An optimization method for traffic information estimation

Abstract

The present invention provides a method for optimizing traffic information estimation. The method includes constructing a neural network model for optimization, the neural network model comprising an input layer, an output layer, and six intermediate layers. Required parameters are input through the input layer, with each parameter input through a neuron. The required parameters include the number of cellular network signals in the current cell, the number of cellular network signals in the previous cell, and the number of cellular network signals in the next cell, as well as at least one piece of estimated traffic information, such as traffic flow or vehicle speed. The intermediate layers optimize the input parameters, and the output layer outputs the optimized estimated traffic information. The present invention has the advantage of effectively reducing estimation errors caused by randomness, thereby ensuring more accurate traffic information estimation.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an optimization method for estimating traffic information. Background Art

[0002] Cellular traffic data can be used to estimate traffic information. The cellular network signals commonly used include HO (Handover / Handoff) signals, NLU (Normal Location Update) signals, CA (Call Arrival) signals, and PLU (Periodic Location Update) signals.

[0003] Currently, there are two types of traffic information estimation methods based on cellular mobile vehicle data that utilize cellular network signal data: the first type estimates traffic information through continuous cellular network signal sequences. This method uses the spatiotemporal characteristics of cellular mobile vehicle data generated by mobile phones during movement, and obtains the time and distance differences between two consecutive cellular network events occurring on the same mobile phone to estimate traffic information. However, this type of method requires continuous tracking of the same mobile phone to obtain all its interactions with the base station during the observation period, which will cause certain privacy issues.

[0004] The second type estimates traffic information by counting cellular network signals. This method uses statistical methods to calculate the number of various cellular network signals occurring in each cell within each time period from cellular network vehicle data, and then uses machine learning or probabilistic statistical methods to estimate traffic information. This method can avoid privacy leakage to a certain extent because the statistical data is insensitive to IMSI. However, how to obtain effective spatiotemporal features from cellular vehicle data remains a problem that needs to be solved. Lin et al. (see Lin BY, Chen CH, Lo C CA, "Traffic Information Estimation Model Using Periodic Location Update Events from Cellular Network [C] / / International Conference on Intelligent Computing and Information Science, Berlin, Heidelberg. 2011, 135:72–77) proposed using probability density functions to obtain user call behavior from cellular vehicle data and extract spatiotemporal features for traffic information estimation. However, due to the randomness of user call behavior, this randomness can cause a deviation between the actual number of cellular network signals and the estimated number, thereby interfering with traffic information estimation. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an optimization method for estimating traffic information, so as to solve the problem of random interference when estimating traffic information using the number of cellular network signals.

[0006] The present invention provides a method for optimizing traffic information estimation, the method comprising:

[0007] Constructing a neural network model for optimization, the neural network model comprising an input layer, an output layer, and six intermediate layers; the output layer comprises one neuron, the input layer comprises at least four neurons, and each of the intermediate layers comprises at least four neurons;

[0008] The required parameters are input through the input layer, and each parameter is input through a neuron. The required parameters include the number of cellular network signals in the current cell, the number of cellular network signals in the previous cell, the number of cellular network signals in the next cell, and at least one estimated traffic information, the traffic information including traffic flow or vehicle speed; the input parameters are optimized through each of the intermediate layers; and the optimized estimated traffic information is output through the output layer.

[0009] Furthermore, the neurons in the input layer, output layer and intermediate layer all use LeakyReLU as the activation function.

[0010] Furthermore, when estimating traffic flow using the HO signal of the cellular network, the parameters input through the input layer include the number of HO signals of the current cell, the number of HO signals of the previous cell, the number of HO signals of the next cell, and the estimated traffic flow.

[0011] Furthermore, the input layer includes four neurons; the number of neurons in the six intermediate layers is 8, 12, 16, 12, 8, and 4 from the input end to the output end.

[0012] Furthermore, when the vehicle speed is estimated using the CA signal of the cellular network, the parameters input through the input layer include the number of CA signals of the current cell, the number of CA signals of the previous cell, the number of CA signals of the next cell, the estimated traffic flow, and the estimated vehicle speed.

[0013] Furthermore, the input layer includes five neurons; the number of neurons in the six intermediate layers is 10, 15, 20, 15, 10, and 5 from the input end to the output end.

[0014] Furthermore, when the vehicle speed is estimated using the PLU signal of the cellular network, the parameters input through the input layer include the number of PLU signals of the current cell, the number of PLU signals of the previous cell, the number of PLU signals of the next cell, and the estimated vehicle speed.

[0015] Furthermore, the input layer includes four neurons; the number of neurons in the six intermediate layers is 8, 12, 16, 12, 8, and 4 from the input end to the output end.

[0016] One or more technical solutions provided in the embodiments of the present invention have at least the following technical effects or advantages: a neural network model is used to optimize the estimated traffic information, and the neural network model includes an input layer, an output layer, and six intermediate layers. The number of cellular network signals in the current cell, the number of cellular network signals in the previous cell, the number of cellular network signals in the next cell, and at least one estimated traffic information are input into the intermediate layers of the neural network model through the input layer for optimization, and the optimized estimated traffic information is output through the output layer; by adopting the above-mentioned technical solutions of the present invention, the estimation error caused by randomness can be effectively reduced, thereby ensuring that the traffic information estimation is more accurate.

[0017] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] Figure 1 A flowchart of the optimization method for estimating traffic information according to the present invention;

[0020] Figure 2 Schematic diagram of the number of HO signals of the cellular network of the cell before and after observation of the present invention;

[0021] Figure 3 Schematic diagram of a neural network model for estimating vehicle speed using HO signals from a cellular network in the present invention;

[0022] Figure 4 Schematic diagram of the number of CA signals of the cellular network of the Cell before and after observation of the present invention;

[0023] Figure 5 This is a schematic diagram of a neural network model for estimating vehicle speed using CA signals from a cellular network in the present invention;

[0024] Figure 6 Schematic diagram of the number of PLU signals of the cellular network of the Cell before and after observation of the present invention;

[0025] Figure 7 This is a schematic diagram of a neural network model for estimating vehicle speed using a PLU signal from a cellular network;

[0026] Figure 8 Schematic diagram of the structure of the cumulative distribution function generator based on call holding time of the present invention;

[0027] Figure 9 Schematic diagram of the structure of the cumulative distribution function generator based on the call interval time of the present invention;

[0028] Figure 10 Schematic diagram of the space where the HO signal occurs in the present invention;

[0029] Figure 11 Schematic diagram of the timing of HO signal generation in the present invention;

[0030] Figure 12 Schematic diagram of the time when CA signal occurs in the present invention;

[0031] Figure 13Schematic diagram of the space where the NLU signal occurs in the present invention;

[0032] Figure 14 This is a spatial schematic diagram of the PLU signal generation in the first case of the present invention;

[0033] Figure 15 This is a schematic diagram of the timing of the PLU signal generation in the first case of the present invention;

[0034] Figure 16 Schematic diagram of the space in which the PLU signal occurs in the second case of the present invention;

[0035] Figure 17 Schematic diagram of the timing of the PLU signal generation in the second case of the present invention;

[0036] Figure 18 A fitting structure diagram of the estimated cumulative distribution function and the actual cumulative distribution function in the present invention;

[0037] Figure 19 This is a graph showing the changing trends of traffic flow, vehicle speed, and cellular network signal according to an example of the present invention. DETAILED DESCRIPTION

[0038] A preferred embodiment of the optimization method for traffic information estimation of the present invention is as follows: Figure 1 As shown, the optimization method includes:

[0039] Constructing a neural network model for optimization, the neural network model comprising an input layer 101, an output layer 103, and six intermediate layers 102; the output layer 103 comprises one neuron, the input layer 101 comprises at least four neurons, and each of the intermediate layers 102 comprises at least four neurons;

[0040] The required parameters are input through the input layer 101, and each parameter is input through a neuron. The required parameters include the number of cellular network signals in the current cell, the number of cellular network signals in the previous cell, the number of cellular network signals in the next cell, and at least one estimated traffic information, where the traffic information includes traffic flow or vehicle speed; the input parameters are optimized through each of the intermediate layers 102; and the optimized estimated traffic information is output through the output layer 103.

[0041] Because the accuracy of traditional probabilistic methods for estimating traffic information based on cellular network signals is related to the frequency of cellular network events, these methods perform well when traffic flow is high. Conversely, their accuracy decreases when traffic flow is low. Furthermore, traffic flow varies continuously throughout the day, so the accuracy of traffic information estimation methods based on probability density functions fluctuates with traffic flow throughout the day. This means that the proposed traffic information estimation model has a higher accuracy when traffic flow is high, but a lower accuracy when traffic flow is low.

[0042] Due to the robustness of neural networks, neural networks are particularly effective in reducing noise interference. Therefore, the present invention uses a neural network model to optimize the estimated traffic information. The neural network model includes an input layer 101, an output layer 103, and six intermediate layers 102. The number of cellular network signals in the current cell, the number of cellular network signals in the previous cell, the number of cellular network signals in the next cell, and at least one estimated traffic information are input into each intermediate layer 102 of the neural network model through the input layer 101 for optimization. The optimized estimated traffic information is output through the output layer 103. By adopting the above-mentioned technical solution of the present invention, the estimation error caused by randomness can be effectively reduced, thereby ensuring that the traffic information estimation is more accurate.

[0043] In a preferred embodiment of the present invention, the neurons in the input layer 101, output layer 103, and intermediate layer 102 all use LeakyReLU as the activation function. LeakyReLU is a deep learning activation function commonly used in computer vision and is a variant of the ReLU function. The advantage of the LeakyReLU function is that it can overcome the potential problem of the ReLU function, namely the vanishing gradient. Because the LeakyReLU function allows negative values ​​to pass through, it is conducive to model convergence and can effectively suppress overfitting. At the same time, the LeakyReLU function can also maintain smoothness within the gradient range of the activation function, which makes backpropagation and gradient updates easier.

[0044] As a first specific embodiment of the present invention, since cellular network events occurring in a certain cell (i.e., a cell, specifically referring to the coverage area of ​​a base station's signal transmission) are random, and the speed of vehicles on a highway varies very little within a certain range, the present invention adopts a method of simultaneously observing the number of cellular network events occurring in the two cells before and after to increase the number of samples and more accurately reflect the traffic status at the cell; Figure 2 As shown, in the estimation of Cell i When the traffic flow at iRepresents the i-th cell, i is a positive integer; at the same time, Cell i-1 and Cell i+1 The number of HO signals of the cellular network occurring at the time is also used as the input of the neural network model, thereby reducing the impact of randomness and improving the accuracy of the estimated traffic flow;

[0045] Therefore, when estimating traffic flow using HO signals of a cellular network, the parameters input through the input layer 101 include the number of HO signals in the current cell, the number of HO signals in the previous cell, the number of HO signals in the next cell, and the estimated traffic flow. More specifically, the input layer 101 includes four neurons; the number of neurons in the six intermediate layers 102 is 8, 12, 16, 12, 8, and 4, respectively, from the input end to the output end.

[0046] like Figure 3 As shown, the present invention uses a neural network model with a structure of 4-8-12-16-12-8-4-1 for optimization. The input of the neural network model is Among them, h i-1 、h i and h i+1 Represents Cell i-1 、Cell i and Cell i+1 The number of HO signals of the cellular network occurring at represents the traffic flow estimated based on the HO signal of the cellular network.

[0047] As a second specific embodiment of the present invention, for optimizing vehicle speed estimation based on CA signals of cellular networks, as shown in FIG. Figure 4 As shown, in the estimation of Cell i At the speed of i Represents the i-th cell, i is a positive integer; at the same time, Cell i-1 and Cell i+1 The number of CA signals of the cellular network occurring at the location is also used as the input of the neural network model, thereby reducing the impact of randomness and improving the accuracy of the estimated vehicle speed;

[0048] Therefore, when estimating vehicle speed using CA signals from a cellular network, the parameters input through the input layer 101 include the number of CA signals in the current cell, the number of CA signals in the previous cell, the number of CA signals in the next cell, the estimated traffic flow, and the estimated vehicle speed. More specifically, the input layer 101 includes five neurons; the number of neurons in the six intermediate layers 102 is 10, 15, 20, 15, 10, and 5, respectively, from the input end to the output end.

[0049] like Figure 5 As shown, the present invention uses a neural network model with a structure of 5-10-15-20-15-10-5-1 for optimization. The input of the neural network model is Among them, c i-1 、c i and c i+1 Represents Cell i-1 、Cell i and Cell i+1 The number of CA signals of the cellular network occurring at It represents the traffic flow estimated based on the NLU signal or HO signal of the cellular network. represents the vehicle speed estimated based on the CA signal of the cellular network; is the output of the neural network model, i.e. the estimated vehicle speed after optimization.

[0050] As a second specific embodiment of the present invention, for optimizing vehicle speed estimation based on PLU signals of cellular networks, as shown in FIG. Figure 6 As shown, in the estimation of Cell i At the speed of i Represents the i-th cell, i is a positive integer; at the same time, Cell i-1 and Cell i+1 The number of PLU signals of the cellular network occurring at each location is also used as the input of the neural network model, thereby reducing the impact of randomness and improving the accuracy of the estimated vehicle speed.

[0051] Therefore, when estimating vehicle speed using the PLU signal of the cellular network, the parameters input through the input layer 101 include the number of PLU signals of the current cell, the number of PLU signals of the previous cell, the number of PLU signals of the next cell, and the estimated vehicle speed; more specifically, the input layer 101 includes four neurons; the number of neurons in the six intermediate layers 102 is 8, 12, 16, 12, 8, and 4, respectively, from the input end to the output end;

[0052] like Figure 7 As shown, the present invention uses a neural network model with a structure of 4-8-12-16-12-8-4-1 for optimization. The input of the neural network model is Among them, p i-1 、p i and p i+1 Represents Cell i-1 、Cell i and Cell i+1 The number of PLU signals of the cellular network occurring at Indicates the vehicle speed estimated based on the PLU signal of the cellular network; Represents the output of the neural network, that is, the optimized estimated vehicle speed.

[0053] In the present invention, traffic information estimation includes:

[0054] Constructing a cumulative distribution function generator, the cumulative distribution function generator comprising an input layer 201, an output layer 203, and an intermediate layer 202; the neurons in the intermediate layer 202 respectively use an exponential distribution function, a normal distribution function, or a lognormal distribution function as activation functions, and remove all biases, that is, a mixed distribution function is formed by the exponential distribution function, the normal distribution function, or the lognormal distribution function, and used as the activation function of the intermediate layer 202; the input layer 201 is used to input the actual call time collected from the cellular mobile vehicle data, the intermediate layer 202 is used to train and learn the call time input by the input layer, and the output layer 203 is used to output the estimated cumulative distribution function of the call time;

[0055] Outputting the cumulative distribution function of the estimated call time using the constructed cumulative distribution function generator, and derivatizing the cumulative distribution function of the estimated call time to obtain a probability density function of the estimated call time;

[0056] Traffic information is estimated by using the estimated probability density function of the talk time and the number of cellular network signals, where the traffic information includes traffic flow or vehicle speed.

[0057] In traffic data estimation models based on cellular network signals, call interval and call holding time are two important influencing factors. In real-world situations, user calling behavior is highly random, and the probability density function of call holding time and call interval time may exhibit a multimodal morphology. Existing probability distribution models all use a single distribution function as the activation function, failing to effectively simulate the actual distribution. The present invention utilizes a cumulative distribution function generator to generate a cumulative distribution function of estimated call time. The neurons in the intermediate layer 202 of the cumulative distribution function generator use an exponential distribution function, a normal distribution function, or a lognormal distribution function as the activation function. The cumulative distribution function of the estimated call time is then differentiated to obtain a probability density function of the estimated call time. Traffic information is then estimated using the probability density function of the estimated call time and the number of cellular network signals. Therefore, the technical solution of the present invention accurately models user calling behavior and fits the cumulative distribution function of the multimodal probability density function, thereby obtaining probability density functions of call holding time and call interval time that conform to actual conditions. This allows for more accurate predictions of call holding time and the arrival time of the next call, effectively improving the accuracy of traffic information estimated by the traffic data estimation model.

[0058] In the present invention, the call duration includes the call hold time and / or call interval time. The call hold time is the duration of a single call from answering to hanging up, and the call interval time is the interval between the arrival of two calls. The estimated probability density function of the call duration includes the probability density function of the estimated call hold time and the probability density function of the estimated call interval time. In practice, if the arrival time of a call is known, the arrival time of the next call can be calculated based on the call interval time. Similarly, the call end time can also be calculated based on the call interval time. Therefore, studying the distribution of these two factors is crucial for predicting the arrival time and call duration of the next call.

[0059] In the present invention, Figure 8 As shown, when the input layer 201 inputs the call holding time t actually collected from the cellular mobile vehicle data h When the exponential distribution function f of the intermediate layer 202 11 (t h ) is as follows:

[0060]

[0061] Where λ1 represents the exponential distribution function f 11 (t h ) parameters;

[0062] The normal distribution function f of the intermediate layer 202 12 (t h ) is as follows:

[0063]

[0064] Where μ1 represents the normal distribution function f 12 (t h ), σ1 represents the normal distribution function f 12 (t h )’s standard deviation;

[0065] The lognormal distribution function f of the intermediate layer 202 13 (t h ) is as follows:

[0066]

[0067] Where μ2 represents the lognormal distribution function f 13 (t h ), σ2 represents the lognormal distribution function f 13 (t h )’s standard deviation;

[0068] The output layer 203 outputs the cumulative distribution function of the estimated call holding time The formula is as follows:

[0069]

[0070] Among them, w 11 represents the exponential distribution function f 11 (t h ) weight, w 12 represents the normal distribution function f 12 (t h ) weight, w 13 represents the lognormal distribution function f 13 (t h )’s weight;

[0071] The formula of the loss function used by the intermediate layer 202 during the training and learning process is as follows:

[0072]

[0073] Among them, F h represents the cumulative distribution function of the actual call holding time obtained by statistics, Represents the cumulative distribution function of the estimated call hold time generated by the cumulative distribution function generator.

[0074] In the present invention, Figure 9 As shown, when the input layer 201 inputs the actual call interval time t collected from the cellular mobile vehicle data a When the exponential distribution function f of the intermediate layer 202 21 (t a ) is as follows:

[0075]

[0076] Where λ2 represents the exponential distribution function f 21 (t a ) parameters;

[0077] The normal distribution function f of the intermediate layer 202 22 (t a ) is as follows:

[0078]

[0079] Where μ3 represents the normal distribution function f 22 (t a ), σ3 represents the normal distribution function f 22 (t a)’s standard deviation;

[0080] The lognormal distribution function f of the intermediate layer 202 23 (t a ) is as follows:

[0081]

[0082] Where μ4 represents the lognormal distribution function f 23 (t a ), σ4 represents the lognormal distribution function f 23 (t a )’s standard deviation;

[0083] The output layer 203 outputs the cumulative distribution function of the estimated call interval time The formula is as follows:

[0084]

[0085] Among them, w 21 represents the exponential distribution function f 21 (t a ) weight, w 22 represents the normal distribution function f 22 (t a ) weight, w 23 represents the lognormal distribution function f 23 (t a )’s weight;

[0086] The formula of the loss function used by the intermediate layer 202 during the training and learning process is as follows:

[0087]

[0088] Among them, F a It represents the cumulative distribution function of the actual call interval obtained by statistics. Represents the cumulative distribution function of the estimated inter-call time generated by the cumulative distribution function generator.

[0089] In the present invention, by deriving the cumulative distribution function of the estimated call holding time and the cumulative distribution function of the call interval time, the probability density function of the estimated call holding time and the probability density function of the estimated call interval time can be obtained; therefore, the call holding time and the time when the next call arrives can be predicted. Taking the call arrival time prediction as an example, the call interval time t h The probability of being greater than a certain time x can be expressed as the following formula:

[0090]

[0091] in, Represents the probability density function of the estimated call holding time.

[0092] When the present invention is implemented, it is necessary to collect a large amount of call holding time and call interval time from actual cellular mobile vehicle data, and then input the collected large amount of call holding time and call interval time into the cumulative distribution function generator through the input layer 201. The cumulative distribution function generator can learn the distribution law of the input call holding time and call interval time through a mixed distribution function composed of an exponential distribution function, a normal distribution function or a log-normal distribution function, and output the estimated cumulative distribution function of the call holding time and the estimated cumulative distribution function of the call interval time through the output layer 203.

[0093] As a specific implementation of the present invention, the number of HO events that occur at a certain place within a certain period of time can be used to estimate the size of traffic flow. The HO event is an important event in a cellular network. In a cellular network, the signal range of the base station responsible for communicating with the mobile phone is limited; if the mobile phone moves out of the base station signal coverage, the mobile phone will connect to the new base station to maintain the call; the process of switching from one base station to another during a call is called a HO event; this action will be recorded by the base station, including the time when the switch occurred, the base stations involved and other information, so the number of HO signals that occurred within a certain Cell range within a certain period of time can be obtained from the cellular mobile vehicle data statistics. In order to better demonstrate the HO mechanism in the cellular network, such as Figure 10 As shown, assuming that the vehicle A call is made / received at When you move forward, you will leave the Cell i Enter Cell i+1 Generate HO signal (ie Similarly, when driving to Cell i+2 The call ends at A second HO signal is generated at

[0094] like Figure 11 As shown, when the mobile phone has a call arrival at time t0 (i.e., making / receiving a call), it enters Cell after time x. i Since the call is not hung up, the mobile phone will disconnect from the old base station and establish a connection with the new base station; at this time, the HO event occurs at time t1. Assuming that the mobile phone hangs up the call at time t2, the following information can be obtained:

[0095] The time difference from t0 to t2 is the call holding time t h ;

[0096] Assume that the vehicle is in Celli The average speed of the vehicles in the i , Cell i The coverage length on the target road segment is l i , the vehicle in the Cell i The travel time within can be calculated by l i / u i to obtain;

[0097] If you want the mobile phone to have a HO event at the cell boundary at time t1, you need to ensure that the call holding time t h Greater than entering the Cell i time, i.e. t h >x.

[0098] Therefore, estimating traffic information using the estimated probability density function of the call duration and the number of cellular network signals specifically includes estimating traffic flow using the estimated probability density function of the call hold time and the number of HO signals. The relationship between the number of HO signals and the estimated traffic flow is as shown in the following formula:

[0099]

[0100] Through the above formula (12), the estimated traffic flow is obtained as follows:

[0101]

[0102] Among them, Pr[t h >x] represents the call holding time t h Greater than entering the Cell i The probability of time x, x is determined by the time the call arrives and the time it enters the cell i The moment is calculated, Cell i represents the i-th cell (i.e., the coverage of the signal transmitted by a base station), where i is a positive integer; The probability density function of the estimated call holding time is expressed by the cumulative distribution function of the estimated call holding time Take the derivative and get; Indicates the estimated Cell i Traffic flow at h i Represents Cell i The number of HO signals recorded at

[0103] Another specific embodiment of the present invention primarily uses the number of CA signals occurring within a cell to estimate vehicle speed. Call arrival (CA) is a key event in cellular networks. It occurs when a call is made or received. The base station connected to the phone records information such as the CA event and the base station ID. Therefore, the number of CA signals occurring within a cell within a certain period of time can be calculated using cellular traffic data statistics.

[0104] like Figure 12 As shown, when the vehicle reaches time t0, the first CA event occurs. After time x, it enters Cell at time t1. i Then, at t2, the second CA event occurred, and finally the vehicle left the cell at t3. i By analyzing the time and location of each event, we can obtain the following information:

[0105] The time difference between t0 and t2 is the call interval time t a ;

[0106] Assume that the vehicle is in Cell i The average speed of the vehicles in the i , Cell i The coverage length on the target road segment is l i , the vehicle in the Cell i The travel time within can be calculated by l i / u i to obtain;

[0107] If the second CA event is to occur on the mobile phone at time t2, the interval between the two calls must be t a Greater than entering the Cell i The time is less than the time to exit the cell i time, that is, t1<t2<t3.

[0108] Therefore, estimating traffic information using the estimated probability density function of the call time and the number of cellular network signals includes estimating vehicle speed using the estimated probability density function of the call interval time and the number of CA signals. The relationship between the number of CA signals and the estimated vehicle speed is as shown in the following formula:

[0109]

[0110] The traffic flow q is used in the factor (14) i As a calculation parameter, the actual traffic flow is unknown, so the traffic flow estimated by the NLU signal or HO signal is used as a parameter. The relationship between the number of CA signals and the estimated vehicle speed is further expressed as follows:

[0111]

[0112] in, Indicates the call interval time t a Greater than entering the Cell i The time x is less than the time to exit the cell i Time The probability that x is reached by the time the call arrives and the time when the call enters the cell i The moment is calculated, Cell i represents the i-th cell (i.e., the coverage of the signal transmitted by a base station), i is a positive integer; l i Represents Cell i Length of coverage on the highway, Indicates the estimated Cell i Speed ​​of the vehicle at The probability density function of the estimated call interval time is expressed by the cumulative distribution function of the estimated call interval time. Take the derivative and get; c i Represents Cell i The number of CA signals recorded at i Represents Cell i Traffic flow within; q N,i Indicates the traffic volume estimated using NLU signal or HO signal;

[0113] Use Newton's downhill method to find an approximate solution that satisfies the error ε less than 0.00001. The solution process is as follows:

[0114]

[0115]

[0116]

[0117]

[0118] in, represents the objective equation of Newton's downhill method, represents the derivative of the objective equation of Newton's downhill method; ω k represents the iteration parameter of Newton's downhill method, Represents the cell of the kth iteration in Newton's downhill method i The estimated vehicle speed value at , k is a positive integer; The cell representing the k+1th iteration of the Newton downhill method i Estimated vehicle speed at .

[0119] In another embodiment of the present invention, when q N,i When the vehicle flow (ie, traffic flow) estimated using the HO signal is indicated, the specific estimation process of the traffic flow is described in detail in the above-mentioned specific implementation manner, which will not be described in detail here.

[0120] When q N,i When the vehicle flow (i.e., traffic flow) estimated using NLU signals, the specific estimation process of traffic flow is as follows: the number of NLU events occurring at a certain location within a certain period of time is used to estimate traffic flow. In cellular networks, in order to ensure normal communication of the cellular network and to facilitate and find oneself more quickly, mobile phones and other communication devices will adopt a series of mechanisms to keep their own location information updated; NLU is one of the update mechanisms, and NLU signals are a very important and frequently occurring cellular network signal, which occurs when the mobile phone crosses LA. Figure 13 As shown, the road contains three LAs. When a vehicle passes through LA j-1 (Right now Drive to LA j (Right now When the phone is between two LAs (i.e. An NLU occurs at LA. Similarly, as the vehicle continues to move forward, when it j Exit and enter LA j+1 , the second NLU event will occur.

[0121] like Figure 13 As shown, assuming that every hour occurs in LA j-1 with LA j The number of NLU signals at the junction is n j , which happened in LA j with LA j+1 The number of NLU signals at the junction is n j+1 ,q i ,q i+1 ,q i+2 Respectively in LA j The traffic flow at the left boundary of the three cells in . Therefore, the number of NLU signals generated by the cell is estimated. i The formula for traffic flow is as follows:

[0122]

[0123] in, In LA j-1 with LA j Estimated traffic flow at the junction, q i Indicates the actual traffic flow through the location; n j Indicates LA jThe number of NLU signals within.

[0124] Because the mobile phone moving inside LA does not trigger NLU events, the Cell inside LA i+1 With Cell i+2 The number of NLU signals that occurred cannot be obtained; for these cells, the formula for traffic flow is as follows:

[0125]

[0126] in, and They represent the traffic flow estimated based on the number of HO signals at two locations, n j+1 Representatives in LA j+1 The number of NLU signals that occurred, i and j are both positive integers.

[0127] As another specific embodiment of the present invention, the number of PLU signals generated in a cell is used to estimate the vehicle speed. As a location update mechanism, periodic location update (PLU) can avoid the problem of location information update delay caused by the mobile phone staying in a certain LA for too long and unable to trigger NLU. For PLU events, since the last location update of the mobile phone, a set cycle time t c , the phone automatically triggers a PLU event. Furthermore, since the base station records the location of a CA event when it occurs, the CA event also updates the phone's location. Therefore, the impact of the CA event on the PLU event must be considered. The characteristics of the PLU event indicate that when the call interval is greater than the cycle time, the PLU event will normally be triggered after the last location update, a certain cycle time has passed. If the call interval is less than the cycle time, the CA event will occur first, and the PLU event originally triggered after the cycle time will not occur. The timer will reset from the moment the CA event occurred.

[0128] Therefore, estimating traffic information by using the estimated probability density function of the call time and the number of cellular network signals includes estimating vehicle speed by using the estimated probability density function of the call interval time and the number of PLU signals. Assuming that the interval between two adjacent PLU signals is one cycle time, in the first case:

[0129] like Figure 14 As shown, when the first PLU event occurs at time t0, after a cycle event t c Later, in Cell i The second PLU event occurred in the cell; the vehicle entered and exited the cell i The times are t1 and t3 respectively; the cycle time t cThe duration of the cycle is usually set by each network operator. In the present invention, the cycle time t is set c The duration is 1 hour. In order to better demonstrate the process of estimating vehicle speed using PLU signal, Figure 15 As shown, from Figure 15 The following information can be obtained:

[0130] The time difference between t0 and t2 is the interval between two PLU events, which is a cycle time t c ;

[0131] Assume that the vehicle is in Cell i The average speed of the vehicles in the i , Cell i The coverage length on the target road segment is l i , the vehicle is in Cell i The travel time within can be calculated by l i / u i to obtain;

[0132] If you want the mobile phone to have a second PLU event at time t2, you need to ensure that the call interval time t a Greater than the cycle time t c In the case of t a >t c , cycle time t c Need to be greater than entering the Cell i The time is less than the time to exit the cell i time, i.e. t c <x+l i / u i .

[0133] Therefore, when the inter-talk time is greater than the cycle time, the relationship between the number of PLU signals and the estimated vehicle speed is as follows:

[0134]

[0135] Among them, p s1 Indicates the Cell in the first case i The number of PLU signals recorded at Cell i represents the i-th cell, i is a positive integer; q i Represents Cell i Traffic flow at the location; Pr(t a >t c ) represents the call interval time t a Greater than the cycle time t c The probability of Represents the cycle time t c Greater than entering the Cell iThe time is less than the time to exit the cell i The probability of time, x, is determined by the time the call arrives and the time the call enters the cell. i The moment is calculated, Indicates the estimated Cell i Vehicle speed at i Represents Cell i Length of coverage on highways;

[0136] The second case: When the call interval time t a Less than the cycle time t c When the timer triggers the PLU event, the phone will update its location through the CA event before the timer triggers the PLU event; therefore, it will affect the time when the second PLU event occurs. Figure 16 As shown, after the first PLU event occurs at time t0, after t a The first CA event occurs after time t, and the call interval time is less than the cycle time; therefore, the timer will be reset and the cycle time t will pass again. c After that, the second PLU event occurs at time t3; at this time, the call interval time is greater than the cycle time, so the second CA event occurs at time t5 after the first CA event. In order to better demonstrate the process of estimating vehicle speed using PLU signals, as shown in the figure below: Figure 17 As shown, from Figure 17 The following information can be obtained:

[0137] The time difference from t0 to t1 is less than the cycle time t c ;

[0138] The time difference from time t1 to time t3 is the cycle time t c ;

[0139] Assume that the vehicle is in Cell j The average speed of the vehicles in the j , Cell j The coverage distance on the target road segment is l j , the vehicle is in Cell j The travel time within can be calculated by l j / u j to obtain;

[0140] If you want the mobile phone to have a second PLU event at time t3, you need to ensure that the first call interval is greater than the cycle time (i.e., t a <t c ), the second call interval is greater than the cycle time (ie t a >t c ), cycle time t c Greater than entering the Cell jThe time is less than the time to exit the cell j time, i.e. t c <y+l j / u j .

[0141] Therefore, when the inter-talk time is less than the cycle time, the relationship between the number of PLU signals and the estimated vehicle speed is as follows:

[0142]

[0143] Among them, p s2 Indicates the Cell in the second case j The number of PLU signals recorded at Cell j represents the jth cell, where j is a positive integer; q j Represents Cell j Traffic flow at the location, Pr(t a <t c ) represents the call interval time t a Less than the cycle time t c probability; Indicates the first call interval time t a Greater than the cycle time t c , the second call interval time t a Greater than the cycle time t c , cycle time t c Greater than entering the Cell j The time is less than the time to exit the cell j The probability y is determined by the time the call arrives and the time the call enters the cell j The moment is calculated, Indicates the estimated Cell j Vehicle speed at j Represents Cell j Length of coverage on highways;

[0144] Combining the above two situations, the relationship between the number of PLU signals and the estimated vehicle speed is as follows:

[0145]

[0146] The probability that the call interval is greater than the cycle time is as follows:

[0147]

[0148] According to formula (24) and formula (25), in Cell m The estimated average vehicle speed within is calculated using the following formula:

[0149]

[0150] Among them, p m Indicates Cell in two cases m The total number of PLU signals within, Indicates that in Cell m The estimated average speed within q m Indicates Cell in two cases m Traffic flow in the a >t c ) represents the call interval time t a Greater than the cycle time t c The probability of l m Represents Cell m Coverage length on the highway, Cell m Represents the mth cell.

[0151] It should be noted that: λ1, μ1, σ1, μ2, σ2, w involved in formula (12)-formula (26) 11 、w 12 、w 13 ,λ2,μ3,σ3,μ4,σ4,w 21 、w 22 、w 23 、 and The meanings are the same as those described above, so they are not repeated in formulas (12) to (26).

[0152] Preferably, in the cumulative distribution function of call holding time, λ1 is 1, μ1 is 1, σ1 is 0.2, μ2 is -0.09116, σ2 is 0.42699, and w 11 The value is 0.2, w 12 The value is 0.4, w 13 The value is 0.4; in the cumulative distribution function of the call interval, λ2 is 60, μ3 is 1 / 60, σ3 is 0.2 / 3600, μ4 is -4.18551, σ4 is 0.42699, w 21 The value is 0.2, w 22 The value is 0.4, w 23 The value is 0.4.

[0153] In a specific implementation of the present invention, in a traffic information estimation model based on cellular network signals, call holding time and call interval data are statistically obtained from cellular mobile vehicle data. Call holding time can be calculated by extracting the call start and end times for the same mobile phone from the cellular mobile vehicle data, while call interval time can be calculated by extracting the time difference between two consecutive call activation (CA) events for each mobile phone from the cellular mobile vehicle data. To calculate the true cumulative distribution function of call holding time and call interval time, the longest and shortest call holding times, or the longest and shortest call interval times, are statistically obtained from the cellular mobile vehicle data. Next, the statistically obtained time range (i.e., within the range from longest to shortest) is divided into several equal parts. The number of call holding times or call interval times within each small time interval is counted, and the number of call holding times or call interval times within each small time interval is calculated. The number of call holding times or call interval times within each small time interval is then compared with the total number to obtain the probability of falling within a certain range. Finally, the probabilities for each interval are cumulatively added up step by step to generate the true cumulative distribution function of call holding time and the true cumulative distribution function of call interval time. Because the actual cumulative distribution function is a discrete set and cannot be used for actual calculations, and the probability density function that fits this distribution is multimodal and cannot be simulated using a single probability distribution model, the cumulative distribution function generator of the present invention can estimate the target cumulative distribution functions of call hold time and call interval time by learning from these two data sets to fit these two mixed distributions.

[0154] Through actual fitting tests, such as Figure 18 As shown, it can be seen that the estimated cumulative distribution function (indicated by asterisks) and the actual cumulative distribution function (indicated by plus signs) almost coincide, where Figure 18 The left half represents the call holding time, and the right half represents the call interval time; this means that the cumulative distribution function generator of the present invention can effectively extract distribution patterns from actual data; therefore, the estimated cumulative distribution functions of the call holding time and call interval time can be well applied to traffic information estimation based on cellular network signals.

[0155] The following example uses the traffic flow, vehicle speed, traffic density, and number of various cellular network signals recorded by Cell4 on a target road section in one day as an example. The specific recorded data is shown in Table 1 below:

[0156] Table 1 Traffic information and cellular network signal data recorded by Cell4 in one day

[0157]

[0158]

[0159] like Figure 19 As shown, the Figure 19 It shows the changes in traffic volume (vehicles / hour), the number of NLU signals (events / hour), and the number of HO signals (events / hour) in a certain day; Figure 19 As shown in (1), the curve of the number of NLU signals of the cellular network is represented by a dotted triangle line, the curve of the number of HO cellular network signals is represented by a dotted circle line segment, and the traffic flow is represented by a dotted star line; the left axis represents the traffic flow, and the right axis represents the number of cellular network signals. As can be seen from the figure, the dotted star line and the dotted triangle line almost overlap, while the dotted circle line segment floats around the dotted triangle line; therefore, it can be seen that the NLU signal and the HO signal are positively correlated with the traffic flow, and it is reasonable to use the number of NLU signals and HO signals of the cellular network to estimate the traffic flow.

[0160] like Figure 19 As shown, the Figure 19 The relationship between vehicle speed (km / h), number of CA signals (events / hour), and number of PLU signals (events / hour) in a given day is shown. Figure 19 As shown in (2), the curve of the change in the number of CA signals of the cellular network is represented by a dotted line segment, the curve of the change in the number of PLU signals of the cellular network is represented by a dotted triangle line, and the vehicle speed is represented by a dashed asterisk line. To show the relationship between them, the speed axis is inverted. As can be seen from the figure, the change trend of the dotted line segment and the dotted triangle line is the same as the change trend of the inverted dashed asterisk line. Therefore, it can be considered that the number of CA and PLU signals of the cellular network is inversely proportional to the vehicle speed; therefore, it is reasonable to use the number of CA signals and the number of PLU signals to estimate the vehicle speed.

[0161] After actual data testing, before optimization, the average MAPE (i.e., mean absolute percentage error) of traffic flow estimated using the HO signal of the cellular network was 11.92%. After optimization using the method of the present invention, the average MAPE can be reduced to 8.07%; the average MAPE of vehicle speed estimated using the CA signal of the cellular network was 13.97%. After optimization using the method of the present invention, the average MAPE can be reduced to 1.80%; the average MAPE of vehicle speed estimated using the PLU signal of the cellular network was 21.56%. After optimization using the method of the present invention, the average MAPE can be reduced to 9.36%. The results show that the technical solution of the present invention can effectively reduce the estimation error caused by randomness, making traffic information estimation more accurate.

[0162] Although the specific embodiments of the present invention are described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for optimizing traffic information estimation, characterized by: The optimization method comprises: Constructing a neural network model for optimization, the neural network model comprising an input layer, an output layer, and six intermediate layers; the output layer comprises one neuron, the input layer comprises at least four neurons, and each of the intermediate layers comprises at least four neurons; Required parameters are input through the input layer, and each parameter is input through a neuron. The required parameters include the number of cellular network signals in the current cell, the number of cellular network signals in the previous cell, the number of cellular network signals in the next cell, and at least one piece of estimated traffic information. Traffic information is estimated using the estimated probability density function of the call time and the number of cellular network signals. The traffic information includes traffic flow or vehicle speed. The input parameters are optimized through each of the intermediate layers. The optimized estimated traffic information is output through the output layer. When estimating traffic flow using HO signals of a cellular network, the parameters input through the input layer include the number of HO signals of a preceding cell, the number of HO signals of a preceding cell, the number of HO signals of a succeeding cell, and the estimated traffic flow.

2. The method for optimizing traffic information estimation according to claim 1, characterized in that: The neurons in the input layer, output layer and intermediate layer all use LeakyReLU as the activation function.

3. The method for optimizing traffic information estimation according to claim 1, characterized in that: The input layer includes four neurons; the number of neurons in the six intermediate layers is 8, 12, 16, 12, 8, and 4 from the input end to the output end.

4. The method for optimizing traffic information estimation according to claim 1, characterized in that: When estimating vehicle speed using CA signals of a cellular network, the parameters input through the input layer include the number of CA signals of the preceding cell, the number of CA signals of the preceding cell, the number of CA signals of the succeeding cell, the estimated traffic flow, and the estimated vehicle speed.

5. The method for optimizing traffic information estimation according to claim 4, characterized in that: The input layer includes five neurons; the number of neurons in the six intermediate layers is 10, 15, 20, 15, 10, and 5 from the input end to the output end.

6. The method for optimizing traffic information estimation according to claim 1, characterized in that: When estimating vehicle speed using the PLU signal of the cellular network, the parameters input through the input layer include the number of PLU signals of the front cell, the number of PLU signals of the previous cell, the number of PLU signals of the next cell, and the estimated vehicle speed.

7. The method for optimizing traffic information estimation according to claim 6, characterized in that: The input layer includes four neurons; the number of neurons in the six intermediate layers is 8, 12, 16, 12, 8, and 4 from the input end to the output end.

Images