Vehicle speed prediction methods, devices, computer equipment, storage media and software products

Abstract

The application relates to a vehicle speed prediction method and device, computer equipment, a storage medium and a computer program product. In the case that a target vehicle is in a target working condition, vehicle speed data in a first time period before a current time is acquired, a combined neural network model is used for feature extraction of the vehicle speed data, high-dimensional complex features are obtained, the high-dimensional complex features are processed by dimension reduction through the combined neural network model, target vehicle speed features are obtained, predicted vehicle speeds in a second time period after the current time are acquired according to the target vehicle speed features, future short-time vehicle speeds can be predicted according to real-time vehicle speed data of the target vehicle, real-time prediction of the vehicle speed is realized, and the vehicle speed prediction accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of automotive technology, and in particular to a vehicle speed prediction method, device, computer equipment, storage medium, and computer program product. Background Technology

[0002] In the field of driver assistance, accurate prediction of future vehicle speed plays a crucial role in vehicle path and collision warning functions. Since vehicle speed itself is highly time-varying, data-driven methods are usually used to predict future vehicle speed in order to ensure the real-time nature of road information prediction.

[0003] Traditional methods use backpropagation (BP) neural networks to learn historical vehicle speed data in order to predict short-term vehicle speeds in the future. However, due to the limitations of the BP neural network model, it is prone to getting trapped in local extrema, resulting in a large deviation in the predicted vehicle speed. Summary of the Invention

[0004] Therefore, it is necessary to provide a vehicle speed prediction method, device, computer equipment, computer-readable storage medium, and computer program product that can improve the accuracy of vehicle speed prediction in response to the above-mentioned technical problems.

[0005] Firstly, this application provides a vehicle speed prediction method, including:

[0006] Given that the target vehicle is under the target operating condition, acquire the vehicle speed data for the first time period prior to the current moment; the vehicle speed data is sequential data.

[0007] The vehicle speed data is feature extracted by a pre-trained ensemble neural network model to obtain high-dimensional complex features. The high-dimensional complex features are then reduced in dimensionality by the ensemble neural network model to obtain the target vehicle speed features. The ensemble neural network model includes at least one complex encoder and at least one complex decoder. The output data of the complex encoder is input to the corresponding complex decoder through a jump connection.

[0008] Based on the target vehicle speed characteristics, obtain the predicted vehicle speed for the second time period after the current moment.

[0009] In one embodiment, the training process of the combined neural network model includes:

[0010] The vehicle speed data is divided into training set data and test set data;

[0011] Determine the initialization parameters and construct an initial combination model based on the initialization parameters;

[0012] The initial combined model is trained based on the training set data, and the trained initial combined model is tested based on the test set data.

[0013] In one embodiment, the combined neural network model includes intermediate layers and an output layer; the steps of training an initial combined model based on training set data and testing the trained initial combined model based on test set data include:

[0014] The initial combined model is processed using the training set data to obtain the output layer error corresponding to the network output layer, and the intermediate layer error corresponding to the network intermediate layer is obtained based on the output layer error.

[0015] Adjust the weights and thresholds of the intermediate layers of the network based on the intermediate layer error, and further adjust the weights and thresholds of the intermediate layers of the network based on the adjusted intermediate layer and output layer errors.

[0016] Update the initialization parameters, return to the steps of building the initial combined model based on the initialization parameters, and continue execution until the initial combined model meets the learning requirements.

[0017] If the initial combined model meets the learning requirements, the adjusted initial combined model is processed using test set data to obtain test results;

[0018] If the test result does not meet the error requirement and the target number of training iterations has not been reached, update the initialization parameters, return to the step of building the initial combined model based on the initialization parameters, and continue execution until the test result meets the error requirement or the target number of training iterations is reached.

[0019] In one embodiment, the intermediate layer of the network includes an encoder layer and a decoder layer; the encoder layer includes a complex encoder, and the decoder layer includes a complex decoder; the step of extracting features from the vehicle speed data using a pre-trained ensemble neural network model to obtain high-dimensional complex features includes:

[0020] The vehicle speed data is convolved sequentially through N encoder layers to obtain high-dimensional complex features; the input data of the first encoder layer is the vehicle speed data.

[0021] Where N is an even number, for the i-th encoder layer among the 2nd to Nth encoder layers, the input data of the i-th encoder layer is the output data of the (i-1)-th encoder layer.

[0022] In one embodiment, the complex encoder includes a real part convolution filter and an imaginary part convolution filter; the step of sequentially convolving the vehicle speed data through N encoder layers to obtain high-dimensional complex features includes:

[0023] For each encoder layer and its corresponding complex encoder, the vehicle speed data is processed by real part convolution filter and imaginary part convolution filter respectively to obtain the corresponding real part convolution features and imaginary part convolution features.

[0024] The real and imaginary convolution features are normalized to obtain the high-dimensional complex features output by the complex encoder.

[0025] In one embodiment, the combined neural network model further includes a separation module; the step of reducing the dimensionality of high-dimensional complex features using the combined neural network model to obtain the target vehicle speed features includes:

[0026] The high-dimensional complex features output from the Nth encoder layer are separated by a separation module to obtain the separated features.

[0027] The target vehicle speed feature is obtained by sequentially performing dimensionality reduction processing on the separated features through M decoder layers; the input data of the first decoder layer is the separated feature.

[0028] Where M and N have the same value, for the m-th decoder layer from the 2nd decoder layer to the M-th decoder layer, the input data of the m-th decoder layer is the output data of the (m-1)-th decoder layer and the output data of the (M+1-m)-th encoder layer.

[0029] In one embodiment, the separation module includes a real part separation module and an imaginary part separation module; the high-dimensional complex features include high-dimensional real part features and high-dimensional imaginary part features; the step of separating the high-dimensional complex features output by the Nth encoder layer through the separation module to obtain the separated features includes:

[0030] By using the real part separation module and the imaginary part separation module, complex number operations are performed on the high-dimensional real part features and the high-dimensional imaginary part features respectively to obtain the separated features.

[0031] Secondly, this application also provides a vehicle speed prediction device, comprising:

[0032] The acquisition module is used to acquire vehicle speed data for the first time period before the current moment, when the target vehicle is under the target operating condition; the vehicle speed data is sequential data.

[0033] The extraction module is used to extract features from vehicle speed data through a pre-trained ensemble neural network model to obtain high-dimensional complex features, and then to perform dimensionality reduction on the high-dimensional complex features through the ensemble neural network model to obtain the target vehicle speed features. The ensemble neural network model includes at least one complex encoder and at least one complex decoder. The output data of the complex encoder is input to the corresponding complex decoder through a jump connection.

[0034] The prediction module is used to obtain the predicted vehicle speed in the second time period after the current moment based on the target vehicle speed characteristics.

[0035] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method steps of any one of the first aspects.

[0036] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method steps of any one of the first aspects.

[0037] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the method steps of any one of the first aspects.

[0038] The aforementioned vehicle speed prediction method, device, computer equipment, storage medium, and computer program product acquire vehicle speed data for a first time period before the current moment when the target vehicle is in the target operating condition. They then extract features from the vehicle speed data using a pre-trained combined neural network model to obtain high-dimensional complex features. Furthermore, they perform dimensionality reduction processing on these high-dimensional complex features using the combined neural network model to obtain target vehicle speed features. Based on these target vehicle speed features, they obtain the predicted vehicle speed for a second time period after the current moment. This allows for the prediction of future short-term vehicle speeds based on real-time vehicle speed data, achieving real-time vehicle speed prediction and thus improving the accuracy of vehicle speed prediction. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is an application environment diagram of the vehicle speed prediction method in one embodiment;

[0041] Figure 2 This is a flowchart illustrating a vehicle speed prediction method in one embodiment;

[0042] Figure 3 This is a schematic diagram of the structure of a combined neural network model in one embodiment;

[0043] Figure 4 This is a flowchart illustrating the training process of a combined neural network model in one embodiment;

[0044] Figure 5 This is a flowchart illustrating the steps for obtaining high-dimensional complex features in one embodiment;

[0045] Figure 6 This is a flowchart illustrating the steps for obtaining high-dimensional complex features in one embodiment;

[0046] Figure 7 This is a flowchart illustrating the steps for obtaining separation features in one embodiment;

[0047] Figure 8 This is a flowchart illustrating a vehicle speed prediction method in one embodiment;

[0048] Figure 9 This is a structural block diagram of a vehicle speed prediction device in one embodiment;

[0049] Figure 10 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0051] The vehicle speed prediction method provided in this application embodiment can be applied to, for example, Figure 1 In the application environment shown, terminal 102 communicates with vehicle speed sensor 104 via a network. Terminal 102 is used to acquire vehicle speed data for a first time period prior to the current moment when the target vehicle is in the target operating condition. It then extracts features from the vehicle speed data using a pre-trained combined neural network model to obtain high-dimensional complex features. These high-dimensional complex features are then reduced in dimensionality using the combined neural network model to obtain the target vehicle speed features. Based on these target vehicle speed features, the predicted vehicle speed for a second time period after the current moment is obtained. Terminal 102 can be an in-vehicle terminal. Vehicle speed sensor 104 can be deployed on the target vehicle.

[0052] In one exemplary embodiment, such as Figure 2 As shown, a vehicle speed prediction method is provided, which is then applied to... Figure 1 Taking terminal 102 as an example, the explanation includes the following steps 202 to 206. Wherein:

[0053] S202: When the target vehicle is in the target operating condition, obtain the vehicle speed data for the first time period before the current moment; the vehicle speed data is sequential data.

[0054] The target operating conditions can be highways, trunk roads, mines, ports, and other operating scenarios. In practical applications, due to the influence of road conditions, altitude, geographical location, and vehicle usage scenarios on vehicle speed prediction, vehicle speed can only be predicted under the same operating conditions. For example, if the obtained vehicle speed data is the data of the target vehicle under highway operating conditions, then the predicted vehicle speed is the short-term future vehicle speed of the target vehicle under highway operating conditions.

[0055] The first time period is typically 10 minutes. When acquiring vehicle speed data, due to the high-speed time-varying nature of vehicle speed, the time factor needs to be considered to ensure prediction accuracy. Vehicle speed data is collected at certain time intervals, resulting in sequential data. Specifically, vehicle speed data can be represented as: V = (v1, v2, ..., vt), where the collection interval Δt is 200ms, and vt is the vehicle speed data collected in the t-th time interval. The total collection time is 10 minutes, yielding 3000 vehicle speed data points, forming the sequential data. In practical applications, due to the volatility and complexity of vehicle speed data, outliers are prone to appear. Therefore, before inputting the vehicle speed data into the model, outliers need to be adjusted and removed, and normal data and time-series data need to be matched one by one. After cleaning and normalization, the resulting dataset serves as the input data for the combined neural network model.

[0056] S204: The vehicle speed data is feature extracted by a pre-trained ensemble neural network model to obtain high-dimensional complex features, and the high-dimensional complex features are then reduced in dimensionality by the ensemble neural network model to obtain the target vehicle speed features. The ensemble neural network model includes at least one complex encoder and at least one complex decoder. The output data of the complex encoder is input to the corresponding complex decoder through a jump connection.

[0057] Among them, the combined neural network model is the BP-CRN network model, which includes the BP neural network model and the CRN neural network model. The BP neural network model is a classic feedforward neural network, which is trained and optimized by the backpropagation algorithm. It consists of multiple neuron layers. Each neuron processes the input through weights and activation functions and passes the result to the next layer. The backpropagation algorithm updates the weights of neurons by calculating the error gradient to minimize the difference between the predicted output and the actual output.

[0058] The CRN neural network model is a convolutional recurrent neural network that combines the advantages of convolutional neural networks (CNNs) and recurrent neural networks (RNNs), making it suitable for processing sequential data. A CRN network mainly includes an input layer, convolutional layers, recurrent layers, and pooling layers. The input layer receives sequential data, such as image or speech sequences. The convolutional layers use convolution operations to extract local features and spatial structure information from the sequence. The recurrent layers use RNN units to iteratively process the feature sequence, learning dependencies between features. The pooling layers perform downsampling operations to reduce the feature map size and improve computational efficiency. The BP-CRN combined neural network model combines BP and CRN network models, using the BP algorithm for training and optimization while leveraging the structure of the CRN network model to improve the network's training performance and generalization ability.

[0059] In traditional methods, RNN units typically employ Long Short-Term Memory (LSTM) networks. However, LSTM involves a large number of gate functions, resulting in high memory consumption. Therefore, this embodiment replaces LSTM with Complex GRU. GRU is a variant of RNN used to process sequential data, such as time series and natural language processing tasks. Unlike standard RNNs, GRU uses a complex separation module to simplify the structure of standard RNN units and solves the vanishing and exploding gradient problems. Complex feature extraction using GRU can more finely represent the two-dimensional complex spectrum, expressing the phase information corresponding to the real value of vehicle speed.

[0060] Specifically, the structural diagram of the BP-CRN network model is as follows: Figure 3 As shown, the BP-CRN network model includes at least one complex encoder and at least one complex decoder. The output data of the complex encoder is input to the corresponding complex decoder through a jump connection. The complex encoder has a real part convolutional layer and an imaginary part convolutional layer. The complex convolution can be regarded as two different real-valued convolutions with shared parameters. The real part and imaginary part of the input complex spectrum pass through the two parts of the convolutional layer respectively, controlling the encoder to perform complex feature extraction. The result simulates complex number operations.

[0061] in, Figure 3The left two layers are schematic diagrams of complex encoders, the middle layer is a schematic diagram of splitting layers, and the right two layers are schematic diagrams of complex decoders corresponding to the complex encoders. The complex encoder mainly consists of stacked convolutional layers and batch normalization layers, used to extract high-dimensional features from the original input. The complex decoder mainly consists of transposed convolutional layers and batch normalization layers, used to restore the extracted high-dimensional features to the original input size. The corresponding complex encoder is connected to its mirror image complex decoder via jump connections, realizing the mapping from a low-resolution feature map to a complete feature map of the same size as the input. During forward and backward propagation, convolutional layers and corresponding deconvolutional layers transmit the response of each layer through jump connections. The high-dimensional features obtained from convolution and the low-dimensional features obtained from deconvolution are complementary, contributing to gradient flow.

[0062] The complex encoder consists of five two-dimensional convolutional (Conv2d) blocks that extract high-dimensional features from the input features to reduce the resolution. The complex decoder reconstructs the low-resolution features to the original size of the input. The encoder-decoder structure is designed symmetrically, with the Conv2d blocks in the complex encoder consisting of convolutional layers and the Conv2d blocks in the complex decoder consisting of deconvolutional layers.

[0063] S206: Based on the target vehicle speed characteristics, obtain the predicted vehicle speed for the second time period after the current moment.

[0064] In this process, the target vehicle speed features extracted by the combined neural network model are used to predict the short-term future speed of the target vehicle. Typically, the second time period is 3 minutes; excessively long prediction times can reduce the accuracy of the predicted speed. In practical applications, to ensure the model's prediction accuracy, a loss function is usually constructed to evaluate the model's prediction accuracy. Specifically, the loss function can use the root mean square error (RMSE) as the evaluation metric. By evaluating the learning time and prediction time of the combined neural network model, the model's prediction efficiency can be assessed. By comparing the model's predicted values ​​with the actual values, the model's prediction efficiency can be evaluated, and backpropagation updates the gradient values ​​to optimize the model's prediction performance, thereby ensuring the accuracy of the obtained predicted vehicle speed.

[0065] In the above-mentioned vehicle speed prediction method, the vehicle speed data for the first time period before the current moment is obtained when the target vehicle is in the target operating condition. The vehicle speed data is then used to extract features from the speed data through a pre-trained combined neural network model to obtain high-dimensional complex features. The high-dimensional complex features are then reduced in dimensionality by the combined neural network model to obtain the target vehicle speed features. Based on the target vehicle speed features, the predicted vehicle speed for the second time period after the current moment is obtained. This method can predict the short-term vehicle speed in the future based on the real-time vehicle speed data of the target vehicle, thereby achieving real-time vehicle speed prediction and improving the accuracy of vehicle speed prediction.

[0066] In an exemplary embodiment, the training process of the combined neural network model includes: dividing the vehicle speed data to obtain training set data and test set data; determining initialization parameters and constructing an initial combined model based on the initialization parameters; training the initial combined model based on the training set data and testing the trained initial combined model based on the test set data.

[0067] The initialization parameters refer to the initial settings of the combined neural network model. Specifically, the model setting parameters are shown in Table 1. The number of iterations refers to the number of training iterations of the combined neural network model. The initial learning rate is set to 0.001. When the loss function increases during each training process, the learning rate is reduced by 0.5. The batch-size refers to the number of training samples input to the model each time. The activation function is the PReLU function, and the input of the PReLU function can be negative. The cGRU layer is a conditional GRU layer. The GRU layer constitutes a separation module, which separates the high-dimensional complex features encoded by the model encoder layer. The separated features are used as the input data of the decoder layer. After dimensionality reduction by the decoder layer, the vehicle speed features are output.

[0068] Table 1 Model Setting Parameters

[0069]

[0070] The encoder and decoder layers each have 6 layers. Correspondingly, the number of channels in the six Conv2d blocks are set to 16, 32, 64, 128, 256, and 256, respectively. The convolution kernel size is 5*2, and the stride is 2*1. The output of each complex encoder is connected to the corresponding complex decoder input via a jump connection to ensure information flow transmission in the network. Therefore, the number of channels in the two-dimensional deconvolution modules of the six complex decoders in the network are set to 256, 256, 128, 64, 32, and 1, respectively. The deconvolution kernel size is 5*2, and the stride is 2×1. Specifically, the initial settings parameters for each convolution block and deconvolution block are shown in Table 2.

[0071] Table 2 Convolution Settings Parameters

[0072]

[0073] Furthermore, the training set data typically consists of the first 2400 vehicle speed data points out of 3000 acquired vehicle speed data points, while the test set data consists of the last 600 vehicle speed data points. The initial combined model is trained using the training set data, and the trained initial combined model is tested using the test set data. The training effect of the model is evaluated using a loss function to ensure the model accuracy.

[0074] In this embodiment, by dividing the vehicle speed data to obtain training set data and test set data, initialization parameters are determined, and an initial combined model is constructed based on the initialization parameters. The initial combined model is trained based on the training set data, and the trained initial combined model is tested based on the test set data. This can improve the model accuracy of the combined neural network model, thereby improving the accuracy of vehicle speed prediction.

[0075] In one exemplary embodiment, such as Figure 4 As shown, the ensemble neural network model includes intermediate layers and an output layer. The steps of training the initial ensemble model based on the training set data and testing the trained initial ensemble model based on the test set data include:

[0076] The initial combined model is processed using the training set data to obtain the output layer error corresponding to the network output layer, and the intermediate layer error corresponding to the network intermediate layer is obtained based on the output layer error.

[0077] Adjust the weights and thresholds of the intermediate layers of the network based on the intermediate layer error, and further adjust the weights and thresholds of the intermediate layers of the network based on the adjusted intermediate layer and output layer errors.

[0078] Update the initialization parameters, return the steps for building the initial combined model based on the initialization parameters, and continue execution until the initial combined model meets the learning requirements.

[0079] If the initial combined model meets the learning requirements, the adjusted initial combined model is processed using test set data to obtain test results.

[0080] If the test result does not meet the error requirement and the target number of training iterations has not been reached, update the initialization parameters, return to the step of building the initial combined model based on the initialization parameters, and continue execution until the test result meets the error requirement or the target number of training iterations is reached.

[0081] The intermediate layer of the network refers to the convolutional layer in the BP-CRN network model. The convolutional layer consists of multiple complex encoders and is used to extract spatial features from the input sequence. The network output layer is used to classify, regress, or generate output according to different task types. Here, the output data of the output layer is the predicted vehicle speed.

[0082] Specifically, the flowchart for training the initial combined model based on the training set data and testing the trained initial combined model based on the test set data is as follows: Figure 4As shown, at the start of training, the model weights first need to be initialized. The Xavier initialization method is used to ensure a relatively uniform distribution of output values ​​across all layers. During training, the weights and thresholds of the intermediate and output layers are continuously adjusted until the model has completed its learning, the model error meets requirements, or the maximum number of training iterations is reached. The depth of model learning can typically be determined by observing changes in training loss, validation metrics, test metrics, model complexity, and prediction results to prevent overfitting.

[0083] In this embodiment, by training the initial combined model based on training set data and testing the trained initial combined model based on test set data, the model accuracy of the combined neural network model can be improved, thereby improving the accuracy of vehicle speed prediction.

[0084] In one exemplary embodiment, such as Figure 5 As shown, the intermediate layers of the network include an encoder layer and a decoder layer; the encoder layer includes a complex encoder, and the decoder layer includes a complex decoder; the steps of extracting features from vehicle speed data using a pre-trained ensemble neural network model to obtain high-dimensional complex features include:

[0085] The vehicle speed data is convolved sequentially through N encoder layers to obtain high-dimensional complex features; the input data of the first encoder layer is the vehicle speed data; where N is an even number, and for the i-th encoder layer in the second to N encoder layers, the input data of the i-th encoder layer is the output data of the (i-1)-th encoder layer.

[0086] The data flow when obtaining high-dimensional complex features by sequentially performing convolution processing on vehicle speed data through N encoder layers is as follows: Figure 5 As shown, the input data of the first encoder layer is the vehicle speed data, and the input data of each subsequent encoder layer is the output data of the previous encoder layer. Furthermore, the output data of the complex encoder in each encoder layer is input to the complex decoder of the corresponding mirror layer through a jump connection.

[0087] In this embodiment, the vehicle speed data is convolved sequentially through N encoder layers to obtain high-dimensional complex features, which can improve the accuracy of feature extraction and thus improve the accuracy of vehicle speed prediction.

[0088] In one exemplary embodiment, such as Figure 6As shown, the complex encoder includes a real part convolution filter and an imaginary part convolution filter. The step of obtaining high-dimensional complex features by sequentially convolving the vehicle speed data through N encoder layers includes: for each encoder layer, extracting features from the vehicle speed data through the real part convolution filter and the imaginary part convolution filter respectively to obtain the corresponding real part convolution features and imaginary convolution features; and normalizing the real part convolution features and the imaginary convolution features to obtain the high-dimensional complex features output by the complex encoder.

[0089] The input vehicle speed data is divided into real and imaginary parts, which are respectively processed by real and imaginary convolutional filters for feature extraction. The output data of the real and imaginary convolutional filters enter the normalization layer. The normalization layer calculates the mean and variance of the input in each channel or batch, and standardizes or normalizes the input according to the mean and variance. The normalized data enters the real PReLu layer, where PReLu is the activation function, and the output is a high-dimensional complex feature.

[0090] In this embodiment, features are extracted from vehicle speed data using real part convolutional filters and imaginary part convolutional filters, and then normalized to obtain high-dimensional complex features output by the complex encoder. This ensures the accuracy of the high-dimensional complex features, thereby improving the accuracy of vehicle speed prediction.

[0091] In one exemplary embodiment, it remains as follows Figure 5 As shown, the combined neural network model also includes a separation module; the step of reducing the dimensionality of high-dimensional complex features to obtain the target vehicle speed features through the combined neural network model includes: separating the high-dimensional complex features output by the Nth encoder layer through the separation module to obtain the separated features; and sequentially reducing the dimensionality of the separated features through M decoder layers to obtain the target vehicle speed features; the input data of the first decoder layer is the separated features; where M and N have the same value, and for the mth decoder layer among the 2nd to Mth decoder layers, the input data of the mth decoder layer is the output data of the (m-1)th decoder layer and the output data of the (M+1-m)th encoder layer.

[0092] The output data of the Nth encoder layer is separated by the separation module and then input to the first decoder layer. The separation module includes a complex GRU module and a linear layer. The complex GRU module is used to perform complex number operations on the high-dimensional complex features to enhance the features, and the linear layer is used to perform linear transformations on the enhanced features to output the separated features.

[0093] Among them, it remains as Figure 5As shown, the input data of the first decoder layer is the separation feature, and the input data of each subsequent decoder layer is the output data of the previous decoder layer and the output data of the complex encoder mirroring the corresponding layer.

[0094] In this embodiment, the high-dimensional complex features output by the Nth encoder layer are separated by a separation module to obtain separated features. The separated features are then reduced in dimensionality by M decoder layers to obtain target vehicle speed features, enabling real-time prediction of vehicle speed based on these features.

[0095] In one exemplary embodiment, such as Figure 7 As shown, the separation module includes a real part separation module and an imaginary part separation module; the high-dimensional complex features include high-dimensional real part features and high-dimensional imaginary part features; the step of separating the high-dimensional complex features output by the Nth encoder layer through the separation module to obtain the separated features includes: performing complex number operations on the high-dimensional real part features and the high-dimensional imaginary part features respectively through the real part separation module and the imaginary part separation module to obtain the separated features.

[0096] Let the expressions for the real part convolution filter and the imaginary part convolution filter in the complex encoder be:

[0097] W = Wr r +jW i

[0098] Among them, W r W represents the real part of the convolution filter. i Represents the imaginary part convolution filter;

[0099] After convolving the vehicle speed data, the resulting convolutional feature F out Represented as:

[0100] F out =(Y r ×W r -Y i ×W i )+j(Y r ×W i -Y i ×W r )

[0101] Among them, Y r Y represents the real part of the vehicle speed data. i This represents the imaginary part of the vehicle speed data;

[0102] The output data of the first encoder layer is represented as follows:

[0103] F1 = PRELu(BN(F out ))

[0104] Where BN() represents the normalization function and PReLu() represents the PReLu activation function;

[0105] Furthermore, after feature extraction through 6 encoder layers, the high-dimensional complex features output by the last encoder layer are represented as follows:

[0106] F6 = F r6 +F i6

[0107] Among them, F r6 F represents the high-dimensional real part feature of the output of the sixth encoder layer. i6 This represents the high-dimensional imaginary part feature of the output of the sixth encoder layer.

[0108] Furthermore, the process of separating the features by performing complex number operations on the high-dimensional real and imaginary parts of the high-dimensional complex features using the separation module is represented as follows:

[0109] F rr =GRU r (F r6 );F ri =GRU i (F r6 );F ir =GRU r (F i6 );F ii =GRU i (F i6 )

[0110] The final separation features are represented as follows:

[0111] F^=(F rr -F ii )+j(F ri +F ir )

[0112] Among them, F rr F is the feature vector output by the real part separation module for high-dimensional real part features. ri F is the feature vector output by the imaginary part separation module for high-dimensional real features. ir F is the feature vector output by the real part separation module for the high-dimensional imaginary part features. ii The feature vector is the output of the imaginary part separation module for high-dimensional imaginary part features.

[0113] Specifically, such as Figure 7 As shown, by performing complex number operations on the obtained feature vectors, the final output separation feature F^ is obtained.

[0114] In this embodiment, the real part separation module and the imaginary part separation module perform complex number operations on the high-dimensional real part features and the high-dimensional imaginary part features respectively to obtain separated features. This can enhance the high-dimensional complex features output by the encoder layer, thereby improving the accuracy of vehicle speed prediction.

[0115] In one exemplary embodiment, such as Figure 8 As shown, a vehicle speed prediction method is provided, including the following steps:

[0116] When the target vehicle is in the target operating condition, obtain the vehicle speed data for the first time period before the current moment; the vehicle speed data is sequential data.

[0117] The vehicle speed data is divided into training set data and test set data; the initialization parameters are determined, and an initial combined model is constructed based on the initialization parameters; the initial combined model is processed through the training set data to obtain the output layer error corresponding to the network output layer, and the intermediate layer error corresponding to the network intermediate layer is obtained based on the output layer error;

[0118] Adjust the weights and thresholds of the intermediate layers of the network based on the intermediate layer error, and adjust the weights and thresholds of the intermediate layers of the network based on the adjusted intermediate layer and output layer errors; update the initialization parameters, return to the step of building the initial combined model based on the initialization parameters, and continue to execute until the initial combined model meets the learning requirements.

[0119] If the initial combined model meets the learning requirements, the adjusted initial combined model is processed using test set data to obtain test results;

[0120] If the test result does not meet the error requirement and the target number of training iterations has not been reached, update the initialization parameters, return to the step of building the initial combined model based on the initialization parameters, and continue execution until the test result meets the error requirement or the target number of training iterations is reached.

[0121] For each encoder layer and its corresponding complex encoder, features are extracted from the vehicle speed data using real part convolution filters and imaginary part convolution filters, respectively, to obtain the corresponding real part convolution features and imaginary part convolution features. The real part convolution features and imaginary part convolution features are then normalized to obtain the high-dimensional complex features output by the complex encoder.

[0122] The input data for the first encoder layer is vehicle speed data; N is an even number, and for the i-th encoder layer from the second to the Nth encoder layers, the input data for the i-th encoder layer is the output data for the (i-1)-th encoder layer.

[0123] By using a real part separation module and an imaginary part separation module, complex number operations are performed on the high-dimensional real part features and the high-dimensional imaginary part features respectively to obtain the separated features; the separated features are then subjected to dimensionality reduction processing by M decoder layers in sequence to obtain the target vehicle speed features; the input data of the first decoder layer is the separated features.

[0124] Where M and N have the same value, for the m-th decoder layer from the 2nd decoder layer to the M-th decoder layer, the input data of the m-th decoder layer is the output data of the (m-1)-th decoder layer and the output data of the (M+1-m)-th encoder layer.

[0125] Based on the target vehicle speed characteristics, obtain the predicted vehicle speed for the second time period after the current moment.

[0126] In this embodiment, the encoding / decoding and separation modules in the deep complex convolutional network simulate complex number operations in both the feature extraction and feature separation stages of the input sequence, ensuring that the feature maps always satisfy the complex number operation relationship, thereby improving the prediction accuracy of vehicle speed signals and the stability of model training.

[0127] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0128] Based on the same inventive concept, this application also provides a vehicle speed prediction device for implementing the vehicle speed prediction method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more vehicle speed prediction device embodiments provided below can be found in the limitations of the vehicle speed prediction method described above, and will not be repeated here.

[0129] In one exemplary embodiment, such as Figure 9 As shown, a vehicle speed prediction device is provided, comprising: an acquisition module 10, an extraction module 20, and a prediction module 40, wherein:

[0130] The acquisition module 10 is used to acquire vehicle speed data within the first time period before the current moment when the target vehicle is in the target operating condition; the vehicle speed data is sequential data.

[0131] The extraction module 20 is used to extract features from vehicle speed data through a pre-trained ensemble neural network model to obtain high-dimensional complex features, and to perform dimensionality reduction processing on the high-dimensional complex features through the ensemble neural network model to obtain the target vehicle speed features. The ensemble neural network model includes at least one complex encoder and at least one complex decoder. The output data of the complex encoder is input to the corresponding complex decoder through a jump connection.

[0132] The prediction module 40 is used to obtain the predicted vehicle speed in the second time period after the current moment based on the target vehicle speed characteristics.

[0133] In an exemplary embodiment, the extraction module 20 is further configured to divide the vehicle speed data to obtain training set data and test set data; determine initialization parameters and construct an initial combined model based on the initialization parameters; train the initial combined model based on the training set data and test the trained initial combined model based on the test set data.

[0134] In an exemplary embodiment, the combined neural network model includes a network intermediate layer and a network output layer. The extraction module 20 is further configured to process the initial combined model using training set data, obtain the output layer error corresponding to the network output layer, and obtain the intermediate layer error corresponding to the network intermediate layer based on the output layer error; adjust the weights and thresholds of the network intermediate layer based on the intermediate layer error, and adjust the weights and thresholds of the network intermediate layer based on the adjusted network intermediate layer and output layer errors; update the initialization parameters, return to the step of constructing the initial combined model based on the initialization parameters, and continue execution until the initial combined model meets the learning requirements; if the initial combined model meets the learning requirements, process the adjusted initial combined model using test set data to obtain test results; if the test results do not meet the error requirements and the target number of training iterations has not been reached, update the initialization parameters, return to the step of constructing the initial combined model based on the initialization parameters, and continue execution until the test results meet the error requirements or the target number of training iterations has been reached.

[0135] In an exemplary embodiment, the intermediate layer of the network includes an encoder layer and a decoder layer; the encoder layer includes a complex encoder, and the decoder layer includes a complex decoder; the extraction module 20 is further configured to perform convolution processing on the vehicle speed data sequentially through N encoder layers to obtain high-dimensional complex features; the input data of the first encoder layer is the vehicle speed data; where N is an even number, and for the i-th encoder layer among the second to N encoder layers, the input data of the i-th encoder layer is the output data of the (i-1)-th encoder layer.

[0136] In an exemplary embodiment, the complex encoder includes a real part convolution filter and an imaginary part convolution filter; the extraction module 20 is further configured to extract features from the vehicle speed data for each complex encoder layer by using the real part convolution filter and the imaginary part convolution filter respectively, to obtain the corresponding real part convolution features and imaginary part convolution features; and to normalize the real part convolution features and imaginary part convolution features to obtain the high-dimensional complex features output by the complex encoder.

[0137] In an exemplary embodiment, the combined neural network model further includes a separation module; the dimensionality reduction module 30 is further configured to perform separation processing on the high-dimensional complex features output by the Nth encoder layer through the separation module to obtain separated features;

[0138] The target vehicle speed feature is obtained by sequentially performing dimensionality reduction processing on the separated features through M decoder layers; the input data of the first decoder layer is the separated feature.

[0139] Where M and N have the same value, for the m-th decoder layer from the 2nd decoder layer to the M-th decoder layer, the input data of the m-th decoder layer is the output data of the (m-1)-th decoder layer and the output data of the (M+1-m)-th encoder layer.

[0140] In an exemplary embodiment, the separation module includes a real part separation module and an imaginary part separation module; the high-dimensional complex features include high-dimensional real part features and high-dimensional imaginary part features; the dimensionality reduction module 30 is further configured to perform complex number operations on the high-dimensional real part features and high-dimensional imaginary part features respectively through the real part separation module and the imaginary part separation module to obtain the separated features.

[0141] The various modules in the aforementioned vehicle speed prediction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0142] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 10As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a vehicle speed prediction method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0143] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0144] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps: when a target vehicle is in a target operating condition, acquire vehicle speed data for a first time period before the current moment; the vehicle speed data is sequential data; extract features from the vehicle speed data using a pre-trained ensemble neural network model to obtain high-dimensional complex features, and perform dimensionality reduction processing on the high-dimensional complex features using the ensemble neural network model to obtain target vehicle speed features; the ensemble neural network model includes at least one complex encoder and at least one complex decoder, and the output data of the complex encoder is input to the corresponding complex decoder through a jump connection; based on the target vehicle speed features, acquire the predicted vehicle speed for a second time period after the current moment.

[0145] In one embodiment, the training process of the combined neural network model involved when the processor executes the computer program includes: dividing the vehicle speed data to obtain training set data and test set data; determining initialization parameters and constructing an initial combined model based on the initialization parameters; training the initial combined model based on the training set data and testing the trained initial combined model based on the test set data.

[0146] In one embodiment, the combined neural network model includes a network intermediate layer and a network output layer. The steps involved in the processor executing the computer program—training an initial combined model based on training set data and testing the trained initial combined model based on test set data—include: processing the initial combined model using training set data to obtain the output layer error corresponding to the network output layer, and obtaining the intermediate layer error corresponding to the network intermediate layer based on the output layer error; adjusting the weights and thresholds of the network intermediate layer based on the intermediate layer error, and adjusting the weights and thresholds of the network intermediate layer based on the adjusted intermediate layer and output layer errors; updating the initialization parameters, returning to the step of constructing the initial combined model based on the initialization parameters, and continuing execution until the initial combined model meets the learning requirements; if the initial combined model meets the learning requirements, processing the adjusted initial combined model using test set data to obtain test results; if the test results do not meet the error requirements and the target number of training iterations has not been reached, updating the initialization parameters, returning to the step of constructing the initial combined model based on the initialization parameters, and continuing execution until the test results meet the error requirements or the target number of training iterations has been reached.

[0147] In one embodiment, the intermediate layer of the network includes an encoder layer and a decoder layer; the encoder layer includes a complex encoder, and the decoder layer includes a complex decoder; the step of extracting features from vehicle speed data through a pre-trained combined neural network model to obtain high-dimensional complex features when the processor executes the computer program includes: sequentially performing convolution processing on the vehicle speed data through N encoder layers to obtain high-dimensional complex features; the input data of the first encoder layer is the vehicle speed data; where N is an even number, and for the i-th encoder layer among the second to N encoder layers, the input data of the i-th encoder layer is the output data of the (i-1)-th encoder layer.

[0148] In one embodiment, the complex encoder includes a real part convolution filter and an imaginary part convolution filter. The step involved in the processor executing the computer program to sequentially convolve the vehicle speed data through N encoder layers to obtain high-dimensional complex features includes: for each encoder layer and its corresponding complex encoder, extracting features from the vehicle speed data through the real part convolution filter and the imaginary part convolution filter respectively to obtain the corresponding real part convolution features and imaginary convolution features; and normalizing the real part convolution features and the imaginary convolution features to obtain the high-dimensional complex features output by the complex encoder.

[0149] In one embodiment, the combined neural network model further includes a separation module; the step of performing dimensionality reduction processing on high-dimensional complex features through the combined neural network model to obtain target vehicle speed features when the processor executes the computer program includes: performing separation processing on the high-dimensional complex features output by the Nth encoder layer through the separation module to obtain separated features; performing dimensionality reduction processing on the separated features sequentially through M decoder layers to obtain target vehicle speed features; the input data of the first decoder layer is the separated features; wherein, M and N have the same value, and for the mth decoder layer among the 2nd to Mth decoder layers, the input data of the mth decoder layer is the output data of the (m-1)th decoder layer and the output data of the (M+1-m)th encoder layer.

[0150] In one embodiment, the separation module includes a real separation module and an imaginary separation module; the high-dimensional complex features include high-dimensional real features and high-dimensional imaginary features; the step of separating the high-dimensional complex features output by the Nth encoder layer through the separation module to obtain the separated features when the processor executes the computer program includes: performing complex number operations on the high-dimensional real features and high-dimensional imaginary features respectively through the real separation module and the imaginary separation module to obtain the separated features.

[0151] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, it performs the following steps: when the target vehicle is in a target operating condition, acquire vehicle speed data for a first time period before the current moment; the vehicle speed data is sequential data; extract features from the vehicle speed data using a pre-trained ensemble neural network model to obtain high-dimensional complex features, and perform dimensionality reduction processing on the high-dimensional complex features using the ensemble neural network model to obtain target vehicle speed features; the ensemble neural network model includes at least one complex encoder and at least one complex decoder, and the output data of the complex encoder is input to the corresponding complex decoder through a jump connection; based on the target vehicle speed features, acquire the predicted vehicle speed for a second time period after the current moment.

[0152] In one embodiment, the training process of the combined neural network model involved when the computer program is executed by the processor includes: dividing the vehicle speed data to obtain training set data and test set data; determining initialization parameters and constructing an initial combined model based on the initialization parameters; training the initial combined model based on the training set data and testing the trained initial combined model based on the test set data.

[0153] In one embodiment, the combined neural network model includes intermediate layers and an output layer. The steps involved in training an initial combined model based on training set data and testing the trained initial combined model based on test set data, when the computer program is executed by a processor, include: processing the initial combined model using the training set data to obtain the output layer error corresponding to the output layer, and obtaining the intermediate layer error corresponding to the intermediate layer based on the output layer error; adjusting the weights and thresholds of the intermediate layers based on the intermediate layer errors, and further adjusting the intermediate layer weights and thresholds based on the adjusted intermediate layer and output layer errors; updating the initialization parameters, returning to the step of constructing the initial combined model based on the initialization parameters, and continuing execution until the initial combined model meets the learning requirements; if the initial combined model meets the learning requirements, processing the adjusted initial combined model using test set data to obtain test results; if the test results do not meet the error requirements and the target number of training iterations has not been reached, updating the initialization parameters, returning to the step of constructing the initial combined model based on the initialization parameters, and continuing execution until the test results meet the error requirements or the target number of training iterations has been reached.

[0154] In one embodiment, the intermediate layer of the network includes an encoder layer and a decoder layer; the encoder layer includes a complex encoder, and the decoder layer includes a complex decoder; the step of extracting features from vehicle speed data through a pre-trained combined neural network model to obtain high-dimensional complex features when the computer program is executed by the processor includes: sequentially performing convolution processing on the vehicle speed data through N encoder layers to obtain high-dimensional complex features; the input data of the first encoder layer is the vehicle speed data; where N is an even number, and for the i-th encoder layer among the second to N encoder layers, the input data of the i-th encoder layer is the output data of the (i-1)-th encoder layer.

[0155] In one embodiment, the complex encoder includes a real part convolution filter and an imaginary part convolution filter. The step involved in the computer program being executed by a processor to sequentially convolve vehicle speed data through N encoder layers to obtain high-dimensional complex features includes: for each encoder layer's corresponding complex encoder, extracting features from the vehicle speed data using both the real part convolution filter and the imaginary part convolution filter to obtain corresponding real part convolution features and imaginary convolution features; and normalizing the real part convolution features and imaginary convolution features to obtain the high-dimensional complex features output by the complex encoder.

[0156] In one embodiment, the combined neural network model further includes a separation module; the step of performing dimensionality reduction processing on high-dimensional complex features through the combined neural network model to obtain target vehicle speed features when the computer program is executed by the processor includes: performing separation processing on the high-dimensional complex features output by the Nth encoder layer through the separation module to obtain separated features; performing dimensionality reduction processing on the separated features sequentially through M decoder layers to obtain target vehicle speed features; the input data of the first decoder layer is the separated features; wherein, M and N have the same value, and for the mth decoder layer among the 2nd to Mth decoder layers, the input data of the mth decoder layer is the output data of the (m-1)th decoder layer and the output data of the (M+1-m)th encoder layer.

[0157] In one embodiment, the separation module includes a real separation module and an imaginary separation module; the high-dimensional complex features include high-dimensional real features and high-dimensional imaginary features; the step of separating the high-dimensional complex features output by the Nth encoder layer through the separation module to obtain the separated features when the computer program is executed by the processor includes: performing complex number operations on the high-dimensional real features and high-dimensional imaginary features respectively through the real separation module and the imaginary separation module to obtain the separated features.

[0158] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps: when the target vehicle is in a target operating condition, acquire vehicle speed data for a first time period prior to the current moment; the vehicle speed data is sequential data; extract features from the vehicle speed data using a pre-trained ensemble neural network model to obtain high-dimensional complex features, and perform dimensionality reduction processing on the high-dimensional complex features using the ensemble neural network model to obtain target vehicle speed features; the ensemble neural network model includes at least one complex encoder and at least one complex decoder, the output data of the complex encoder being input to the corresponding complex decoder via a jump connection; and based on the target vehicle speed features, acquire the predicted vehicle speed for a second time period after the current moment.

[0159] In one embodiment, the training process of the combined neural network model involved when the computer program is executed by the processor includes: dividing the vehicle speed data to obtain training set data and test set data; determining initialization parameters and constructing an initial combined model based on the initialization parameters; training the initial combined model based on the training set data and testing the trained initial combined model based on the test set data.

[0160] In one embodiment, the combined neural network model includes intermediate layers and an output layer. The steps involved in training an initial combined model based on training set data and testing the trained initial combined model based on test set data, when the computer program is executed by a processor, include: processing the initial combined model using the training set data to obtain the output layer error corresponding to the output layer, and obtaining the intermediate layer error corresponding to the intermediate layer based on the output layer error; adjusting the weights and thresholds of the intermediate layers based on the intermediate layer errors, and further adjusting the intermediate layer weights and thresholds based on the adjusted intermediate layer and output layer errors; updating the initialization parameters, returning to the step of constructing the initial combined model based on the initialization parameters, and continuing execution until the initial combined model meets the learning requirements; if the initial combined model meets the learning requirements, processing the adjusted initial combined model using test set data to obtain test results; if the test results do not meet the error requirements and the target number of training iterations has not been reached, updating the initialization parameters, returning to the step of constructing the initial combined model based on the initialization parameters, and continuing execution until the test results meet the error requirements or the target number of training iterations has been reached.

[0161] In one embodiment, the intermediate layer of the network includes an encoder layer and a decoder layer; the encoder layer includes a complex encoder, and the decoder layer includes a complex decoder; the step of extracting features from vehicle speed data through a pre-trained combined neural network model to obtain high-dimensional complex features when the computer program is executed by the processor includes: sequentially performing convolution processing on the vehicle speed data through N encoder layers to obtain high-dimensional complex features; the input data of the first encoder layer is the vehicle speed data; where N is an even number, and for the i-th encoder layer among the second to N encoder layers, the input data of the i-th encoder layer is the output data of the (i-1)-th encoder layer.

[0162] In one embodiment, the complex encoder includes a real part convolution filter and an imaginary part convolution filter. The step involved in the computer program being executed by a processor to sequentially convolve vehicle speed data through N encoder layers to obtain high-dimensional complex features includes: for each encoder layer's corresponding complex encoder, extracting features from the vehicle speed data using both the real part convolution filter and the imaginary part convolution filter to obtain corresponding real part convolution features and imaginary convolution features; and normalizing the real part convolution features and imaginary convolution features to obtain the high-dimensional complex features output by the complex encoder.

[0163] In one embodiment, the combined neural network model further includes a separation module; the step of performing dimensionality reduction processing on high-dimensional complex features through the combined neural network model to obtain target vehicle speed features when the computer program is executed by the processor includes: performing separation processing on the high-dimensional complex features output by the Nth encoder layer through the separation module to obtain separated features; performing dimensionality reduction processing on the separated features sequentially through M decoder layers to obtain target vehicle speed features; the input data of the first decoder layer is the separated features; wherein, M and N have the same value, and for the mth decoder layer among the 2nd to Mth decoder layers, the input data of the mth decoder layer is the output data of the (m-1)th decoder layer and the output data of the (M+1-m)th encoder layer.

[0164] In one embodiment, the separation module includes a real separation module and an imaginary separation module; the high-dimensional complex features include high-dimensional real features and high-dimensional imaginary features; the step of separating the high-dimensional complex features output by the Nth encoder layer through the separation module to obtain the separated features when the computer program is executed by the processor includes: performing complex number operations on the high-dimensional real features and high-dimensional imaginary features respectively through the real separation module and the imaginary separation module to obtain the separated features.

[0165] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0166] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0167] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A vehicle speed prediction method characterized by comprising: The method comprises: In the case that the target vehicle is in a target working condition, vehicle speed data in a first time period before the current time is acquired; the vehicle speed data is sequence data; Feature extraction is performed on the vehicle speed data by a pre-trained combined neural network model to obtain high-dimensional complex features, and dimension reduction processing is performed on the high-dimensional complex features by the combined neural network model to obtain target vehicle speed features; the combined neural network model comprises at least one complex encoder and at least one complex decoder, and output data of the complex encoder is input to the corresponding complex decoder through a jump connection; the combined neural network model further comprises a separation module; the separation module comprises a complex GRU module and a linear layer, the complex GRU module is used for complex operation on the high-dimensional complex features to enhance the features, and the linear layer is used for linear transformation on the enhanced features to output separation features; According to the target vehicle speed features, predicted vehicle speeds in a second time period after the current time are acquired.

2. The method of claim 1, wherein, The training process of the combined neural network model comprises: The vehicle speed data is divided to obtain training set data and test set data; Initialization parameters are determined, and an initial combined model is constructed according to the initialization parameters; The initial combined model is trained based on the training set data, and the trained initial combined model is tested based on the test set data.

3. The method of claim 2, wherein, The combined neural network model comprises a network intermediate layer and a network output layer; the training of the initial combined model based on the training set data and the testing of the trained initial combined model based on the test set data comprise: The initial combined model is processed by the training set data, output layer errors corresponding to the network output layer are respectively acquired, and intermediate layer errors corresponding to the network intermediate layer are acquired according to the output layer errors; The weights and thresholds of the network intermediate layer are adjusted according to the intermediate layer errors, and the weights and thresholds of the network intermediate layer are adjusted according to the adjusted network intermediate layer and the output layer errors; The initialization parameters are updated, the step of constructing the initial combined model according to the initialization parameters is returned, and the execution is continued until the initial combined model meets the learning requirement condition; In the case that the initial combined model meets the learning requirement condition, the adjusted initial combined model is processed by the test set data to acquire a test result; In the case that the test result does not meet the error requirement condition and the target training number is not reached, the initialization parameters are updated, the step of constructing the initial combined model according to the initialization parameters is returned, and the execution is continued until the test result meets the error requirement condition or the target training number is reached.

4. The method according to any one of claims 1 to 3, characterized in that, The network intermediate layer comprises an encoder layer and a decoder layer; the encoder layer comprises a complex encoder, and the decoder layer comprises a complex decoder; the feature extraction on the vehicle speed data by the pre-trained combined neural network model to obtain high-dimensional complex features comprises: The vehicle speed data is sequentially convoluted by N encoder layers to obtain high-dimensional complex features; input data of a first encoder layer is the vehicle speed data; wherein N is an even number, and for an i-th encoder layer in a second encoder layer to an N-th encoder layer, input data of the i-th encoder layer is output data of an (i-1)-th encoder layer.

5. The method of claim 4, wherein, The complex encoder comprises a real part convolution filter and an imaginary part convolution filter; and the sequentially convoluting the vehicle speed data by N encoder layers to obtain high-dimensional complex features comprises: For each complex encoder corresponding to each encoder layer, the real part convolution filter and the imaginary part convolution filter are used to extract features from the vehicle speed data respectively to obtain corresponding real part convolution features and imaginary part convolution features; The real part convolution features and the imaginary part convolution features are normalized to obtain high-dimensional complex features output by the complex encoder.

6. The method of claim 4, wherein, The high-dimensional complex features are processed by the combined neural network model to obtain target vehicle speed features, which comprises: The high-dimensional complex features output by the N-th encoder layer are separated by the separation module to obtain separated features; The separated features are sequentially processed by M decoder layers to obtain target vehicle speed features; input data of a first decoder layer is the separated features; wherein M and N have the same value, and for an m-th decoder layer in a second decoder layer to an M-th decoder layer, input data of the m-th decoder layer is output data of an (m-1)-th decoder layer and output data of an (M+1-m)-th encoder layer.

7. The method of claim 6, wherein, The separation module comprises a real part separation module and an imaginary part separation module; the high-dimensional complex features comprise high-dimensional real part features and high-dimensional imaginary part features; and the high-dimensional complex features output by the N-th encoder layer are separated by the separation module to obtain separated features, which comprises: The real part separation module and the imaginary part separation module are used to perform complex operation processing on the high-dimensional real part features and the high-dimensional imaginary part features respectively to obtain separated features.

8. A vehicle speed prediction device characterized by comprising: The device comprises: An acquisition module is configured to acquire vehicle speed data in a first time period before a current time when a target vehicle is in a target working condition; the vehicle speed data is sequence data; An extraction module is configured to extract features from the vehicle speed data by a combined neural network model pre-trained, to obtain high-dimensional complex features, and to process the high-dimensional complex features by the combined neural network model to obtain target vehicle speed features; the combined neural network model comprises at least one complex encoder and at least one complex decoder, and output data of the complex encoder is input to a corresponding complex decoder through a skip connection; the combined neural network model further comprises a separation module; the separation module comprises a complex GRU module and a linear layer, the complex GRU module is configured to perform complex operation on high-dimensional complex features to enhance features, and the linear layer is configured to perform linear transformation on the enhanced features to output separated features; A prediction module is configured to acquire a predicted vehicle speed in a second time period after the current time according to the target vehicle speed feature. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor implements the steps of the method of any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 7.

Images