A parallel prediction method for preceding vehicle speed based on deep belief and gated recurrent unit

Abstract

The present application belongs to the field of intelligent transportation and machine learning, and specifically relates to a preceding vehicle speed parallel prediction method based on deep belief and gated recurrent unit. Step one: collect preceding vehicle speed and acceleration to construct original time series data set, set rolling window segmentation and normalization, and use the same as training and verification data; step two: construct and train deep belief network to perform multi-layer feature extraction on the data; step three: input the extracted high-order features into the gated recurrent unit after splicing to learn time series dependence and realize speed prediction; step four: design a multi-condition fusion event triggering mechanism, and when the data reaches the triggering threshold, start offline parameter learning and synchronously update the online module, and the online prediction is performed according to the updated model to realize real-time speed prediction. Through deep feature extraction, time series modeling and event-driven updating mechanism, the present application realizes high-precision online speed prediction, improves the adaptability to dynamic scenes, and is suitable for advanced auxiliary driving systems, vehicle networking and the like.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation and machine learning, specifically a parallel prediction method for the speed of the vehicle ahead based on deep belief and gated recurrent units. Background Technology

[0002] In autonomous driving systems, accurate and rapid prediction of the speed of the vehicle ahead not only helps the controlled vehicle plan its route in advance and optimize energy consumption, but also avoids potential collision risks in complex traffic scenarios, while improving the overall efficiency of traffic flow. While the prediction of the speed of the vehicle ahead has promising applications, this technology also faces many challenges. As intelligent driving scenarios become increasingly complex, prediction models struggle to capture deep structural features in the data when dealing with complex nonlinear relationships or time-series problems, or they may encounter issues such as vanishing gradients and dimensionality explosion. Furthermore, autonomous driving systems need to perform real-time vehicle speed prediction under limited computing resources, requiring models to possess not only high accuracy but also rapid response capabilities. Summary of the Invention

[0003] This invention provides a parallel prediction method for the speed of vehicles ahead based on deep belief networks and gated recurrent units. The method proposes a hybrid online and offline parallel architecture based on deep belief networks and gated neural units. An offline module performs fine-tuning on historical data, while an online module performs real-time prediction of the speed of vehicles ahead. A conditional triggering mechanism is designed to update the parameters of the online module. This architecture effectively improves the adaptability of the prediction model while ensuring high accuracy, and also meets real-time requirements, making it suitable for engineering applications. Attached Figure Description

[0004] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0005] Figure 1 This is a schematic diagram of the overall structure of the present invention;

[0006] Figure 2 This is a schematic diagram of the offline learning structure for steps two and three in this invention;

[0007] Figure 3 A comparison chart of the prediction results of the three methods;

[0008] Figure 4 A comparison chart of the mean absolute error and root mean square error of the three methods;

[0009] Figure 5A comparison chart of the computation time for the three methods; Detailed Implementation

[0010] The parallel prediction method for the speed of the vehicle ahead based on deep belief and gated recurrent units includes the following steps:

[0011] Step 1: Collect the speed and acceleration of the vehicle in front to construct the original dataset, set a rolling window to segment the time series and perform normalization preprocessing, which will serve as the training and validation dataset for the prediction model.

[0012] Step 2: Construct and train a deep belief network (DBN) with a convolutional restricted Boltzmann machine to extract multi-level features from the normalized traffic data and obtain high-order semantic feature representations.

[0013] Step 3: The extracted high-order features are concatenated and input into a gated recurrent unit (GRU) to learn temporal dependencies and achieve velocity prediction;

[0014] Step 4: Design a multi-condition fusion event triggering mechanism. When the data reaches the trigger threshold, the parameter learning module is started and the online module parameters are updated asynchronously. The online prediction is then used to make real-time speed predictions.

[0015] The specific method for step one is as follows:

[0016] Historical data of the preceding vehicle, including speed and acceleration, is obtained through network connectivity and sensors. The scrolling window length is set to n, and the historical preceding vehicle speed v at time t is segmented by the scrolling window. h and acceleration a h The data sample is represented as follows:

[0017] (1)

[0018] Where T represents the sampling time interval. After selecting the dataset, this invention uses the maximum-minimum method to normalize the data, ensuring that the scale of each feature is consistent, as expressed in the following formula:

[0019] (2)

[0020] Among them, X max X represents the maximum value of the input data. min This represents the minimum value of the input data.

[0021] The specific method for step two is as follows:

[0022] To address the nonlinearity of vehicle speed and multi-factor coupling, this invention designs a feature extraction method based on deep belief networks. The core of the deep belief network is a restricted Boltzmann machine (RBM). To reduce computational complexity and improve the ability to express temporal local nonlinearity and multi-source coupling, this invention designs a convolutional restricted Boltzmann machine in each RBM layer, as follows:

[0023] (3)

[0024] In the formula, The input data to the network is a two-dimensional matrix with dimensions (2, T), where T is the sampling time interval. This represents the entire hidden layer information matrix. t and k are the time index and feature index, respectively, b is the hidden layer bias, and h is the neuron state of the hidden layer. It is the bias of the c-th channel (c=1 is the bias of the velocity channel, c=2 is the bias of the acceleration channel). This represents an element in the input data (c=1 is a velocity element, c=2 is an acceleration element). Let represent the local matching degree of the k-th filter at time point t of the input sequence, where It is a weight matrix. This represents the cross-correlation operation. The learning process of the visible and hidden layers is represented as follows:

[0025] (4)

[0026] in, It's the sigmoid activation function. The training objective is to optimize the network parameters by maximizing the log-likelihood of the data. The log-likelihood function of an RBM is expressed as:

[0027] (5)

[0028] in, This is the assignment function, used to ensure the normalization of the probability distribution. To maximize the log-likelihood, we need to calculate its value with respect to the parameters. The gradient is used for parameter updates via contrastive divergence, and the overall expression is as follows:

[0029] (6)

[0030] and These represent the expected value calculated from the data and the expected value calculated after one model reconstruction, respectively. Approximate values ​​of the parameter gradients are calculated using a divergence algorithm, followed by updating the model parameters using gradient descent rules, allowing the DBN to gradually learn the probability distribution of the input data. The hidden layer output after training the first RBM layer is used as the visible layer input for the second RBM layer, and the pre-trained parameters are recursively applied layer by layer. After completing the pre-training of multiple RBM layers, the high-dimensional feature representation of the DBN network output is denoted as X. DBN This feature contains deep abstract information about the input data. After training, the DBN network waits to receive data from the next time window before starting a new learning cycle. Each time window outputs a response X. DBN The trained feature X DBN The data is then fed into step three to achieve modeling and prediction of the time series data.

[0031] The specific method for step three is as follows:

[0032] After extracting high-order features from a deep belief network and concatenating the data, the data is fed into a GRU network to learn the temporal dependencies of these features. The GRU network is a simple, low-parameter, and computationally efficient recurrent neural network that effectively captures long-term dependencies in sequences and avoids the vanishing and exploding gradient problems common in traditional recurrent neural networks (RNNs). It excels at capturing dependencies in long-term sequences. Compared to methods such as Long Short-Term Memory (LSTM) networks and transformer models, the lightweight structure of the GRU network effectively reduces computational cost, thereby improving the real-time performance of the system.

[0033] The specific GRU network design is as follows:

[0034] (7)

[0035] Among them, z t To update the gate, r t It's a door reset. This represents the system's output state. A fully connected layer is used to map the final state to the predicted value, as shown below:

[0036] (8)

[0037] in, and It is the weight of the fully connected layer. It is the hidden state from the previous moment.

[0038] These are the weight parameters that need to be trained, and Ä represents element-wise multiplication. The goal of the GRU network is to predict, for example, the speed of the car in front, from time-series data. Therefore, the entire process can be optimized through end-to-end training, with the goal of minimizing the following loss function:

[0039] (9)

[0040] Where n is the length of the time window, V t It is the actual vehicle speed collected at time t. This represents the predicted vehicle speed of the GRU network. After the GRU network learns, this invention employs a self-attention mechanism to generate the final predicted value of the vehicle ahead. The attention mechanism, through weight allocation, assigns sufficient attention to key information, highlighting the influence of important information, thereby improving the accuracy of the neural network model. The attention mechanism is represented as follows:

[0041] (10)

[0042] Where e t The attention weights at time t are represented by the output vectors from the GRU network layer. Let u be the weight coefficient matrix. e is the bias parameter. j This is the attention score, typically obtained by calculating the relationship between the input and the query. α t The attention weights are obtained by normalizing the attention scores of each input; this invention uses the softmax function. Using the above method, the final predicted speed of the vehicle ahead at time t can be obtained. .

[0043] Steps two and three are the overall structure. Figure 1 The offline learning module has the following structure: Figure 2 As shown.

[0044] The specific method for step four is as follows:

[0045] In the process of predicting the speed of the vehicle in front, in order to solve the problem of training efficiency and adaptive capability of deep learning models under large-scale data and multiple parameters, this invention designs a parameter parallel optimization mode.

[0046] Based on four single parameters—acceleration of the vehicle ahead, rate of change of acceleration, traffic density, and road grade—a multi-condition fusion trigger condition is designed. The single-parameter acceleration trigger condition is expressed as follows:

[0047] (11)

[0048] Wherein, the acceleration threshold a thIt is used to characterize the degree of change in the longitudinal motion state of a vehicle, and its value is set according to the vehicle dynamics characteristics, preferably 0.5 to 2.0 m / s².

[0049] The trigger condition for the single-parameter rate of change of acceleration is expressed as:

[0050] (12)

[0051] Wherein, the acceleration change rate threshold j th The value used to characterize the drastic change in vehicle acceleration is set according to the requirements of driving smoothness, and is preferably 0.5 to 2.5 m / s³.

[0052] The trigger condition for a single-parameter traffic density change rate is expressed as follows:

[0053] (13)

[0054] Among them, the traffic density threshold ρ th The normalized traffic density threshold is used to characterize changes in traffic flow state. Its value is set according to the characteristics of traffic flow density distribution, and is preferably 0.2 to 0.5.

[0055] The single-parameter road grade trigger condition is expressed as follows:

[0056] (14)

[0057] Among them, road grade R perv Represented by discrete variables, it is triggered when the road grade at the previous moment is inconsistent with the road grade at the current moment;

[0058] The above thresholds can be adaptively adjusted according to different vehicle types, traffic environments and data sampling frequencies, and can be optimized and determined through historical data statistical analysis or model training processes.

[0059] By fusing the above single-parameter triggers, a multi-condition fusion triggering strategy is designed, as follows:

[0060] (15)

[0061] in, These represent the weight coefficients of the single triggering conditions in Formulas 11-14, respectively. The weight represents a dynamic threshold, which is designed in this invention as... This design allows for real-time adjustment of the trigger threshold, enabling a lower trigger threshold in high-traffic-density scenarios and a conservative triggering strategy in low-density scenarios, thereby balancing the sensitivity of model updates with computational efficiency. Wherein, ρ normIt is a normalized traffic density, obtained based on the connected environment. Through a multi-condition fusion triggering mechanism, it fully considers changes in the speed of the vehicle ahead, traffic density, and road type, making it more adaptable to dynamic traffic scenarios.

[0062] This invention employs an asynchronous update rule, with an online prediction module handling real-time prediction requests and using the current stable parameters. Once the trigger mechanism is activated, the background executes an incremental update mechanism to generate new parameters. The system switches offline-obtained parameters to the online prediction module via atomic pointer swapping, a process that takes very little time, approximately 1µs. In practical applications, a hardware FPGA environment can be used. This further improves the system's real-time performance, achieving <10ms, which meets the real-time requirements of autonomous driving control closed loops.

[0063] In summary, this invention first designs a scrolling window to segment and normalize the sensor-acquired data. Each preprocessed window of data is then input into a deep belief network, where different levels of higher-order representations are learned layer by layer. These higher-order representations are then concatenated according to feature dimensions and input into a GRU network to learn the temporal dependencies of these features. During the learning process, a parallel parameter optimization structure is designed. When multiple event triggering conditions are met, the prediction model parameters are updated in real time through pointer swapping, thereby improving the adaptive capability of the prediction method. A comparison of test results is attached. Figure 3 and attached Figure 4 As shown.

[0064] As can be seen from the comparison chart, compared to BPNN and LSTM predictions, the method of this invention can more smoothly and accurately follow the actual vehicle speed, while the other two methods produce larger fluctuations, especially the BPNN network, which generates a large error when performing multi-step predictions. (Appendix) Figure 4 The comparison of mean absolute error and root mean square error under test conditions shows that the method of the present invention not only achieves higher overall prediction accuracy, but also effectively avoids large prediction errors, thereby obtaining more stable prediction results that are closer to the actual vehicle speed trajectory.

[0065] Regarding computation time, the comparison results of the computation time of the three prediction algorithms are shown in the appendix. Figure 5 As shown, the maximum time consumed during the rolling update and prediction processes are 0.58 seconds, 0.14 seconds, and 0.18 seconds, respectively. However, the average time consumed by the three algorithms is approximately 0.52 seconds, 0.12 seconds, and 0.08 seconds, respectively. In the initial setup phase of the test, the data sampling interval of the vehicle sensor was set to 0.5 seconds. Based on the time consumption data of the prediction algorithm, the method of this invention can meet the real-time sampling requirements of vehicle applications.

Claims

1. A parallel prediction method for the speed of the preceding vehicle based on deep belief and gated recurrent units, characterized in that, Includes the following steps: Step 1: Acquire historical speed and acceleration data of the vehicle in front through vehicle-mounted sensors, segment the time series using a rolling time window of length n, construct time series samples from speed and acceleration data at multiple consecutive moments, and preprocess the sample data using a normalization method to obtain the input dataset for model training and prediction. The input data is a two-dimensional time series matrix containing two channels: speed and acceleration. Step 2: Construct a Deep Belief Network (DBN) consisting of stacked Restricted Boltzmann Machines (RBMs). Unsupervised, layer-by-layer pre-training is used to learn features from the normalized input data. The contrastive divergence algorithm is employed in each layer of the RBM to iteratively update the network parameters, approximating the probability distribution features of the input data. The output of the previous hidden layer is used as the input to the next layer for recursive training. After pre-training, a high-dimensional feature vector X is output. DBN And then sent to step three; wherein, the input data is a two-dimensional matrix of size (2, T), corresponding to the two channels of the vehicle speed and acceleration respectively; the restricted Boltzmann machine includes a visible layer, a hidden layer and corresponding weight parameters and bias parameters, and feature extraction is achieved by probabilistic modeling of the input data; Step 3: Input the high-order features output by the deep belief network into the gated recurrent unit network, use its gating structure to perform time-series modeling of the feature sequence, and output the speed prediction results of the vehicle ahead at future moments; Step 4: Construct a parallel prediction architecture consisting of an offline learning module and an online prediction module. The online prediction module predicts the speed of the vehicle in front based on real-time collected vehicle data, and stores the real-time data in a data cache module. An event triggering mechanism based on multi-condition fusion is designed. This event triggering mechanism constructs a multi-condition fusion trigger function based on the vehicle's acceleration, rate of change of acceleration, traffic density, and road type. By weighting and fusing each triggering condition and comparing it with a preset triggering threshold, when the triggering condition is met, the offline learning module is activated to update the model parameters. The updated parameters are then asynchronously switched to the online prediction module to achieve real-time prediction of the vehicle's speed.

2. The parallel prediction method for the speed of the vehicle ahead based on deep belief and gated recurrent units according to claim 1, characterized in that, The process described in step one involves collecting historical speeds and accelerations of vehicles ahead to construct the original dataset, setting a rolling window to segment the time series, and performing normalization preprocessing. This dataset serves as the training and validation dataset for the prediction model. The process is as follows: Historical data of the vehicle ahead, including speed and acceleration, is obtained through network connectivity and sensors. The scrolling window length is set to... n After being split by a scrolling window t The speed of history in the moment v h and acceleration a h Data samples X t It is expressed as follows: (1) in, T This indicates the sampling time interval. After selecting the dataset, this invention uses the maximum-minimum method to normalize the data, ensuring that the scale of each feature is consistent. The formula is as follows: (2) in, X max The maximum value of the input data. X min This represents the minimum value of the input data.

3. The parallel prediction method for the speed of the preceding vehicle based on deep belief and gated recurrent units according to claim 1, characterized in that, In step two, a deep belief network (DBN) with a convolutional restricted Boltzmann machine is constructed and trained to perform multi-level feature extraction on the normalized traffic data to obtain high-order semantic feature representations. The process is as follows: To address the nonlinearity of vehicle speed and multi-factor coupling, this invention designs a feature extraction method based on a deep belief network, where the core of the deep belief network is a restricted Boltzmann machine (RBM). To reduce computational complexity and improve the ability to express temporal local nonlinearity and multi-source coupling, a convolutional restricted Boltzmann machine is designed in each RBM layer, as shown below: (3) In the formula, The input data for the network is a two-dimensional matrix with dimensions (2, ...). T ); T It is the sampling time interval; This represents the entire hidden layer information matrix; t and k These are the time index and the feature index, respectively. b It is the bias of the hidden layer; h It refers to the neuronal state in the hidden layer; It is the first c The bias of each channel ( c =1 is the offset of the velocity channel. c =2 is the offset of the acceleration channel; This represents the elements in the input data ( c =1 is the velocity element. c =2 is an acceleration element); Indicates the first k Each filter at time point of the input sequence t The local matching degree at, where, Indicates cross-correlation operation. This is the weight matrix; the learning process of the visible and hidden layers is represented as follows: (4) in, It uses the sigmoid activation function; the training objective is to optimize the network parameters by maximizing the log-likelihood of the data. The log-likelihood function of RBM is expressed as: (5) in, It is the assignment function, used to ensure the normalization of the probability distribution; to maximize the log-likelihood, we need to calculate its value with respect to the parameters. The gradient is used for parameter updates via contrastive divergence, and the overall expression is as follows: (6) and These are the expected values ​​calculated from the data and the expected values ​​calculated after one model reconstruction, respectively. Approximate values ​​of the parameter gradients are calculated using a divergence algorithm. Then, the model parameters are updated using gradient descent, allowing the DBN to gradually learn the probability distribution of the input data. The hidden layer output after the first RBM training is used as the visible layer input for the second RBM. The pre-training parameters are iterated layer by layer. After completing the pre-training of multiple RBMs, the high-dimensional feature representation of the DBN network output is denoted as... X DBN and the trained features X DBN Send it to step three.

4. The parallel prediction method for the speed of the preceding vehicle based on deep belief and gated recurrent units according to claim 1, characterized in that, Step 3 involves concatenating the extracted high-order features and inputting them into a gated recurrent unit (GRU) to learn temporal dependencies and achieve velocity prediction. The specific process is as follows: After extracting high-order features from a deep belief network and concatenating the data, the data is fed into a GRU network to learn the temporal dependencies of these features. Gated recurrent unit networks are a type of recurrent neural network with a simple structure, few parameters, and high computational efficiency. They can effectively capture long-term dependencies in sequences and effectively avoid the gradient vanishing and exploding problems in traditional recurrent neural networks (RNNs). They are good at capturing dependencies in long-term sequences. Compared with the methods of Long Short-Term Memory (LSTM) networks and transformer models, the lightweight structure of GRU networks effectively reduces the amount of computation, thereby improving the real-time performance of the system. The specific GRU network design is as follows: (7) in, X DBN These are high-dimensional features output by deep belief networks; z t To update the door; r t It's a door reset; This represents the system's output state; a fully connected layer is used to map the final state to the predicted value, as shown below: (8) in, and These are the weights of the fully connected layer; It is the hidden state from the previous moment; and These are the weight parameters that need to be trained; Ä represents element-wise multiplication; The goal of a GRU network is to predict, for example, the speed of the car in front, using time-series data. Therefore, the entire process can be optimized through end-to-end training, with the objective of minimizing the following loss function: (9) in, n It is the length of the time window; V t yes t Real-time vehicle speed data; This represents the predicted vehicle speed of the GRU network. After the GRU network learns, this invention uses a self-attention mechanism to generate the final predicted value of the vehicle ahead. The attention mechanism allocates sufficient attention to key information through weight distribution, highlighting the influence of important information, thereby improving the accuracy of the neural network model. The attention mechanism is represented as follows: (10) in e t Indicates the first t The attention weights at time step are determined by the output vector from the GRU network layer. Decide; u This is the weight coefficient matrix; These are bias parameters; e j It is the attention score, obtained by calculating the relationship between the input and the query; α t The attention weights are obtained by normalizing the attention scores of each input; this invention uses the softmax function, and through the above method, the attention weights of the preceding vehicle at time [time value missing] can be obtained. t Final prediction speed .

5. The parallel prediction method for the speed of the preceding vehicle based on deep belief and gated recurrent units according to claim 1, characterized in that, Step four involves designing a multi-condition fusion event triggering mechanism. When the data reaches the trigger threshold, the parameter learning module is activated to update the parameters and asynchronously update the parameters of the online module. The online prediction then performs real-time speed prediction based on this. The specific process is as follows: Design a multi-condition fusion event triggering mechanism. When the data reaches the trigger threshold, start the parameter learning module and asynchronously update the online module parameters. The online prediction then performs real-time speed prediction based on this. In the process of predicting the speed of the vehicle in front, in order to solve the problem of training efficiency and adaptive capability of deep learning models under large-scale data and multiple parameters, this invention designs a parameter parallel optimization mode. Design multi-condition fusion triggering conditions based on four single parameters: acceleration of the vehicle ahead, rate of change of acceleration, traffic density, and road grade. The single-parameter acceleration trigger condition is expressed as: (11) Among them, acceleration threshold a th It is used to characterize the degree of change in the longitudinal motion state of a vehicle, and its value is set according to the vehicle dynamics characteristics, preferably 0.5 to 2.0 m / s². The trigger condition for the single-parameter rate of change of acceleration is expressed as: (12) Among them, the acceleration rate of change threshold j th The value used to characterize the drastic change in vehicle acceleration is set according to the requirements of driving smoothness, and is preferably 0.5 to 2.5 m / s³. The trigger condition for a single-parameter traffic density change rate is expressed as follows: (13) Among them, traffic density threshold ρ th The normalized traffic density threshold is used to characterize changes in traffic flow state. Its value is set according to the characteristics of traffic flow density distribution, and is preferably 0.2 to 0.

5. The single-parameter road grade trigger condition is expressed as follows: (14) Among them, road grade R perv Represented by discrete variables, it is triggered when the road grade at the previous moment is inconsistent with the road grade at the current moment; The above thresholds can be adaptively adjusted according to different vehicle types, traffic environments and data sampling frequencies, and can be optimized and determined through historical data statistical analysis or model training process; By fusing the above single-parameter triggers, a multi-condition fusion triggering strategy is designed, as follows: (15) in, These represent the weight coefficients of the single triggering conditions in Formulas 11-14, respectively. The weight represents a dynamic threshold, which is designed in this invention as... This design allows for real-time adjustment of the trigger threshold, enabling a lower trigger threshold in high-traffic-density scenarios and a conservative triggering strategy in low-density scenarios, thereby balancing the sensitivity of model updates with computational efficiency. ρ norm It is a normalized traffic density, obtained based on the connected environment. Through a multi-condition fusion triggering mechanism, it fully considers changes in the speed of the vehicle ahead, traffic density, and road type, making it more adaptable to dynamic traffic scenarios. Simultaneously, asynchronous update rules are designed, and an online prediction module handles real-time prediction requests. Currently stable parameters are used. Once the trigger mechanism is activated, the background executes an incremental update mechanism to generate new parameters. The parameters obtained offline are switched to the online prediction module by atomic pointer swapping. This process takes very little time, less than 1µs. In actual application, a hardware FPGA environment can be used to further improve the real-time performance of the system, which can basically reach <10ms, thus meeting the real-time requirements of the autonomous driving control closed loop.

Images