Train position prediction method and device, equipment, storage medium and program product
By using the H∞ diffusion Kalman filter data fusion algorithm and long and short-term memory network in the train positioning system, the problem that a single sensor system cannot provide high-precision positioning is solved, and more accurate train position prediction is achieved.
Patent Information
- Application Number
- CN202411867784.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-18
- Publication Date
- 2025-05-06
AI Technical Summary
In the prior art, a single sensor system cannot provide continuous high accuracy and high reliability train positioning, resulting in poor train position prediction results.
The data fusion algorithm based on H∞ diffusion Kalman filtering is used to clean, characterize and normalize the multidimensional train data, and input it into the long and short-term memory network to output the train position prediction information.
Through data fusion and feature processing, the accuracy and reliability of train positioning are improved, and more accurate train position prediction information is provided.
Smart Images
Figure CN119939498A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to a train position prediction method, device, equipment, storage medium, and program product. Background Art
[0002] In the field of train positioning technology, traditional positioning methods rely on a single sensor system, such as the BeiDou Navigation Satellite System (BDS) or the odometer (ODO). Each of these systems has its own limitations: BDS may be affected by satellite errors, signal propagation errors, multipath effects, etc., while ODO has error accumulation problems due to idling, sliding, and wheel wear. Therefore, a single sensor cannot provide continuous high-precision and high-reliability positioning.
[0003] Therefore, how to effectively predict train locations has become a problem that needs to be solved urgently in the industry. Summary of the invention
[0004] The present invention provides a train position prediction method, device, equipment, storage medium, and program product to solve the problem of how to effectively predict the train position in the prior art.
[0005] The present invention provides a train position prediction method, comprising the following steps: After data cleaning of the original multi-dimensional train data information of the train, multi-dimensional train data information is obtained; Based on the H∞ diffusion Kalman filter data fusion algorithm, the multi-dimensional train data information is fused and the fused train data is output; wherein the H∞ diffusion Kalman filter data fusion algorithm is a data fusion algorithm that replaces the covariance prediction formula of the Kalman filter with the covariance prediction formula of the H∞ filter; The fused train data is characterized and normalized to obtain normalized features, and the normalized features are input into a long short-term memory network to output the position prediction information of the train.
[0006] According to a train position prediction method provided by the present invention, the normalized features are input into a long short-term memory network, and the train position prediction information is output, including: Inputting the normalized features into the long short-term memory network, and outputting instantaneous speed prediction information of the train; Based on the train position information, train travel direction information and instantaneous speed prediction information of the train at the previous moment, the position prediction information of the train at the current moment is predicted.
[0007] According to a train position prediction method provided by the present invention, the H∞ diffusion Kalman filter data fusion algorithm is used to fuse the multi-dimensional train data information and output fused train data, including: Initializing a state vector of a system model according to the odometer data information in the multi-dimensional train data information; Initializing an observation vector of an observation model according to the satellite positioning data information in the multidimensional train data information, and calculating an observation noise covariance matrix according to the observation vector; In each time step, the H∞ diffusion Kalman filter data fusion algorithm is used to evaluate the uncertainty of the system model and the observation model, and the state estimation and error covariance are updated to achieve data fusion of the multidimensional train data information and output fused train data.
[0008] According to a train position prediction method provided by the present invention, after data cleaning of original multi-dimensional train data information of a train, multi-dimensional train data information is obtained, including: Correcting or deleting abnormal values in the original multi-dimensional train data information; Average data filling is performed on the data missing values in the original multi-dimensional train data information; wherein the average data is determined based on the average value of the data adjacent to the missing data value.
[0009] According to a train position prediction method provided by the present invention, the fused train data is characterized and normalized to obtain normalized features, including: After respectively performing time feature extraction, position feature conversion, periodic encoding of travel direction and speed feature processing on the fused train data, normalization processing is performed to obtain normalized features; Wherein, the time feature extraction includes: relative time feature extraction and period feature extraction; the position feature conversion includes: differential feature extraction and polar coordinate conversion.
[0010] The present invention also provides a train position prediction device, comprising the following modules: A cleaning module is used to clean the original multi-dimensional train data information of the train to obtain multi-dimensional train data information; A fusion module is used to perform data fusion on the multi-dimensional train data information based on an H∞ diffusion Kalman filter data fusion algorithm, and output fused train data; wherein the H∞ diffusion Kalman filter data fusion algorithm is a data fusion algorithm that replaces the covariance prediction formula of the Kalman filter with the covariance prediction formula of the H∞ filter; The output module is used to perform characterization and normalization processing on the fused train data to obtain normalized features, and input the normalized features into a long short-term memory network to output the position prediction information of the train.
[0011] According to a train position prediction device provided by the present invention, the device is also used for: Inputting the normalized features into the long short-term memory network, and outputting instantaneous speed prediction information of the train; Based on the train position information, train travel direction information and instantaneous speed prediction information of the train at the previous moment, the position prediction information of the train at the current moment is predicted.
[0012] According to a train position prediction device provided by the present invention, the device is also used for: Initializing a state vector of a system model according to the odometer data information in the multi-dimensional train data information; Initializing an observation vector of an observation model according to the satellite positioning data information in the multidimensional train data information, and calculating an observation noise covariance matrix according to the observation vector; In each time step, the H∞ diffusion Kalman filter data fusion algorithm is used to evaluate the uncertainty of the system model and the observation model, and the state estimation and error covariance are updated to achieve data fusion of the multidimensional train data information and output fused train data.
[0013] According to a train position prediction device provided by the present invention, the device is also used for: Correcting or deleting abnormal values in the original multi-dimensional train data information; Average data filling is performed on the data missing values in the original multi-dimensional train data information; wherein the average data is determined based on the average value of the data adjacent to the missing data value.
[0014] According to a train position prediction device provided by the present invention, the device is also used for: After respectively performing time feature extraction, position feature conversion, periodic encoding of travel direction and speed feature processing on the fused train data, normalization processing is performed to obtain normalized features; Wherein, the time feature extraction includes: relative time feature extraction and period feature extraction; the position feature conversion includes: differential feature extraction and polar coordinate conversion.
[0015] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the train position prediction method described above is implemented.
[0016] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the train position prediction method as described in any one of the above is implemented.
[0017] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the train position prediction method as described above is implemented.
[0018] The train position prediction method, device, equipment, storage medium, and program product provided by the present invention improve the accuracy and reliability of train positioning through a data fusion algorithm based on H∞ diffusion Kalman filtering. The method achieves effective fusion of multi-dimensional train data information by evaluating the uncertainty of the system model and the observation model, and updating the state estimation and error covariance. Through this method, more accurate fused train data can be output, providing strong support for train positioning. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0020] Figure 1 It is a flow chart of the train position prediction method provided by the present invention.
[0021] Figure 2 Schematic diagram of the working principle of the FCL layer provided for the implementation of this application.
[0022] Figure 3 Schematic diagram of the long short-term memory network model.
[0023] Figure 4 A schematic diagram of the structure of a train position prediction device provided in an embodiment of the present application; Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION
[0024] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0025] Figure 1 FIG. 1 is a flow chart of the train position prediction method provided by the present invention, such as Figure 1 As shown, the method includes the following: Step 110, after data cleaning of the original multi-dimensional train data information of the train, multi-dimensional train data information is obtained; In an embodiment of the present application, the original multi-dimensional train data information may specifically include: timestamp (t); longitude (lon_BDS), latitude (lat_BDS), ground heading (dir_BDS) and signal-to-noise ratio (SNR) obtained by BDS; speed (v_ODO) obtained by ODO; longitude (lon_map) and latitude (lat_map) of the section in the electronic map.
[0026] In the embodiment of the present application, data cleaning is performed on the original multi-dimensional train data information of the train, which specifically means that in the data preprocessing stage, outlier detection is first performed on each data point in the data set, including checking key fields such as longitude, latitude, speed and driving direction.
[0027] For detected outliers, such as data points with negative speed or driving direction not between 0 and 360 degrees, correction or deletion is required. These outliers may be caused by sensor errors, data recording errors or other abnormal conditions.
[0028] Correcting outliers may involve adjusting speed values to a reasonable range, or adjusting driving directions to a standard range of 0 to 360 degrees.
[0029] If the outliers cannot be reasonably corrected, or the number of outliers is large, you may need to delete the data points containing outliers to prevent them from adversely affecting subsequent analysis.
[0030] After outlier processing, it is necessary to detect missing values in the dataset. Missing values may be caused by sensor failure, data transmission interruption, or other technical issues.
[0031] There are many ways to handle missing data points. A common method is to fill the missing values with the average of the previous and next data points, which can maintain the continuity of the data.
[0032] Another approach is to directly delete data points containing missing values, which is simple and effective, especially when the number of missing values is small.
[0033] In some cases, if the missing data points have little impact on the overall analysis, you can choose to retain the missing values, but the impact of these missing values needs to be considered in the subsequent analysis to finally obtain multi-dimensional train data information.
[0034] Step 120, performing data fusion on the multi-dimensional train data information based on an H∞ diffusion Kalman filter data fusion algorithm, and outputting fused train data; wherein the H∞ diffusion Kalman filter data fusion algorithm is a data fusion algorithm that replaces the covariance prediction formula of the Kalman filter with the covariance prediction formula of the H∞ filter; In the embodiment of the present application, during the travel of a heavy-load railway train, the BDS receiver may be affected by the train passing through a section where the signal is blocked or lost (such as a tunnel), and the ODO data may also produce errors due to road surface and mechanical changes. Therefore, a filter is needed that can improve the accuracy and robustness of the estimation while maintaining real-time processing capabilities. Combining the robustness of the H∞ filter and the nonlinear processing capability of the EKF, a data fusion algorithm combining H∞ and the standard EKF is proposed.
[0035] EKF is very suitable for handling nonlinear systems by linearizing at each estimation step. However, EKF has three shortcomings: the accuracy is reduced when the system model and observation model are not accurately designed or the signal noise varies greatly; it assumes that the process noise and measurement noise are white noise and follow Gaussian distribution, but the noise characteristics in actual applications are often more complicated; it may show unstable convergence behavior when the initial conditions are not good or the noise is too large.
[0036] In contrast, H∞ filters are very robust in the face of model uncertainty and non-Gaussian noise. H∞ filtering does not rely on the probabilistic statistical properties of the Gaussian assumption, but instead uses a deterministic approach to deal with uncertainty, ensuring system stability and performance in the worst case by optimizing some energy or gain metric.
[0037] By combining H∞ and EKF, and using the error covariance of H∞ to update the error covariance matrix of EKF, we can simultaneously utilize the robustness of H∞ and the nonlinear processing capability of EKF to improve the overall performance of the system. This combination can provide more accurate state estimation in the complex background of uncertainty and noise.
[0038] In practical applications, the design of H∞ filter is mainly suitable for linear systems, but has poor ability in dealing with nonlinear systems, and the design and implementation are relatively complex and require careful adjustment of parameters to obtain optimal performance. Therefore, by combining H∞ and EKF, the shortcomings of EKF in robustness can be made up, while the advantages of H∞ filter in dealing with uncertainty can be used to achieve better data fusion effect. Step 130, characterizing and normalizing the fused train data to obtain normalized features, and inputting the normalized features into a long short-term memory network to output the position prediction information of the train.
[0039] In the embodiment of the present application, after respectively performing time feature extraction, position feature conversion, periodic encoding of travel direction and speed feature processing on the fused train data, normalization processing is performed to obtain normalized features; Wherein, the time feature extraction includes: relative time feature extraction and period feature extraction; the position feature conversion includes: differential feature extraction and polar coordinate conversion.
[0040] Time features are very important for time series prediction models. The following features can be extracted from timestamps: Relative time, which converts the timestamp to a relative time from the start of the trajectory, such as seconds or minutes. This helps the model understand the sequence and time intervals of events. Periodic features, which extract periodic features such as hours and days of the week, which can help the model capture patterns of periodic changes.
[0041] In this application, the UTC timestamp is converted to the relative number of seconds from the start of the track. The relative time can be obtained by subtracting the first timestamp in the track, as shown in the following formula: ; In the formula, is the timestamp of the start of the trajectory. The relative time is then normalized to scale it to a reasonable range, as shown in the following formula: ; In the formula, and are the minimum and maximum relative time in the training set, respectively.
[0042] Position feature conversion: Longitude and latitude coordinates are not the best representation for the model. They can be converted in the following ways: Differential features: Calculate the longitude and latitude differences between adjacent time points, which can help the model better understand the changing trend of the position. Polar coordinate conversion: Convert the longitude and latitude differences into distances and angles in the polar coordinate system, which can more intuitively represent the position change.
[0043] Since the range of longitude and latitude is known, longitude is from -180 to 180, and latitude is from -90 to 90, this article chooses the following formula to normalize longitude and latitude to the interval [0, 1], as shown in the following formula: ; ; Periodic encoding of driving direction. Driving direction is a periodic feature, which can be converted into sine and cosine values to better represent this periodicity, as shown in the following formula: ; ; This transformation helps the model understand continuity and periodicity in directions, for example, 359 degrees and 0 degrees are very close in direction.
[0044] Speed feature processing, speed can be used directly as a feature, but the data can also be divided into different intervals according to the speed value, such as static, low speed, medium speed and high speed, which helps the model capture the behavior patterns at different speeds. This article uses Z-score normalization to process speed, as shown in the following formula: ; In the formula, and are the mean and standard deviation of the speed in the training set, respectively.
[0045] In the embodiment of the present application, the accuracy and reliability of train positioning are improved by using a data fusion algorithm based on H∞ diffusion Kalman filtering. This method achieves effective fusion of multi-dimensional train data information by evaluating the uncertainty of the system model and the observation model, and updating the state estimation and error covariance. In this way, more accurate fused train data can be output, providing strong support for train positioning.
[0046] Optionally, the performing data fusion on the multi-dimensional train data information based on the H∞ diffusion Kalman filter data fusion algorithm and outputting fused train data includes: Initializing a state vector of a system model according to the odometer data information in the multi-dimensional train data information; Initializing an observation vector of an observation model according to the satellite positioning data information in the multidimensional train data information, and calculating an observation noise covariance matrix according to the observation vector; In each time step, the H∞ diffusion Kalman filter data fusion algorithm is used to evaluate the uncertainty of the system model and the observation model, and the state estimation and error covariance are updated to achieve data fusion of the multidimensional train data information and output fused train data.
[0047] In the embodiment of the present application, two models, the system model and the observation model, must first be defined in the multi-sensor information fusion positioning process of the train. The system model then pre-estimates the longitude and latitude of the train and the system error, and the observation model corrects the prior estimates of the longitude and latitude of the train and the system error. This paper constructs the system model with the motion model of ODO and constructs the observation model with the motion model of GPS.
[0048] ODO system model, the state vector is defined as The state prediction equation is shown in formula (1): (1) In the formula is the state vector at time k-1 and system control input The nonlinear state transfer function is is the Gaussian white noise of the system at time k-1, which is normally distributed. Because the sampling calculation period is 500ms, The train can be regarded as moving at a constant speed during the time, so the state transfer function The equation is shown in formula (2): (2) In the formula is the angle calculated based on the line segment at time k-1 on the electronic map, and It is calculated and The change in direction is shown in formulas (3) and (4): (3) (4) In the formula is the radian calculation formula, is the radius of the Earth.
[0049] Linearize the system state model and find About Status The Jacobian matrix , as shown in formula (5): (5) The error covariance prediction is shown in formula (6): (6) In the formula is the error covariance estimated at time k-1, is the k-1 moment process noise covariance matrix, as shown in formula (7): (7) In the formula and is the standard deviation of process noise and should be adjusted based on actual conditions.
[0050] BDS observation model, the actual measurement value observation vector at the kth moment As shown in formula (8): (8) In the formula is the observation noise at time k, which is Gaussian normally distributed, so the observation noise covariance matrix As shown in formula (9): (9) In the formula and is the standard deviation of the observation noise and is adjusted according to actual conditions.
[0051] Linearize the observation model and find the k moment About Status The Jacobian matrix , as shown in formula (10): (10) Calculate Kalman gain , as shown in formula (11): (11) It can be deduced from the formula that when The larger the Kalman gain is, the The larger it is, the more trust is placed on the BDS measurement, and vice versa, the more trust is placed on the ODO state estimate.
[0052] Update the system state and error covariance as shown in equations (12) and (13): (12) (13) The prior estimated state vector established by ODO is updated and corrected by the Kalman gain calculated by the observation model established by BDS to obtain the final estimated state vector and the optimal solution of error covariance.
[0053] The mathematical model of the filter, the linear combination of the system equation and the state quantity after linearization and discretization of the nonlinear system formula is shown in formula (14). (14) In the formula, is the estimated output, is a linear matrix. Filtering is used to ensure the transfer function The norm is minimized so that The filtering has better robustness, and its cost function is shown in formula (15): (15) In the formula, for The covariance matrix of =I, so and Same. If the upper bound of the cost function is 1 / ,but , are pre-set parameters. The derivation process of filtering is shown in formulas (16) to (20). (16) (17) (18) (19) (20) in Need to meet: (twenty one) (twenty two) The idea used in this paper is to combine the nonlinear processing capability of EKF and The main use case for filtering robustness is to introduce The covariance prediction formula of the H∞ filter replaces the covariance prediction formula of the EKF mentioned above. The covariance prediction formula of the H∞ filter does not follow the traditional Kalman filter form. It usually relies on solving an optimization problem to minimize the estimation error in the worst case, thereby improving the robustness of the filter and providing robust performance under model uncertainty and external interference. The replacement result is shown in formula (23), (twenty three) In the formula, Adjust the covariance prediction formula and modify the weight of the observations, thereby affecting the subsequent Kalman filter gain and error covariance Calculation of the EKF improves the robustness of the EKF.
[0054] Optionally, inputting the normalized features into a long short-term memory network and outputting the position prediction information of the train includes: Inputting the normalized features into the long short-term memory network, and outputting instantaneous speed prediction information of the train; Based on the train position information, train travel direction information and instantaneous speed prediction information of the train at the previous moment, the position prediction information of the train at the current moment is predicted.
[0055] In the embodiment of the present application, there are two steps to obtain the final result output layer. The first step is to obtain the first goal of the model proposed in this section through the fully connected layer FCL processing, which is to predict the instantaneous speed of the train. Based on the LSTM layer result feature vector, extract the feature space mapping relationship, perform nonlinear transformation, adjust the digital display output dimension, and map the result to the output space. Figure 2 The schematic diagram of the working principle of the FCL layer provided for the implementation of this application is as follows: Figure 2 shown.
[0056] The second step is to calculate the instantaneous speed based on the instantaneous speed obtained in the previous step. , combined with the train position at the previous moment and And the driving direction d of the section at the previous moment can be known from the electronic map, and the latitude and longitude of the train at this moment are obtained through dead reckoning and , the calculation formulas are shown in (37) and (38), where E is the radius of the earth.
[0057] (37) (38) Optionally, a LSTM network is a special type of recurrent neural network architecture. Figure 3 This is a schematic diagram of the long short-term memory network model. Figure 3 As shown, Aokuo is used for the KF layer for data fusion, the LSTM layer for train speed prediction analysis, and the output layer for finally outputting the train location information.
[0058] The LSTM layer is designed to solve the gradient vanishing and gradient exploding problems of traditional RNN when processing long sequence data, as well as the limitation of difficulty in capturing long-term dependencies. LSTM effectively manages and transmits information by introducing gating units, including input gates, forget gates, and output gates, as well as memory units, thereby achieving the ability to better capture long-term dependencies. LSTM is widely used in various sequence data processing tasks, such as natural language processing, speech recognition, time series prediction, image processing, recommendation systems, bioinformatics, time series data processing, generative models, etc.
[0059] Memory Cell: The core of the LSTM network model is to store and transmit information through memory cells. Memory cells consist of a linear unit and a nonlinear unit. The linear unit is a long-term memory storage unit that can retain long-term dependencies. The nonlinear unit is a sigmoid function that is used to control the flow of information.
[0060] Input Gate: The input gate is used to control the input of information. It consists of a sigmoid function and a dot multiplication operation. The sigmoid function is used to convert the input information into a value between 0 and 1, and the dot multiplication operation is used to multiply the input information with the output of the sigmoid function. The input gate determines which information should be included in the memory unit based on the current input and the state of the previous moment.
[0061] Forget Gate, the forget gate is used to control the forgetting of information. It consists of a sigmoid function and a dot multiplication operation. The sigmoid function is used to convert the memory cell of the previous moment and the input of the current moment into a value between 0 and 1, and the dot multiplication operation is used to multiply the memory cell of the previous moment with the output of the sigmoid function. The forget gate determines which information should be forgotten based on the current input and the state of the previous moment.
[0062] Output Gate: The output gate is used to control the output of information. It consists of a sigmoid function and a dot multiplication operation. The sigmoid function is used to convert the current memory cell and the current input into a value between 0 and 1, and the dot multiplication operation is used to multiply the current memory cell with the output of the sigmoid function. The current input and the state of the previous moment determine how the information in the memory cell is output to the current state.
[0063] LSTM working steps 1) Initialization Initialize weights and bias , initialize the cell state and hidden state .
[0064] 2) Iteration Enter the input for the current time step and the hidden state at the previous time step .
[0065] Calculate the forget gate as shown in formula (31) (31) Calculate the input gate as shown in formula (32) (32) Calculate the cell state update as shown in formula (33) (33) Update the cell state as shown in formula (34) (34) Calculate the output gate as shown in formula (35) (35) Update the hidden state as shown in formula (36) (36) 3) Output Output the hidden state of the current time step .
[0066] In the embodiment of the present application, through field research on the operating environment and sensor usage of heavy-duty railway trains, the positioning prediction model designed by the present invention is more in line with actual work needs. This field research ensures that the model design can be targeted at specific operating conditions and challenges, thereby improving the applicability and effectiveness of the model.
[0067] Based on the LSTM positioning prediction model, the present invention combines the H∞ and EKF data fusion algorithms to optimize the data input of the prediction model. This combination utilizes the robustness of the H∞ filter in processing model uncertainty and non-Gaussian noise, and the ability of the EKF in processing nonlinear systems, thereby improving the accuracy of positioning prediction.
[0068] By applying the above data fusion algorithm, the present invention can more accurately predict the location information of the train. This improvement in accuracy is crucial for optimizing train scheduling and improving the safety of train operation.
[0069] The method of the present invention is not only applicable to trains under normal operating conditions, but can also maintain high prediction accuracy under different operating conditions such as acceleration, deceleration and cruising, which shows the strong adaptability and robustness of the model.
[0070] According to the simulation results, after adopting the optimization strategy of the present invention, the energy consumption of train operation is reduced, and the operation efficiency of the railway line is improved. This is of great significance for improving the overall efficiency and economy of railway transportation.
[0071] The train position prediction device provided by the present invention is described below. The train position prediction device described below and the train position prediction method described above can be referenced to each other.
[0072] Figure 4 A schematic diagram of the structure of a train position prediction device provided in an embodiment of the present application is shown in FIG. Figure 4 As shown, including: The cleaning module 410 is used to clean the original multi-dimensional train data information of the train to obtain multi-dimensional train data information; The fusion module 420 is used to perform data fusion on the multi-dimensional train data information based on the H∞ diffusion Kalman filter data fusion algorithm, and output fused train data; wherein the H∞ diffusion Kalman filter data fusion algorithm is a data fusion algorithm that replaces the covariance prediction formula of the Kalman filter with the covariance prediction formula of the H∞ filter; The output module 430 is used to perform characterization and normalization processing on the fused train data to obtain normalized features, and input the normalized features into the long short-term memory network to output the position prediction information of the train.
[0073] In the embodiment of the present application, the accuracy and reliability of train positioning are improved by using a data fusion algorithm based on H∞ diffusion Kalman filtering. This method achieves effective fusion of multi-dimensional train data information by evaluating the uncertainty of the system model and the observation model, and updating the state estimation and error covariance. In this way, more accurate fused train data can be output, providing strong support for train positioning.
[0074] Figure 5 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the train position prediction method, which includes: performing data cleaning on the original multi-dimensional train data information of the train to obtain the multi-dimensional train data information; Based on the H∞ diffusion Kalman filter data fusion algorithm, the multi-dimensional train data information is fused and the fused train data is output; wherein the H∞ diffusion Kalman filter data fusion algorithm is a data fusion algorithm that replaces the covariance prediction formula of the Kalman filter with the covariance prediction formula of the H∞ filter; The fused train data is characterized and normalized to obtain normalized features, and the normalized features are input into a long short-term memory network to output the position prediction information of the train.
[0075] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.
[0076] On the other hand, the present invention further provides a computer program product, the computer program product includes a computer program, the computer program can be stored in a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the train position prediction method provided by the above methods, the method comprising: after data cleaning the original multi-dimensional train data information of the train, obtaining the multi-dimensional train data information; Based on the H∞ diffusion Kalman filter data fusion algorithm, the multi-dimensional train data information is fused and the fused train data is output; wherein the H∞ diffusion Kalman filter data fusion algorithm is a data fusion algorithm that replaces the covariance prediction formula of the Kalman filter with the covariance prediction formula of the H∞ filter; The fused train data is characterized and normalized to obtain normalized features, and the normalized features are input into a long short-term memory network to output the position prediction information of the train.
[0077] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, the computer program being implemented when executed by a processor to execute the train position prediction method provided by the above methods, the method comprising: performing data cleaning on original multi-dimensional train data information of the train to obtain multi-dimensional train data information; Based on the H∞ diffusion Kalman filter data fusion algorithm, the multi-dimensional train data information is fused and the fused train data is output; wherein the H∞ diffusion Kalman filter data fusion algorithm is a data fusion algorithm that replaces the covariance prediction formula of the Kalman filter with the covariance prediction formula of the H∞ filter; The fused train data is characterized and normalized to obtain normalized features, and the normalized features are input into a long short-term memory network to output the position prediction information of the train.
[0078] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.
[0079] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A train position prediction method, characterized in that: include: After data cleaning of the original multi-dimensional train data information of the train, multi-dimensional train data information is obtained; Based on the H∞ diffusion Kalman filter data fusion algorithm, the multi-dimensional train data information is fused and the fused train data is output; wherein the H∞ diffusion Kalman filter data fusion algorithm is a data fusion algorithm that replaces the covariance prediction formula of the Kalman filter with the covariance prediction formula of the H∞ filter; The fused train data is characterized and normalized to obtain normalized features, and the normalized features are input into a long short-term memory network to output the position prediction information of the train.
2. The train position prediction method according to claim 1, characterized in that: Inputting the normalized features into a long short-term memory network and outputting the position prediction information of the train, including: Inputting the normalized features into the long short-term memory network, and outputting instantaneous speed prediction information of the train; Based on the train position information, train travel direction information and instantaneous speed prediction information of the train at the previous moment, the position prediction information of the train at the current moment is predicted.
3. The train position prediction method according to claim 1, characterized in that: The H∞ diffusion Kalman filter data fusion algorithm is used to fuse the multi-dimensional train data information and output fused train data, including: Initializing a state vector of a system model according to the odometer data information in the multi-dimensional train data information; Initializing an observation vector of an observation model according to the satellite positioning data information in the multidimensional train data information, and calculating an observation noise covariance matrix according to the observation vector; In each time step, the H∞ diffusion Kalman filter data fusion algorithm is used to evaluate the uncertainty of the system model and the observation model, and the state estimation and error covariance are updated to achieve data fusion of the multidimensional train data information and output fused train data.
4. The train position prediction method according to claim 1, characterized in that: After data cleaning of the original multi-dimensional train data information of the train, multi-dimensional train data information is obtained, including: Correcting or deleting abnormal values in the original multi-dimensional train data information; Average data filling is performed on the data missing values in the original multi-dimensional train data information; wherein the average data is determined based on the average value of the data adjacent to the missing data value.
5. The train position prediction method according to claim 1, characterized in that: The fused train data is characterized and normalized to obtain normalized features, including: After respectively performing time feature extraction, position feature conversion, periodic encoding of travel direction and speed feature processing on the fused train data, normalization processing is performed to obtain normalized features; Wherein, the time feature extraction includes: relative time feature extraction and period feature extraction; the position feature conversion includes: differential feature extraction and polar coordinate conversion.
6. A train position prediction device, characterized in that: include: A cleaning module is used to clean the original multi-dimensional train data information of the train to obtain multi-dimensional train data information; A fusion module is used to perform data fusion on the multi-dimensional train data information based on an H∞ diffusion Kalman filter data fusion algorithm, and output fused train data; wherein the H∞ diffusion Kalman filter data fusion algorithm is a data fusion algorithm that replaces the covariance prediction formula of the Kalman filter with the covariance prediction formula of the H∞ filter; The output module is used to perform characterization and normalization processing on the fused train data to obtain normalized features, and input the normalized features into a long short-term memory network to output the position prediction information of the train.
7. The train position prediction device according to claim 6, characterized in that: The device is also used for: Inputting the normalized features into the long short-term memory network, and outputting instantaneous speed prediction information of the train; Based on the train position information, train travel direction information and instantaneous speed prediction information of the train at the previous moment, the position prediction information of the train at the current moment is predicted.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the train position prediction method as described in any one of claims 1 to 5 is implemented.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the train position prediction method as described in any one of claims 1 to 5 is implemented.
10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the train position prediction method as described in any one of claims 1 to 5 is implemented.