CN110942211A - Prediction arrival time prediction method and device based on deep neural network - Google Patents
Prediction arrival time prediction method and device based on deep neural network Download PDFInfo
- Publication number
- CN110942211A CN110942211A CN201911277290.3A CN201911277290A CN110942211A CN 110942211 A CN110942211 A CN 110942211A CN 201911277290 A CN201911277290 A CN 201911277290A CN 110942211 A CN110942211 A CN 110942211A
- Authority
- CN
- China
- Prior art keywords
- neural network
- deep neural
- time
- driving time
- estimated
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/29—Geographical information databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Economics (AREA)
- General Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- General Business, Economics & Management (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Tourism & Hospitality (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Game Theory and Decision Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Development Economics (AREA)
- Remote Sensing (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Traffic Control Systems (AREA)
Abstract
The embodiment of the invention provides a prediction method and a prediction device of estimated time of arrival based on a deep neural network, wherein the method comprises the following steps: processing a path planning query request of a user based on a path planning algorithm, and outputting an action track; and inputting the action track and the external attribute information into a trained deep neural network, and outputting the estimated driving time. The prediction arrival time prediction method and device based on the deep neural network can capture spatial correlation and time correlation, integrate attribute information of external factors and combine local and global driving time estimation methods. Therefore, the network model obtains accurate driving time estimation and has good robustness.
Description
Technical Field
The invention relates to the technical field of space-time big data mining, in particular to a prediction arrival time prediction method and device based on a deep neural network.
Background
For a given path and departure Time, Estimated Time of Arrival (Arrival Time of Arrival) prediction is the most fundamental problem in the travel services industry. Such as a path planning system for navigation services, an online taxi intelligent-dispatch system, etc. Accurate driving time estimation helps to better plan routes and avoid congested roads, thereby helping to alleviate traffic congestion. The problem has been widely studied and is a basic function of the navigation electronic map, but the precision of the estimated arrival time prediction in the current navigation electronic map is still to be improved. The pre-estimated time of arrival prediction may be translated into a driving time estimation problem. The pre-estimated arrival time is calculated and updated by repeated invocations of the driving time estimation module.
For a given path and departure time, driving time estimation is a complex problem. It is influenced by various complex factors, spatial correlation, temporal correlation, and also some external factors, such as weather information, driving habits, week periodicity, etc. The complex spatial correlation is mainly embodied in complex tracks such as straight going and turning at the intersection. And secondly, going up and down the ramp, driving the main road and switching the auxiliary road. In addition, there are also instances of a u-turn, where the upstream and downstream speeds are often inconsistent for the same segment of the road. Modeling difficulties are therefore high for these complex cases. The time relevance is usually reflected in the periodicity of time, and the passing speed in the peak working period on the same road and off duty is usually greatly different from the statistical average passing speed. The travel tracks of the holidays and the like of the next festival can be greatly different.
The current mainstream algorithms for driving time estimation are mainly divided into two types: one is a local driving time estimation algorithm. He is based on a method of road segment cumulative time summation. The route planning result from the starting point to the end point is divided into small road sections, the passing time of each small road section is calculated respectively, and the time obtained by summing the passing times of all the road sections is used as the estimated arrival time. The method can accurately predict the driving time of each road section, but cannot model complex traffic conditions such as intersections, traffic lights and steering. In addition, if there are too many road segments, local errors may accumulate. Another is a method of global driving time estimation, which directly estimates the travel time of the entire journey from the start to the end. The global approach can capture traffic condition information. According to the characteristics of the path length distribution, the longer the path is, the less the track history data can be collected, and the track data is provided by a small number of drivers. The accuracy of the prediction result of the arrival time of the long track is low, and the prediction result is a problem to be solved in the global travel time estimation.
Disclosure of Invention
The embodiment of the invention provides a prediction method and a prediction device for estimated arrival time based on a deep neural network, which are used for solving the problems of larger error and lower accuracy of the existing driving time estimation method.
In a first aspect, an embodiment of the present invention provides a prediction method for estimated time of arrival based on a deep neural network, including:
processing a path planning query request of a user based on a path planning algorithm, and outputting an action track;
and inputting the action track and the external attribute information into a trained deep neural network, and outputting the estimated driving time.
Further, before the inputting the action track and the external attribute information into the trained deep neural network, the method further comprises:
and training a preset deep neural network model based on the historical track data to obtain a trained deep neural network.
Further, the training a preset deep neural network model based on the historical track data to obtain a trained deep neural network includes:
carrying out data preprocessing on the historical track data;
establishing the deep neural network model, and designing a loss function corresponding to the deep neural network model;
and training the deep neural network model to obtain a trained deep neural network.
Further, the deep neural network model includes:
the external factor attribute value processing submodule is used for representing all external attributes by vectors and inputting the external attributes into a neural network, and modeling external attribute factors;
the space-time convolution module is used for extracting features;
and the multi-task learning submodule is used for accurately estimating the global driving time.
Further, the training the deep neural network model to obtain a trained deep neural network includes:
calculating true values and estimated values of local path driving time and global driving time for each section of historical track data;
the difference between the weight minimization pre-truth value and the estimate value for each neuron is reconciled based on a mechanism of back propagation.
Further, the processing a path planning query request of a user based on a path planning algorithm and outputting an action trajectory includes:
extracting position information and departure time information in a path planning query request of a user;
and calculating a plurality of action tracks based on a path planning algorithm.
Further, the inputting the action track and the external attribute information into a trained deep neural network and outputting the estimated driving time includes:
and converting the input action track and external attribute information into the estimated driving time for output based on the forward propagation mode of the deep neural network.
In a second aspect, an embodiment of the present invention further provides an estimated time of arrival prediction apparatus based on a deep neural network, including:
the action track prediction module is used for processing a path planning query request of a user and outputting an action track based on a path planning algorithm;
and the driving time prediction module is used for inputting the action track and the external attribute information into the trained deep neural network and outputting the predicted driving time.
In a third aspect, an embodiment of the present invention provides an electronic device, including:
a processor, a memory, a communication interface, and a bus; the processor, the memory and the communication interface complete mutual communication through the bus; the memory stores program instructions executable by the processor, the processor invoking the program instructions to perform the above-described deep neural network-based estimated time of arrival prediction method.
In a fourth aspect, an embodiment of the present invention provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the method for predicting estimated time of arrival based on a deep neural network.
According to the prediction method and device based on the deep neural network, a deep learning neural network model is trained in advance based on historical track data, the neural network model can capture spatial correlation and time correlation and integrate attribute information of external factors, and a local driving time estimation method and a global driving time estimation method are combined. Therefore, the action track and the external attribute information are input into the deep neural network, accurate estimated driving time can be output, and the deep neural network has good robustness.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
Fig. 1 is a schematic flow chart of a predicted arrival time prediction method based on a deep neural network according to an embodiment of the present invention;
fig. 2 is a schematic structural diagram of an estimated time of arrival prediction apparatus based on a deep neural network according to an embodiment of the present invention;
fig. 3 is a block diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some embodiments, but not all embodiments, of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Fig. 1 is a schematic flow chart of a prediction time-of-arrival prediction method based on a deep neural network according to an embodiment of the present invention, as shown in fig. 1, including:
101. processing a path planning query request of a user based on a path planning algorithm, and outputting an action track;
102. and inputting the action track and the external attribute information into a trained deep neural network, and outputting the estimated driving time.
It should be noted that, in the embodiment of the present invention, an end-to-end network model is substantially designed, which integrates a local driving time estimation algorithm and a global driving time estimation algorithm, and utilizes a cyclic neural network processed by a geographical convolution layer and a time series, and on the basis, external factors are embedded into the model as attributes. The model can capture spatial correlation, temporal correlation and fuse attribute information of external factors, and combines local and global driving time estimation methods. Therefore, the network model obtains accurate driving time estimation and has good robustness. The model has no requirement on computing capacity in a prediction stage, and can achieve the effect of real-time response. Specifically, in the embodiment of the present invention, a time-space convolution module is used to extract a feature map layer of a time and space dependency relationship from an original GPS track sequence. Based on the geographical convolution layer, an original GPS track sequence can be converted into a series of characteristic layers, and the characteristic layers can extract local spatial correlation. Temporal dependencies are learned from feature layers derived from geography-based convolutional layers and from the embedding of extrinsic features using a recurrent neural network (Long Short Term memories).
And secondly, adding embedded representation of road attribute information into the space-time convolution network model. The road attribute information includes road grade, whether the road contains complex intersections, and whether the road is a tunnel. Since the properties of a road have a direct relationship to the transit time through the road segment.
The embodiment of the invention also integrates the external factors into the network model in an attribute embedding mode. External factors include time, weather, driver information. And resolving the attribute of the month to which the departure time belongs, the attribute of the time segment in the day to which the week belongs, and the attribute of the holiday or not. In addition to the vehicle number, and weather attributes. These attributes are not represented in the model in a One-Hot-only manner, but by applying the method of embedding vector representation separately for different attributes. The overall network model designs a loss function consisting of two tasks. One is a local route driving time estimation task, and the other is a global route driving time estimation task from the start point to the end point. The network can thus learn both local and global travel times, balancing the loss of local and global path estimation by means of weighting.
Finally, in order to improve the accuracy of global travel time prediction, a multi-factor attention mechanism is added into the model to learn the local path weight in estimating the global path driving time. The theoretical basis for introducing this mechanism is that the uncertainty of the global driving time is often caused by several critical local paths. Such as intersections, traffic lights, congested road sections, etc. The attention mechanism learns the weights of the different local impairments based on the hidden representation of the different local paths and external factors.
The prediction method for the estimated arrival time based on the deep neural network provided by the embodiment of the invention is characterized in that a deep learning neural network model is trained in advance based on historical track data, and the neural network model can capture spatial correlation and temporal correlation and integrate attribute information of external factors, and is combined with a local driving time estimation method and a global driving time estimation method. Therefore, the action track and the external attribute information are input into the deep neural network, accurate estimated driving time can be output, and the deep neural network has good robustness.
On the basis of the above embodiment, before the inputting the action track and the external attribute information into the trained deep neural network, the method further includes:
and training a preset deep neural network model based on the historical track data to obtain a trained deep neural network.
In essence, the embodiment of the invention designs a deep neural network model, through processing a large amount of historical track data, since the historical track data has a time stamp, the time stamp can obtain a true value (actual driving time) of each track sequence through calculation, the network reversely propagates through a minimization loss function (difference between the actual driving time and a pre-estimated driving time), and the weight of each neuron is updated by using a gradient descent algorithm. And repeating iteration of the network model in the existing historical track data until the network converges, thus obtaining the complex ETA model. And storing the trained weight in the network model, thereby obtaining the trained neural network model.
On the basis of the above embodiment, the training a preset deep neural network model based on the historical trajectory data to obtain a trained deep neural network includes:
carrying out data preprocessing on the historical track data;
establishing the deep neural network model, and designing a loss function corresponding to the deep neural network model;
and training the deep neural network model to obtain a trained deep neural network.
The embodiment of the invention firstly needs to preprocess data, and the data preprocessing comprises the following steps:
preprocessing historical track data: abnormal value processing in track data: and deleting track points with abnormal numerical values in longitude, latitude, speed and timestamp fields and track points with inconsistent data types.
Track sequence outlier deletion.
Track point deviation correction: for a point with the speed of 0, namely a parking point, if a vehicle uploads a few or dozens of track points, the track points need to be combined into one track point according to the continuity of the front and rear track sequences. Specifically, the method comprises the steps of averaging and then fine-tuning according to sequence continuity.
And (3) longitude and latitude coordinate projection: and converting the longitude and latitude coordinates of the GPS track points into a Gaussian plane coordinate system, and then calculating the distance between two adjacent GPS track points.
Resampling trace points: the purpose of resampling the track points is to ensure that the distance intervals of any two adjacent track points are uniformly distributed. The resampling standard is that the distance between any two adjacent track points (except the last two track points) is in a fixed value range after resampling, and the distance is two hundred meters or more in an experiment. This speeds up the convergence of the network.
Calculating a road attribute value: and calculating the road attribute value of a road section consisting of two continuous track points through map data according to the sequence consisting of the resampled track points. Road grade, whether including complex intersections, whether being a tunnel.
Local path calculation: and calculating the accumulated distance and the accumulated running time by taking the starting point as the starting point according to the trajectory points output by resampling. These two attributes are input to the geo-convolved layers together with the longitude and latitude coordinates.
Statistical analysis of trajectory data: and counting the longitude, latitude, accumulated distance, accumulated time, the maximum value, the minimum value, the mean value and the variance of all track points, wherein the maximum value, the minimum value, the mean value and the variance are the total mileage and the whole driving time. This information will be used for normalization processing of the data. The normalization processing of the input sample data can simplify the initialization work, accelerate the network learning speed and accelerate the convergence.
Calculating the attribute value of the external factor: a set of extrinsic information attribute values is extracted for each segment of the track. The time stamp of the first track point of each track segment is used as the departure time. And taking the longitude and latitude coordinates of the first track point as a starting point. And inquiring weather information of the departure place at the departure time point according to the departure point information and the departure time information, and converting the weather information into attribute values of 0-3, wherein the attribute values of the weather (0-3 respectively represent sunny days, rainy days, foggy days and snowy days). Whether or not the holiday attribute belongs to the holiday is judged according to the departure time, and an attribute value (0/1,0 represents non-holiday) of the holiday attribute is calculated. And calculates a month attribute (0-11), a week attribute value (0-6), a time segment attribute value (0-1439), one time segment per minute, based on the departure time. The user ID is converted into a numerical number by means of a dictionary.
On the basis of the above embodiment, the deep neural network model includes:
the external factor attribute value processing submodule is used for representing all external attributes by vectors and inputting the external attributes into a neural network, and modeling external attribute factors;
the space-time convolution module is used for extracting features;
and the multi-task learning submodule is used for accurately estimating the global driving time.
It can be known from the content of the above embodiment that a neural network model is constructed in the embodiment of the present invention, and the model can be subdivided into three parts, namely, an external factor attribute value processing submodule, a space-time convolution module and a multitask learning submodule, specifically, the external factor attribute value processing submodule is used for inputting all external attributes into the neural network by vector representation, and modeling external attribute factors. Firstly, embedding processing is carried out on the external attribute values calculated by preprocessing, and the value of each external attribute is mapped into an expression vector. Then all the expression vectors of the external attributes and the global path length after the z-score standardization are connected in series to obtain the final output.
The reason for the embedding process is that the external attribute values are all class values and cannot be directly input into the neural network for processing. The attribute values are thus converted into low-dimensional vectors by means of an embedded representation. There are two methods for converting attribute values into vector representations, one is a one-hot encoding method, and the other is an embedding method. Compared with a single-hot coding mode, the embedded mode can reduce dimensionality and reduce calculation consumption. Embedded representations can represent semantic information, with vector representations of semantically similar attribute values often being more similar.
The space-time convolution module is composed of two parts, wherein the first part is a geographical convolution neural network layer. The sub-module processes the original GPS sequence into a series of feature maps. It can extract local spatial correlations. The second part is a recurrent neural network layer which can extract the time series correlation from the feature map layer generated by the first part.
Geographical convolutional neural network layer: the module mainly processes three attributes of longitude, latitude sequence and accumulated distance, namely a road section attribute vector. And converting the longitude and latitude of each GPS track point into a vector with a fixed length through nonlinear layer transformation. The vector can represent a geographic feature. And combining the feature vectors of all track points into a feature vector sequence according to the time sequence of the track. And performing 1-dimensional convolution operation on the feature vector sequence. The convolution kernel of the 1-dimensional convolution operation is a one-dimensional sliding window. The size of the sliding window and the number of filters for the convolution operation are set by human experimentation. The length of the sliding window is the length of the partial path. Finally, the outputs of all filters are connected in series to obtain a 2-dimensional vector. Since it is difficult to extract the geographical distance information by the 1-dimensional convolution, the length of the local path is also concatenated to the two-dimensional vector and output as a feature layer extracted as the geographical convolution layer. The feature map layer can represent the spatial dependence of all local paths. Which is the input to the recurrent neural network layer.
Specifically, whether the road grade contains a complex intersection or not and whether the three attributes are the tunnel or not are respectively expressed by attribute embedding (the same as the external factor attribute embedding processing) to obtain three attribute feature vectors. Respectively adding a latitude to the longitude and latitude sequences, then connecting the latitude sequences with the three attribute feature vectors in series, passing through a linear processing neural unit, connecting a nonlinear activation function behind the neural unit, then passing through a 1-dimensional convolution neural unit, and obtaining an output O1 after the output is activated by the nonlinear activation function. And slicing the attribute sequence of the accumulated distance to obtain two local sequence segments, solving the difference value of the two track sequence segments, and carrying out z-score standardization to obtain an output O2. And serially connecting O1 and O2 to obtain an output O of the geographical convolutional network model.
A recurrent neural network layer: the time dependence of the local path is learned by using the memory function of the long and short device memory network layer. The memory function of the long-term and short-term memory network is to control input and output streams through three gate control units (an input gate, a forgetting gate and an updating gate), so that some information which is not important enough can be forgotten, and the problems of gradient explosion and gradient disappearance are effectively avoided. The output of the layer of the geographical convolutional neural network is the input of the layer. The purpose of this layer is to further extract the temporal dependencies of the local paths.
The purpose of the multi-task learning submodule is to combine the local driving time estimation method and the global driving time estimation method to obtain more accurate global driving time estimation. The module mainly comprises two processing parts, namely local path driving time estimation and global path driving time estimation.
Local path driving time estimation: and each local characteristic sequence output by the space-time convolution module passes through two full-connection layers and then outputs the driving time estimation value of each local path.
Global path driving time estimation: since the length of each local signature sequence of the space-time convolution output is variable, each local signature sequence is first converted into a vector of fixed length by a pooling process. There are various pooling ways, including mean pooling, maximum pooling, etc. The mean pooling considers each local path as equally important, and the eigenvectors of the local paths are weighted and summed in an equally weighted form. An attention mechanism is adopted in the patent. The basis for this approach is that the uncertainty in the global driving time is often caused by several critical local paths. Such as intersections, traffic lights, congested road sections, etc. More attention is therefore required to these critical local paths. Attention mechanism pooling is also a weighted summation of the feature vectors of all global paths. Unlike mean pooling: the weight of the feature vector of each section of local path is obtained by the model through self learning. The weights of each segment of the local path are learned by attention mechanism pooling layers.
Attention mechanism pooling layer: the input of the layer comprises the local characteristic sequence of the space-time convolution output and the output of the external factor attribute value processing submodule. And converting the external factor attribute expression vector into a vector with the same length as the local characteristic sequence through a nonlinear network layer, and then obtaining the local path weight attention vector through inner product operation with the local characteristic sequence vector.
Residual fully connected sub-modules: and the output of the attention pooling layer is serially connected with the external factor vector and then input into the residual full-connection sub-module. The sub-modules are connected with a full connection layer through a plurality of staggered layers for processing, and finally, a network output layer of a neuron is added. The output layer outputs the estimated value of the global path driving time. The use of the spread connection can capture the spread of the nonlinear transformation, making the neural network training simpler and more robust.
On the basis of the above embodiment, the training of the deep neural network model to obtain a trained deep neural network includes:
calculating true values and estimated values of local path driving time and global driving time for each section of historical track data;
the difference between the weight minimization pre-truth value and the estimate value for each neuron is reconciled based on a mechanism of back propagation.
From the content of the above embodiment, it can be seen that the embodiment of the present invention constructs a neural network model, and on the basis of the neural network model, training and tuning are required, and specifically, the embodiment of the present invention calculates the true values of the local path driving time and the global driving time for each segment of historical track data. During the training process, the network will output an estimate of the local path driving time and an estimate of the global path driving time. The loss function of the network model is weighted by the local path driving time estimation loss and the global driving time estimation loss. The weights are set as hyper-parameters of the model based on empirical values. Reconciling the weight of each neuron through a back-propagation mechanism minimizes the difference between the pre-estimated driving time and the actual driving time truth.
On the basis of the above embodiment, the processing a path planning query request of a user based on a path planning algorithm and outputting an action trajectory includes:
extracting position information and departure time information in a path planning query request of a user;
and calculating a plurality of action tracks based on a path planning algorithm.
As can be seen from the above description, the embodiment of the present invention needs to process the path planning query request of the user and output the action trajectory. Specifically, a general ETA query includes information such as location information, departure time, and an attached user ID.
Then, the track is calculated by the above information, and the position information includes the positions of the start point and the end point. And calculating a plurality of vehicle running track points of different paths through a path planning algorithm. And resampling the trace points according to the intervals, and ensuring that the trace points are distributed at equal intervals.
And acquiring the rough arrival time, the whole mileage and the sub-segment average speed output by the path planning algorithm. And calculating the rough arrival time of each track point according to the average speed of the sub-segments.
The current weather attribute may be acquired based on the location information and the departure time information when calculating the external factor attribute value. According to the departure time selected by the user, a month attribute (0-11), a week attribute value (0-6), a time segment attribute value (0-1439, a time segment is every minute), a holiday information attribute value (0/1,0 represents non-holidays), and a weather attribute value (0-3, respectively represent sunny days, rainy days, foggy days, and snowy days) are calculated. The user ID is converted into a numerical number by means of a dictionary.
On the basis of the above embodiment, the inputting the action trajectory and the external attribute information into the trained deep neural network and outputting the estimated driving time includes:
and converting the input action track and external attribute information into the estimated driving time for output based on the forward propagation mode of the deep neural network.
According to the content of the embodiment, the trained network model is obtained, the track sequence and the external factor attribute value are calculated, and the information, the rough estimated arrival time and the rough estimated mileage are input into the constructed network model. The network converts all inputs into pre-estimated arrival times as outputs by means of forward propagation.
Fig. 2 is a schematic structural diagram of an estimated time of arrival prediction apparatus based on a deep neural network according to an embodiment of the present invention, as shown in fig. 2, including: an action track prediction module 201 and a driving time prediction module 202, wherein:
the action track prediction module 201 is configured to process a path planning query request of a user based on a path planning algorithm, and output an action track;
the driving time estimation module 202 is configured to input the action trajectory and the external attribute information into a trained deep neural network, and output estimated driving time.
Specifically, how to use the action trajectory prediction module 201 and the driving time prediction module 202 to execute the technical scheme of the prediction arrival time prediction method embodiment based on the deep neural network shown in fig. 1 is similar to the implementation principle and the technical effect, and details are not described here.
The prediction arrival time prediction device based on the deep neural network provided by the embodiment of the invention is characterized in that a deep learning neural network model is trained in advance based on historical track data, and the neural network model can capture spatial correlation and temporal correlation and integrate attribute information of external factors, and is combined with a local driving time estimation method and a global driving time estimation method. Therefore, the action track and the external attribute information are input into the deep neural network, accurate estimated driving time can be output, and the deep neural network has good robustness.
On the basis of the above embodiment, the apparatus further includes:
and the training module is used for training a preset deep neural network model based on the historical track data to obtain a trained deep neural network.
On the basis of the above embodiment, the training module is configured to:
carrying out data preprocessing on the historical track data;
establishing the deep neural network model, and designing a loss function corresponding to the deep neural network model;
and training the deep neural network model to obtain a trained deep neural network.
On the basis of the above embodiment, the deep neural network model includes:
the external factor attribute value processing submodule is used for representing all external attributes by vectors and inputting the external attributes into a neural network, and modeling external attribute factors;
the space-time convolution module is used for extracting features;
and the multi-task learning submodule is used for accurately estimating the global driving time.
On the basis of the above embodiment, the training of the deep neural network model to obtain a trained deep neural network includes:
calculating true values and estimated values of local path driving time and global driving time for each section of historical track data;
the difference between the weight minimization pre-truth value and the estimate value for each neuron is reconciled based on a mechanism of back propagation.
On the basis of the above embodiment, the processing a path planning query request of a user based on a path planning algorithm and outputting an action trajectory includes:
extracting position information and departure time information in a path planning query request of a user;
and calculating a plurality of action tracks based on a path planning algorithm.
On the basis of the above embodiment, the inputting the action trajectory and the external attribute information into the trained deep neural network and outputting the estimated driving time includes:
and converting the input action track and external attribute information into the estimated driving time for output based on the forward propagation mode of the deep neural network.
Fig. 3 is a block diagram of an electronic device according to an embodiment of the present invention, and referring to fig. 3, the electronic device includes: a processor (processor)301, a communication Interface (communication Interface)302, a memory (memory)303 and a bus 304, wherein the processor 301, the communication Interface 302 and the memory 303 complete communication with each other through the bus 304. Processor 301 may call logic instructions in memory 303 to perform the following method: processing a path planning query request of a user based on a path planning algorithm, and outputting an action track; and inputting the action track and the external attribute information into a trained deep neural network, and outputting the estimated driving time.
An embodiment of the present invention discloses a computer program product, which includes a computer program stored on a non-transitory computer readable storage medium, the computer program including program instructions, when the program instructions are executed by a computer, the computer can execute the methods provided by the above method embodiments, for example, the method includes: processing a path planning query request of a user based on a path planning algorithm, and outputting an action track; and inputting the action track and the external attribute information into a trained deep neural network, and outputting the estimated driving time.
Embodiments of the present invention provide a non-transitory computer-readable storage medium, which stores computer instructions, where the computer instructions cause the computer to perform the methods provided by the above method embodiments, for example, the methods include: processing a path planning query request of a user based on a path planning algorithm, and outputting an action track; and inputting the action track and the external attribute information into a trained deep neural network, and outputting the estimated driving time.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method according to each embodiment or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A prediction arrival time prediction method based on a deep neural network is characterized by comprising the following steps:
processing a path planning query request of a user based on a path planning algorithm, and outputting an action track;
and inputting the action track and the external attribute information into a trained deep neural network, and outputting the estimated driving time.
2. The deep neural network-based prediction of time of arrival method of claim 1, wherein before the inputting the action trajectory and external attribute information into the trained deep neural network, the method further comprises:
and training a preset deep neural network model based on the historical track data to obtain a trained deep neural network.
3. The method for predicting estimated arrival time based on the deep neural network according to claim 2, wherein the training a preset deep neural network model based on the historical track data to obtain a trained deep neural network comprises:
carrying out data preprocessing on the historical track data;
establishing the deep neural network model, and designing a loss function corresponding to the deep neural network model;
and training the deep neural network model to obtain a trained deep neural network.
4. The method of claim 3, wherein the deep neural network model comprises:
the external factor attribute value processing submodule is used for representing all external attributes by vectors and inputting the external attributes into a neural network, and modeling external attribute factors;
the space-time convolution module is used for extracting features;
and the multi-task learning submodule is used for accurately estimating the global driving time.
5. The method of claim 3, wherein the training the deep neural network model to obtain the trained deep neural network comprises:
calculating true values and estimated values of local path driving time and global driving time for each section of historical track data;
the difference between the weight minimization pre-truth value and the estimate value for each neuron is reconciled based on a mechanism of back propagation.
6. The method for predicting estimated arrival time based on the deep neural network as claimed in claim 1, wherein the processing the path planning query request of the user based on the path planning algorithm and outputting the action track comprises:
extracting position information and departure time information in a path planning query request of a user;
and calculating a plurality of action tracks based on a path planning algorithm.
7. The method for predicting estimated time of arrival based on deep neural network as claimed in claim 1, wherein said inputting said action track and external attribute information into a trained deep neural network and outputting estimated driving time comprises:
and converting the input action track and external attribute information into the estimated driving time for output based on the forward propagation mode of the deep neural network.
8. A predicted arrival time prediction device based on a deep neural network is characterized by comprising the following components:
the action track prediction module is used for processing a path planning query request of a user and outputting an action track based on a path planning algorithm;
and the driving time prediction module is used for inputting the action track and the external attribute information into the trained deep neural network and outputting the predicted driving time.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program implements the steps of the method for predicting estimated time of arrival based on a deep neural network as claimed in any one of claims 1 to 7.
10. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements the steps of the method for predicting estimated time of arrival of a multi-industry deep neural network as claimed in any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911277290.3A CN110942211A (en) | 2019-12-12 | 2019-12-12 | Prediction arrival time prediction method and device based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911277290.3A CN110942211A (en) | 2019-12-12 | 2019-12-12 | Prediction arrival time prediction method and device based on deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110942211A true CN110942211A (en) | 2020-03-31 |
Family
ID=69911050
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911277290.3A Pending CN110942211A (en) | 2019-12-12 | 2019-12-12 | Prediction arrival time prediction method and device based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110942211A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111667693A (en) * | 2020-06-15 | 2020-09-15 | 腾讯科技（深圳）有限公司 | Method, apparatus, device and medium for determining estimated time of arrival |
| CN111724586A (en) * | 2020-05-11 | 2020-09-29 | 清华大学 | Commuting time prediction method, and training method and device of commuting time prediction model |
| CN111982138A (en) * | 2020-07-09 | 2020-11-24 | 北京百度网讯科技有限公司 | Prediction model obtaining and path planning method, device and storage medium |
| CN112541638A (en) * | 2020-12-21 | 2021-03-23 | 北京邮电大学 | Method for estimating travel time of vehicle connected with Internet |
| CN112859848A (en) * | 2021-01-06 | 2021-05-28 | 国电内蒙古东胜热电有限公司 | Wireless navigation method and system of pipeline robot |
| CN113837268A (en) * | 2021-09-18 | 2021-12-24 | 北京百度网讯科技有限公司 | Method, device, equipment and medium for determining track point state |
| CN113902427A (en) * | 2021-12-09 | 2022-01-07 | 腾讯科技（深圳）有限公司 | Method, device and equipment for determining estimated arrival time and storage medium |
| CN114116944A (en) * | 2021-11-30 | 2022-03-01 | 重庆七腾科技有限公司 | Trajectory prediction method and device based on time attention convolution network |
| WO2022161045A1 (en) * | 2021-01-26 | 2022-08-04 | 深圳市优必选科技股份有限公司 | Estimated time of arrival calculation method, and system and mobile machine using same |
| CN115394088A (en) * | 2022-10-31 | 2022-11-25 | 江苏博宇鑫信息科技股份有限公司 | Crossing traffic time prediction method based on space-time attention mechanism |
| CN117975178A (en) * | 2024-04-02 | 2024-05-03 | 北京市计量检测科学研究院 | Taxi track data analysis method based on big data analysis |
| CN118052347A (en) * | 2024-04-16 | 2024-05-17 | 北京航空航天大学 | Travel time estimation method and system based on travel track sequence of floating car |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109035761A (en) * | 2018-06-25 | 2018-12-18 | 复旦大学 | Travel time estimation method based on back-up surveillance study |
| CN110073426A (en) * | 2017-11-23 | 2019-07-30 | 北京嘀嘀无限科技发展有限公司 | The system and method for Estimated Time of Arrival |
-
2019
- 2019-12-12 CN CN201911277290.3A patent/CN110942211A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110073426A (en) * | 2017-11-23 | 2019-07-30 | 北京嘀嘀无限科技发展有限公司 | The system and method for Estimated Time of Arrival |
| CN109035761A (en) * | 2018-06-25 | 2018-12-18 | 复旦大学 | Travel time estimation method based on back-up surveillance study |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111724586B (en) * | 2020-05-11 | 2022-03-11 | 清华大学 | Commuting time prediction method, and training method and device of commuting time prediction model |
| CN111724586A (en) * | 2020-05-11 | 2020-09-29 | 清华大学 | Commuting time prediction method, and training method and device of commuting time prediction model |
| CN111667693A (en) * | 2020-06-15 | 2020-09-15 | 腾讯科技（深圳）有限公司 | Method, apparatus, device and medium for determining estimated time of arrival |
| CN111982138A (en) * | 2020-07-09 | 2020-11-24 | 北京百度网讯科技有限公司 | Prediction model obtaining and path planning method, device and storage medium |
| CN112541638A (en) * | 2020-12-21 | 2021-03-23 | 北京邮电大学 | Method for estimating travel time of vehicle connected with Internet |
| CN112541638B (en) * | 2020-12-21 | 2022-07-12 | 北京邮电大学 | Method for estimating travel time of vehicle connected with Internet |
| CN112859848A (en) * | 2021-01-06 | 2021-05-28 | 国电内蒙古东胜热电有限公司 | Wireless navigation method and system of pipeline robot |
| CN112859848B (en) * | 2021-01-06 | 2023-03-10 | 国电内蒙古东胜热电有限公司 | Wireless navigation method and system of pipeline robot |
| WO2022161045A1 (en) * | 2021-01-26 | 2022-08-04 | 深圳市优必选科技股份有限公司 | Estimated time of arrival calculation method, and system and mobile machine using same |
| US11933615B2 (en) | 2021-01-26 | 2024-03-19 | Ubkang (Qingdao) Technology Co., Ltd. | Estimated time of arrival calculating method and system and mobile machine using the same |
| CN113837268A (en) * | 2021-09-18 | 2021-12-24 | 北京百度网讯科技有限公司 | Method, device, equipment and medium for determining track point state |
| CN113837268B (en) * | 2021-09-18 | 2024-03-12 | 北京百度网讯科技有限公司 | Method, device, equipment and medium for determining track point state |
| CN114116944A (en) * | 2021-11-30 | 2022-03-01 | 重庆七腾科技有限公司 | Trajectory prediction method and device based on time attention convolution network |
| CN114116944B (en) * | 2021-11-30 | 2024-06-11 | 重庆七腾科技有限公司 | Track prediction method and device based on time attention convolution network |
| CN113902427A (en) * | 2021-12-09 | 2022-01-07 | 腾讯科技（深圳）有限公司 | Method, device and equipment for determining estimated arrival time and storage medium |
| CN115394088A (en) * | 2022-10-31 | 2022-11-25 | 江苏博宇鑫信息科技股份有限公司 | Crossing traffic time prediction method based on space-time attention mechanism |
| CN117975178A (en) * | 2024-04-02 | 2024-05-03 | 北京市计量检测科学研究院 | Taxi track data analysis method based on big data analysis |
| CN117975178B (en) * | 2024-04-02 | 2024-05-28 | 北京市计量检测科学研究院 | Taxi track data analysis method based on big data analysis |
| CN118052347A (en) * | 2024-04-16 | 2024-05-17 | 北京航空航天大学 | Travel time estimation method and system based on travel track sequence of floating car |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110942211A (en) | Prediction arrival time prediction method and device based on deep neural network | |
| CN109670277B (en) | Travel time prediction method based on multi-mode data fusion and multi-model integration | |
| US11725953B2 (en) | Private and personalized estimation of travel time | |
| Wang et al. | When will you arrive? Estimating travel time based on deep neural networks | |
| Tran et al. | DeepTRANS: a deep learning system for public bus travel time estimation using traffic forecasting | |
| CN113642633A (en) | Method, apparatus, device and medium for classifying driving scene data | |
| US11536582B2 (en) | System and method for estimating travel time and distance | |
| CN114450557B (en) | Route deviation quantification and vehicle route learning based thereon | |
| Gao et al. | Aggressive driving saves more time? Multi-task learning for customized travel time estimation. | |
| CN111667693B (en) | Method, apparatus, device and medium for determining estimated time of arrival | |
| CN110598917B (en) | Destination prediction method, system and storage medium based on path track | |
| US20240046785A1 (en) | System and method for predicting road traffic speed | |
| CN114360239A (en) | Traffic prediction method and system for multilayer space-time traffic knowledge map reconstruction | |
| CN114742280B (en) | Road condition prediction method and corresponding model training method, device, equipment and medium | |
| Gao et al. | CTTE: Customized travel time estimation via mobile crowdsensing | |
| CN113888867B (en) | Parking space recommendation method and system based on LSTM (least squares) position prediction | |
| Pan et al. | Data-driven time series prediction based on multiplicative neuron model artificial neuron network | |
| Kumar et al. | Analytical approach to identify the optimum inputs for a bus travel time prediction method | |
| Sheng et al. | Deep spatial-temporal travel time prediction model based on trajectory feature | |
| CN116698075B (en) | Road network data processing method and device, electronic equipment and storage medium | |
| CN111613052B (en) | Traffic condition determining method and device, electronic equipment and storage medium | |
| CN113706857A (en) | Method, device and equipment for determining road trafficability and storage medium | |
| CN114419877B (en) | Vehicle track prediction data processing method and device based on road characteristics | |
| CN112686457B (en) | Route arrival time estimation method and device, electronic equipment and storage medium | |
| CN113850297B (en) | Road data monitoring method and device, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200331 |