being greater than a preset level, wherein the adjusted signal coverage of the neighboring cells includes part or all of the target road section.

[0111] In a feasible implementation manner, the cell load data includes at least one of the following data: the average number of connected users of the cell radio resource control, the number of activated users, the uplink medium access control layer traffic, the downlink medium access control layer traffic, and the utilization rate of the physical resource blocks.

[0112] In a feasible implementation manner, the user terminal switching data includes at least one of the following data: the number of times the user terminal successfully switches into the target cell, the number of times the user terminal successfully switches out of the target cell, the number of times the user terminal requests to switch into the target cell, and the number of times the user terminal requests to switch out of the target cell.

[0113] In a feasible implementation manner, the first building block is further used for:

[0114] Preprocessing each item of the cell load data and the user terminal switching data to obtain a characteristic quantity corresponding to each item of the data;

[0115] The feature quantities corresponding to each data item are combined to obtain the target feature vector.

[0116] In a feasible implementation manner, the preset classification model includes an input layer, at least one hidden layer and an output layer, and the recognition module is further used for:

[0117] Inputting the target feature vector into the hidden layer through the input layer;

[0118] After the input vector is processed by the hidden layer, the processed vector is input into the output layer;

[0119] The output layer identifies the input vector and obtains the congestion level of the target road section within the target time period.

[0120] In one feasible implementation, the identification module is further used for:

[0121] After performing linear processing on the input vector through the weight matrix of the neurons in the hidden layer, performing activation processing on the linearly processed vector through the activation function in the hidden layer;

[0122] The vector obtained by the activation process of the last hidden layer is input into the output layer.

[0123] In one feasible implementation, the identification module is further used for:

[0124] The input vector is linearly processed by the weight matrix of the neurons in the output layer to obtain output values ​​corresponding to each preset congestion level;

[0125] Normalizing the output values ​​corresponding to the preset congestion levels by the normalization function in the output layer to obtain the probability that the target road section belongs to each preset congestion level in the target time period;

[0126] The congestion level of the target road section within the target time period is determined from various preset congestion levels according to the probabilities.

[0127] In addition, the present application also provides a classification model training device, the classification model training device comprising:

[0128] A second acquisition module is used to acquire a sample data set, wherein the sample data set includes a plurality of sample data, each of which includes cell load data and user terminal switching data of a cell in a period of time, and an actual congestion level of a road section covered by the cell signal;

[0129] A second construction module is used to construct a feature vector according to the cell load data and the user terminal switching data in each piece of the sample data, and to construct label data according to the actual congestion level in each piece of the sample data;

[0130] The training module is used to use the feature vector corresponding to each sample data as model input data, and the label data corresponding to each sample data as model training labels, to train the classification model to be trained, and obtain a target classification model for predicting the road congestion level.

[0131] In a feasible implementation manner, the second acquisition module is further used for:

[0132] Collecting cell load data and user terminal switching data of at least one cell in each hour period from the network management module, the network element module or the data service module, wherein each hour period is obtained by dividing a period of preset time length according to a preset time granularity;

[0133] Collecting the actual congestion level of the road section covered by the cell signal in each of the hourly periods;

[0134] A piece of sample data is obtained based on the cell load data, user terminal switching data and actual congestion level of the same cell collected in the same hour period, and the sample data set is obtained based on multiple pieces of sample data.

[0135] In a feasible implementation manner, the second acquisition module is further used for:

[0136] Collecting surveillance videos of surveillance cameras installed on the road sections covered by the cell signals in each of the hourly periods;

[0137] Image analysis is performed on the surveillance video of each of the hourly segments to obtain the actual congestion level of the road section covered by the cell signal in each of the hourly segments.

[0138] The embodiment of the present application also provides a traffic condition prediction device, the traffic condition prediction device comprising: a memory, a processor, and a traffic condition prediction program stored in the memory and executable on the processor, the traffic condition prediction program when executed by the processor implements the steps of the traffic condition prediction method described in the above embodiments. Figure 5 As shown, the traffic condition prediction device may also include a network interface, a user interface and a communication bus. Among them, the communication bus is used to realize the connection and communication between these components. The user interface may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface may also include a standard wired interface and a wireless interface. The network interface may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a disk memory. The memory may optionally be a storage device independent of the aforementioned processor. As a computer storage medium, the memory may include an operating system, a network communication module, a user interface module, and a traffic condition prediction program. The operating system is a program that manages and controls the hardware and software resources of the device, and supports the operation of the traffic condition prediction program and other software or programs. Those skilled in the art will understand that Figure 5 The device structure shown in the figure does not constitute a limitation on the traffic condition prediction device, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0139] The present application also provides a classification model training device, the classification model training device comprising: a memory, a processor, and a classification model training program stored in the memory and executable on the processor, the classification model training program when executed by the processor implements the steps of the classification model training method described in the above embodiments. In a specific implementation, the classification model training device may be a line card device in a network element device. In a specific implementation, Figure 6 As shown, the training device of the classification model may also include a network interface, a user interface and a communication bus. Among them, the communication bus is used to realize the connection and communication between these components. The user interface may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface may also include a standard wired interface and a wireless interface. The network interface may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. The memory may also be a storage device independent of the aforementioned processor. As a computer storage medium, the memory may include an operating system, a network communication module, a user interface module, and a training program for the classification model. The operating system is a program that manages and controls the hardware and software resources of the device, and supports the operation of the training program of the classification model and other software or programs. Those skilled in the art will understand that Figure 6 The device structure shown in the figure does not constitute a limitation on the training device of the classification model, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0140] Reference Figure 7 The present application also provides a computer-readable storage medium on which a traffic condition prediction program is stored. When the traffic condition prediction program is executed by a processor, the steps of the traffic condition prediction method described in the above embodiments are implemented. Figure 7 In a specific implementation, a memory as a computer-readable storage medium can be connected to communicate with the processor via a communication bus.

[0141] Reference Figure 8 The present application also provides a computer-readable storage medium on which a classification model training program is stored. When the classification model training program is executed by a processor, the steps of the classification model training method described in the above embodiments are implemented. Figure 8 In a specific implementation, a memory as a computer-readable storage medium can be connected to communicate with the processor via a communication bus.

[0142] The various embodiments of the traffic condition prediction device, prediction equipment, classification model training device, training equipment and computer-readable storage medium of the present application can all refer to the various embodiments of the traffic condition prediction method and classification model training method of the present application, and will not be repeated here.

[0143] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0144] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0145] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0146] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A traffic condition prediction method, characterized in that: The method comprises: Acquire cell load data and user terminal switching data of a target cell in a target period, wherein the target cell is a cell whose signal covers a target road section; Constructing a target feature vector according to the cell load data and the user terminal switching data; The target feature vector is input into a preset classification model for identification to obtain the congestion level of the target road section within the target time period, wherein the preset classification model is trained using a feature vector constructed using cell load data and user terminal switching data as model input data and label data constructed using actual congestion levels of the road section covered by the cell as model training labels.

2. The traffic condition prediction method according to claim 1, characterized in that: After the step of inputting the target feature vector into a preset classification model for identification to obtain the congestion level of the target road section within the target time period, the method further includes: In response to the congestion level being greater than a preset level, the signal coverage of a cell adjacent to the target cell is adjusted, wherein the adjusted signal coverage of the adjacent cell includes a portion or all of the target road section.

3. The traffic condition prediction method according to claim 1, characterized in that: The cell load data includes at least one of the following data: the average number of connected users, the number of activated users, the uplink medium access control layer flow, the downlink medium access control layer flow, and the utilization rate of physical resource blocks in the cell radio resource control.

4. The traffic condition prediction method according to claim 1, characterized in that: The user terminal switching data includes at least one of the following data: the number of times the user terminal successfully switches into the target cell, the number of times the user terminal successfully switches out of the target cell, the number of times the user terminal requests to switch into the target cell, and the number of times the user terminal requests to switch out of the target cell.

5. The traffic condition prediction method according to claim 3, characterized in that: The step of constructing a target feature vector according to the cell load data and the user terminal switching data comprises: Preprocessing each item of the cell load data and the user terminal switching data to obtain a characteristic quantity corresponding to each item of the data; The feature quantities corresponding to each data item are combined to obtain the target feature vector.

6. The traffic condition prediction method according to any one of claims 1 to 5, characterized in that: The preset classification model includes an input layer, at least one hidden layer and an output layer. The step of inputting the target feature vector into the preset classification model for identification to obtain the congestion level of the target road section in the target time period includes: Inputting the target feature vector into the hidden layer through the input layer; After the input vector is processed by the hidden layer, the processed vector is input into the output layer; The output layer identifies the input vector and obtains the congestion level of the target road section within the target time period.

7. The traffic condition prediction method according to claim 6, characterized in that: After the input vector is processed by the hidden layer, the step of inputting the processed vector into the output layer comprises: After performing linear processing on the input vector through the weight matrix of the neurons in the hidden layer, performing activation processing on the linearly processed vector through the activation function in the hidden layer; The vector obtained by the activation process of the last hidden layer is input into the output layer.

8. The traffic condition prediction method according to claim 6, characterized in that: The step of obtaining the congestion level of the target road section within the target time period by identifying the input vector through the output layer includes: The input vector is linearly processed by the weight matrix of the neurons in the output layer to obtain output values ​​corresponding to each preset congestion level; Normalizing the output values ​​corresponding to the preset congestion levels by the normalization function in the output layer to obtain the probability that the target road section belongs to each preset congestion level in the target time period; The congestion level of the target road section within the target time period is determined from various preset congestion levels according to the probabilities.

9. A classification model training method, characterized in that: The training method of the classification model includes: Acquire a sample data set, wherein the sample data set includes a plurality of sample data, each sample data including cell load data and user terminal switching data of a cell in a time period, and an actual congestion level of a road section covered by the cell signal; Constructing feature vectors according to the cell load data and user terminal switching data in each of the sample data, and constructing label data according to the actual congestion level in each of the sample data; The feature vector corresponding to each piece of sample data is used as model input data, and the label data corresponding to each piece of sample data is used as model training label, and the classification model to be trained is trained to obtain a target classification model for predicting the road congestion level.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a traffic condition prediction program, and when the traffic condition prediction program is executed by a processor, the traffic condition prediction method according to any one of claims 1 to 7 is implemented.

Images