Adaptive high-precision timing method and system based on deep learning

Abstract

The present invention discloses an adaptive high-precision timing method and system based on deep learning, which monitors low-frequency time code signals and Beidou signals collected in real time by user equipment and saves key parameters of the low-frequency time code signal; when the low-frequency time code signal and the Beidou signal are both normal, or when the low-frequency time code signal is normal and the Beidou signal is abnormal, the quality of the low-frequency time code signal is scored using a signal quality evaluation model and key parameters; when the scoring result is not less than a preset value, the timing result corresponding to the low-frequency time code signal is compensated based on a timing error predicted by a long short-term memory model with attention and the key parameters, and the timing is completed through a low-frequency time code timing mode using the compensated timing result; when the low-frequency time code signal is normal and the Beidou signal is abnormal and the scoring result is less than a preset value, the timing result corresponding to the low-frequency time code signal is compensated using the timing error and a window sliding average, and the timing is completed through the low-frequency time code timing mode using the compensated timing result.

Description

Technical Field

[0001] The present invention belongs to the field of time and frequency technology, and relates to but is not limited to an adaptive high-precision timing method and system based on deep learning. Background Art

[0002] As the modern information society develops towards intelligence and ubiquity, high-precision timing technology has become a core technology supporting critical infrastructure such as mobile communications, electricity, finance, the industrial internet, and navigation. The Beidou Navigation Satellite System (BDS) and the low-frequency time code (BPC) timing system are two representative timing methods, each with its own unique advantages. The Beidou Navigation Satellite System, with its high accuracy and global coverage, provides stable, high-precision timing services in open areas and good weather. However, in indoor environments, inclement weather, or heavily obstructed conditions, signal degradation or even interruption can occur, significantly reducing timing accuracy or even rendering the system inoperable. The low-frequency time code timing system, leveraging the low propagation loss of long-wave signals, offers a certain degree of stability in complex environments. Compared to the Beidou Navigation Satellite System, it is less susceptible to weather influences and has stronger anti-interference capabilities. However, its timing accuracy still lags behind that of the Beidou Navigation Satellite System.

[0003] In related technologies, high-precision timing is performed through a low-frequency time code timing system or a Beidou satellite navigation system. However, this method is difficult to meet the current high-precision timing requirements in complex and changing environments.

[0004] Therefore, how to meet the demand for high-precision time synchronization in modern society has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the embodiments of the present invention provide an adaptive high-precision timing method and system based on deep learning, which at least solves the problem that related technologies cannot meet the needs of modern society for high-precision timing.

[0006] According to a first aspect of an embodiment of the present invention, an adaptive high-precision timing method based on deep learning is provided, comprising:

[0007] Monitor low-frequency time code signals and BeiDou signals collected in real time by user equipment, and save key parameters of the low-frequency time code signals during the monitoring process; the key parameters include signal strength, carrier phase, amplitude modulation, and signal-to-noise ratio;

[0008] When the monitored low-frequency time code signal and the Beidou signal are both normal, or when the monitored low-frequency time code signal is normal and the Beidou signal is abnormal, scoring the quality of the low-frequency time code signal based on the signal quality assessment model and the key parameters to obtain a first scoring result;

[0009] When the first scoring result is not less than a preset value, compensating the timing result corresponding to the low-frequency time code signal based on the long short-term memory model with attention and the timing error predicted by the key parameters to obtain a first target timing result, and using the first target timing result to complete timing of the user equipment through a low-frequency time code timing mode;

[0010] When the monitored low-frequency time code signal is normal but the Beidou signal is abnormal and the first scoring result is less than a preset value, the timing result corresponding to the low-frequency time code signal is compensated by the timing error and the window sliding average to obtain a second target timing result, and the second target timing result is used to complete the timing of the user equipment through the low-frequency time code timing mode.

[0011] According to a second aspect of an embodiment of the present invention, there is provided an adaptive high-precision timing system based on deep learning, comprising:

[0012] The signal acquisition and monitoring module is used to monitor the low-frequency time code signal and Beidou signal collected in real time by the user equipment, and save the key parameters of the low-frequency time code signal during the monitoring process; the key parameters include signal field strength, carrier phase, amplitude modulation, and signal-to-noise ratio;

[0013] a low-frequency time code signal quality assessment module, configured to score the quality of the low-frequency time code signal based on a signal quality assessment model and the key parameters to obtain a first scoring result when both the monitored low-frequency time code signal and the Beidou signal are normal, or when the monitored low-frequency time code signal is normal and the monitored Beidou signal is abnormal;

[0014] a timing error prediction and compensation module, configured to, when the first scoring result is not less than a preset value, compensate the timing result corresponding to the low-frequency time code signal based on the timing error predicted by the long short-term memory model with attention and the key parameters to obtain a first target timing result;

[0015] The timing error prediction and compensation module is further configured to, when the monitored low-frequency time code signal is normal but the Beidou signal is abnormal and the first scoring result is less than a preset value, compensate the timing result corresponding to the low-frequency time code signal by using the timing error and the window sliding average to obtain a second target timing result;

[0016] A dynamic timing decision and execution module is used to complete the timing of the user equipment through the low-frequency time code timing mode using the first target timing result or to complete the timing of the user equipment through the low-frequency time code timing mode using the second target timing result.

[0017] According to a third aspect of an embodiment of the present invention, an electronic device is provided, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; the memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform an operation corresponding to the method described in the first aspect or the second aspect.

[0018] According to a fourth aspect of an embodiment of the present invention, a computer storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the method according to the first aspect or the second aspect is implemented.

[0019] According to the solution provided by the embodiment of the present invention, the low-frequency time code signal and the Beidou signal collected in real time by the user equipment are monitored, and the key parameters of the low-frequency time code signal in the monitoring process are saved; the key parameters include signal field strength, carrier phase, amplitude modulation and signal-to-noise ratio; when the monitored low-frequency time code signal and the Beidou signal are both normal, or when the monitored low-frequency time code signal is normal and the Beidou signal is abnormal, the quality of the low-frequency time code signal is scored based on the signal quality evaluation model and the key parameters to obtain a first scoring result; when the first scoring result is not less than a preset value, the long-term and short-term evaluation based on attention is performed. The timing error predicted by the memory model and the key parameters is used to compensate the timing result corresponding to the low-frequency time code signal to obtain the first target timing result, and the first target timing result is used to complete the timing of the user device through the low-frequency time code timing mode; when the monitored low-frequency time code signal is normal and the Beidou signal is abnormal and the first scoring result is less than the preset value, the timing result corresponding to the low-frequency time code signal is compensated by the timing error and the window sliding average to obtain the second target timing result, and the second target timing result is used to complete the timing of the user device through the low-frequency time code timing mode. In this process, by introducing a long-short-term memory model with attention, a signal quality scoring model, and a hierarchical compensation strategy, this method can intelligently switch the timing mode based on the real-time status of the BDS signal and the BPC signal, and when the BDS signal is unavailable, it can achieve precise regulation of error compensation based on the signal quality of the BPC, significantly improving the timing accuracy and stability in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive work, among which:

[0021] Figure 1A flowchart of an adaptive high-precision timing system and method based on deep learning provided by an embodiment of the present invention;

[0022] Figure 2 A framework diagram of a long short-term memory model with attention provided by an embodiment of the present invention;

[0023] Figure 3 An architectural diagram of an adaptive high-precision timing system based on deep learning provided by an embodiment of the present invention;

[0024] Figure 4 The figure is a schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0026] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0027] It should be pointed out that the terms "first\second\third" involved in the embodiments of the present invention are only used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present invention described here can be implemented in an order other than that illustrated or described here.

[0028] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art in the art to which the embodiments of the present invention pertain. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and, unless specifically defined as herein, should not be interpreted in an idealized or overly formal sense.

[0029] Figure 1This is a flow chart of an adaptive high-precision timing system and method based on deep learning provided in an embodiment of the present invention. The adaptive high-precision timing system and method based on deep learning provided in an embodiment of the present invention can be executed by an electronic device, such as a computer, a server, etc.

[0030] like Figure 1 As shown, the adaptive high-precision timing system and method based on deep learning include:

[0031] S101. Monitor the low-frequency time code signal and Beidou signal collected in real time by the user equipment, and save the key parameters of the low-frequency time code signal during the monitoring process; the key parameters include signal field strength, carrier phase, amplitude modulation, and signal-to-noise ratio.

[0032] In an embodiment of the present invention, the user equipment collects Beidou signals (BDS signals) and low-frequency time code signals (BPC signals) in real time, monitors the collected low-frequency time code signals and Beidou signals, and saves the low-frequency time code signals including signal field strength during the monitoring process. E , carrier phase , amplitude modulation m and signal-to-noise ratio SNR The signal strength is a key parameter used to measure the timing error of low-frequency time code signals. The changing trend of the signal field strength can effectively reflect the reception status of user devices within the signal coverage area. The rate of change of the carrier phase over a certain period of time can reflect the stability of the BPC broadcasting system. The amplitude modulation, reflected by the depth of the negative pulse dive of the pulse width modulation signal, can reflect the correctness of the broadcast signal. The signal-to-noise ratio represents the ratio of signal power to noise, reflecting the clarity of the signal in a noisy environment and can, to a certain extent, reflect the reliability of the timing signal. For example, a high signal-to-noise ratio is conducive to accurate timing, while a low signal-to-noise ratio increases the difficulty of demodulation.

[0033] S102. When the monitored low-frequency time code signal and the Beidou signal are both normal, or when the monitored low-frequency time code signal is normal and the Beidou signal is abnormal, score the quality of the low-frequency time code signal based on the signal quality assessment model and key parameters to obtain a first scoring result.

[0034] In an embodiment of the present invention, when both the monitored low-frequency time code signal and the Beidou signal are normal, or when the monitored low-frequency time code signal is normal but the monitored Beidou signal is abnormal, a signal quality assessment model for the BPC signal based on deep learning can be designed. The quality of the low-frequency time code signal is scored using the signal quality assessment model and key parameters to obtain a scoring result. This score is used to coordinate the compensation strategy after adjustment and enhance the robustness of the timing system. A higher score indicates a more reliable BPC signal. When the score falls below a threshold, an alarm is triggered or the timing mode is switched.

[0035] Furthermore, when the monitored low-frequency time code signal and Beidou signal are both normal, the actual error is calculated and saved using the low-frequency time code signal and Beidou signal. This error is then used to train a long short-term memory (LSTM) model with attention, namely the LSTM-Attention hybrid model.

[0036] S103. When the first scoring result is not less than a preset value, the timing result corresponding to the low-frequency time code signal is compensated based on the timing error predicted by the long short-term memory model with attention and the key parameters to obtain a first target timing result, and the first target timing result is used to complete the timing of the user equipment through the low-frequency time code timing mode.

[0037] In an embodiment of the present invention, the timing error refers to the timing error between the BDS timing result and the BPC timing result. Because the timing accuracy of BDS is higher than that of BPC, the timing error between the BPC timing result and the BDS timing result is predicted and the timing result of BPC is compensated, so that BPC can also achieve a timing accuracy close to that of BDS. The first scoring result is compared with the preset value. When the first scoring result is not less than the preset value, the key parameters are input into the trained LSTM model with attention to obtain the timing error of the low-frequency time code signal during the transmission process, and the timing result obtained from the low-frequency time code signal is compensated by the timing error to obtain the compensated accurate first target timing result. Finally, the first target timing result is used to complete the timing of the user device through the low-frequency time code timing mode to ensure that it is consistent with the time of the low-frequency time code system.

[0038] A hybrid LSTM (Long Short-Term Memory-Attention Neural Network) model with attention was selected as the deep learning model. LSTM is a special recurrent temporal network with memory cells and a gating mechanism, effectively handling long-term dependencies in time series data. Furthermore, the attention mechanism allows the model to dynamically focus on different parts of the input, thereby improving its attention to key information. For example, key parameters (such as sudden changes in carrier phase and sudden drops in field strength) may have a greater impact on timing error during certain time periods. Traditional LSTMs cannot effectively capture this information at these critical time points. By introducing the attention mechanism, the LSTM model with attention can automatically learn weights for different time steps, weightedly fusing the hidden states across all time steps, thereby improving prediction accuracy. Therefore, the LSTM model with attention is more suitable for handling the complex relationship between the time-varying characteristics of key BPC parameters and timing error.

[0039] In an embodiment of the present invention, the LSTM model with attention is divided into two stages: training and compensation:

[0040] During the training phase, we select the conditions where both BeiDou satellite and BPC signals are good and continuously collect a large amount of data. The key parameters of BPC signal quality monitoring are obtained by arranging them in time series. 、BDS timing result T BD (t) and BPC timing result T BPC Error of (t) , which is used as a training sample for the LSTM model with attention; As input to the LSTM model with attention, As the ideal output of the LSTM model with attention, the purpose of training is to enable the model to master the mapping relationship between input and output, that is, to construct the function , so that it approaches the one that satisfies the following relationship ,in, a Refers to a specific moment, which is a variable value, such as the first second, second or third second.

[0041] ;

[0042] in, is the trained LSTM model with attention.

[0043] Among them, when training the LSTM model with attention, the mean square error ( MSE ) and attention weight sparse constraints The sum of the two is used as the loss function, denoted as L, and the specific formula is as follows:

[0044] ;

[0045] ;

[0046] ;

[0047] In the above formula, is the BPC signal quality data sequence at time t The attention weights corresponding to different features in y i It is i The actual timing error of the key parameters, It is i The predicted timing error of the key parameters, n is the sum of the data points, is the mean square error, is the attention weight sparse constraint, is the loss function.

[0048] Compensation stage: The BPC signal quality features in the current time window (60s) are used as the input of the LSTM model with attention. The output of the LSTM model with attention is calculated as t Compensation value of BPC timing result at the moment :

[0049] ;

[0050] The default value can be 0.5. In actual use, it defines the dynamic quality score of the BPC signal. When the signal is high quality, The low-quality signal is used to dynamically adjust the LSTM model with attention to compensate for the BPC timing error.

[0051] S104. When the monitored low-frequency time code signal is normal and the Beidou signal is abnormal and the first scoring result is less than a preset value, the timing result corresponding to the low-frequency time code signal is compensated by the timing error and the window sliding average to obtain a second target timing result, and the second target timing result is used to complete the timing of the user equipment through the low-frequency time code timing mode.

[0052] In an embodiment of the present invention, a sliding time window mechanism can collect input data within a preset time period and input the collected input data into an attention-based LSTM model to obtain the timing error. The window sliding average refers to the average of all timing error predictions within a given time window (e.g., 60 seconds). After obtaining the timing error and the window sliding average, the timing result obtained from the low-frequency time code signal is compensated using the timing error and the window sliding average to obtain a second, accurate, compensated target timing result. Finally, the second target timing result is used to complete the timing of the user device through the low-frequency time code timing mode, ensuring that it remains consistent with the time of the low-frequency time code system.

[0053] It can be understood that in an embodiment of the present invention, the low-frequency time code signal and the Beidou signal collected in real time by the user equipment are monitored, and the key parameters of the low-frequency time code signal during the monitoring process are saved; the key parameters include signal field strength, carrier phase, amplitude modulation and signal-to-noise ratio; when the monitored low-frequency time code signal and the Beidou signal are normal, or when the monitored low-frequency time code signal is normal and the Beidou signal is abnormal, the quality of the low-frequency time code signal is scored based on the signal quality assessment model and the key parameters to obtain a first scoring result; when the first scoring result is not less than a preset value, the timing result corresponding to the low-frequency time code signal is compensated based on the timing error predicted by the LSTM model with attention and the key parameters to obtain a first target timing result, and the first target timing result is used to complete the timing of the user equipment through the low-frequency time code timing mode;

[0054] When the monitored low-frequency time code signal is normal but the Beidou signal is abnormal and the first scoring result is less than the preset value, the timing result corresponding to the low-frequency time code signal is compensated by using the timing error and the window sliding average to obtain the second target timing result. This second target timing result is then used to complete the timing of the user device using the low-frequency time code timing mode. In this process, by introducing an LSTM model with attention, a signal quality scoring model, and a hierarchical compensation strategy, this method can intelligently switch the timing mode based on the real-time status of the BDS and BPC signals. When the BDS signal is unavailable, the error compensation is precisely adjusted based on the BPC signal quality, significantly improving the timing accuracy and stability in complex environments.

[0055] In the embodiment of the present invention, S101 further includes S105 to S107, which are described through the following steps.

[0056] S105. When the monitored low-frequency time code signal is abnormal and the Beidou signal is normal, or when the monitored low-frequency time code signal and the Beidou signal are both normal, score the quality of the low-frequency time code signal based on the signal quality assessment model and key parameters to obtain a second scoring result.

[0057] S106: When the second scoring result is less than a preset value, the user equipment is time-synchronized using the Beidou signal timing mode.

[0058] In some embodiments of the present invention, when the monitored low-frequency time code signal is abnormal and the Beidou signal is normal, or when both the monitored low-frequency time code signal and the Beidou signal are normal, key parameters are input into a signal quality assessment model to score the quality of the low-frequency time code signal, obtaining a second scoring result. When the second scoring result is less than a preset value, timing of the user equipment is performed using the Beidou signal timing mode. When both the monitored low-frequency time code signal and the Beidou signal are normal, the second scoring result is equal to the first scoring result.

[0059] S107. When the monitored low-frequency time code signal and Beidou signal are both abnormal, alarm information is continuously issued.

[0060] In some embodiments of the present invention, when both the monitored low-frequency time code signal and the Beidou signal are abnormal, alarms are issued continuously. These alarms alert users to potential unreliability of current data, preventing safety risks associated with erroneous positioning information. Promptly detecting and reporting signal anomalies helps maintenance personnel quickly troubleshoot and repair the problem.

[0061] In some embodiments of the present invention, scoring the quality of the low-frequency time code signal based on the signal quality assessment model and key parameters to obtain the first scoring result in S1022 can be achieved by S1021, which is explained in the following steps.

[0062] S1021. Use the input layer to input the key parameters into the bidirectional gated recurrent layer to obtain a merged feature vector; and input the merged feature vector into the fully connected layer to obtain a first scoring result.

[0063] In some embodiments of the present invention, the signal quality assessment model can be a bidirectional gated recurrent model; the bidirectional gated recurrent model includes an input layer, a bidirectional gated recurrent layer, and a fully connected layer. The input layer is used to convert the key parameters into a sequence. The data is fed into a bidirectional gated recurrent layer in the form of . The bidirectional gated recurrent layer processes the key input parameters from the beginning to the end of the sequence, generating a hidden state for each time step in the sequence. Simultaneously, the bidirectional gated recurrent layer processes the same input data from the end to the beginning of the sequence, also generating a hidden state for each time step. For each time step in the sequence, the bidirectional gated recurrent layer generates two hidden states—one from the forward processing and the other from the backward processing. These hidden states are typically merged in one or more ways to form the final merged feature vector. The fully connected layer maps the merged feature vector to a single-value score, resulting in the first score.

[0064] Specifically, the input layer: receives the time series data sequence of the BPC signal The initial design time window is 30s and the step size is 1s; Bidirectional gated recurrent layer (Bi-GRU layer): 1 layer of bidirectional GRU, 64 neurons in each direction; Fully connected layer: maps the output of the bidirectional gated recurrent layer to a single value score , select Sigmoid as the activation function, and the output range is [0,1]. When training the bidirectional gated loop model, the BDS timing result T BD (t) and BPC timing result T BPC Error of (t) To classify. Define the sample data is the maximum value hour, ;and is the minimum value hour, ; intermediate values are scored using linear interpolation.

[0065] In some embodiments of the present invention, the timing error predicted based on the long short-term memory model with attention and key parameters in S103 compensates the timing result corresponding to the low-frequency time code signal, and before obtaining the first target timing result, it also includes S10 to S13, which is explained through the following steps.

[0066] S10, normalize the key parameters to obtain normalized data; and input the normalized data into the three-layer long short-term memory layer through the input layer to obtain the first latent feature vector. In some embodiments of the present invention, the key parameters are recorded as a time series data sequence , because the value ranges and dimensions of different parameters vary greatly, it is necessary to normalize the parameter data to the [0,1] interval. The following formula is used to normalize the key parameters:

[0067] ;

[0068] In the above formula, is the normalized data, is the original key parameter, and are the minimum and maximum values of key parameters respectively, to improve the model training effect and convergence speed.

[0069] Among them, the LSTM layer mainly captures the time series dependency between input features. Unlike the traditional method, the present invention is a three-layer LSTM cascade structure, that is, a three-layer stacked LSTM is set, the output of the previous layer is used as the input of the next layer, the number of neurons in each layer increases, and the initial number is set to 64, 128, and 256 neurons respectively, gradually extracting deep temporal features, and adding residual connections after each LSTM layer to alleviate the gradient vanishing problem. The specific formula is as follows:

[0070] ;

[0071] in, Represents the number of layers, Representative l The hidden state output of the layer, Representative l The hidden state of the layer is processed by the LSTM unit, which can enhance the training stability and feature extraction ability of the deep network, and finally output the hidden state of all time steps H , that is, the first latent eigenvector is obtained.

[0072] S12. Input the first latent feature vector into the attention layer to obtain the second latent feature vector.

[0073] S13. Use the fully connected layer to extract and combine the second latent feature vector to obtain a third latent feature vector, and input the third latent feature vector into the output layer to obtain the timing error.

[0074] In some embodiments of the present invention, after the first latent feature vector is input into the attention layer, linear transformation, attention score calculation, attention distribution calculation, and weighted summation are performed to obtain a second latent feature vector. The second latent feature vector is input into the fully connected layer for feature extraction and combination to obtain a third latent feature vector, which is then input into the output layer to obtain the timing error.

[0075] Among them, the LSTM layer and the attention layer also include a dropout layer (Dropout layer). The Dropout layer randomly sets the output of some neurons to zero (that is, "drops" these neurons), thereby forcing the LSTM model with attention to learn more robust features. It is mainly activated during the training of the LSTM model with attention. When the LSTM model with attention is trained, the input of the attention layer is the output of the LSTM layer. By calculating the weighted sum of the input features and the weight matrix, the importance score of each feature is obtained, and the relative contribution between the features is adjusted accordingly; then the Dropout layer is added to prevent overfitting; finally, the fully connected layer is added to Mapping to error prediction value, realizing the conversion from feature to prediction value, outputting the prediction value of BPC timing error. The specific formula is as follows:

[0076] ;

[0077] ;

[0078] in, Represents the BPC signal quality data sequence at time t The attention weights corresponding to different features in ; represents the output of the LSTM layer at time t; 、 is a trainable parameter; b is a bias term; is the output of the attention layer at time t. This vector integrates the information of all time steps in the time window, but focuses on the time periods with higher weights.

[0079] In some embodiments of the present invention, S103 may be implemented through S1031, which is explained through the following steps.

[0080] S1031. Obtain a first target timing result through the timing error, the timing result, and the first target timing result formula.

[0081] In some embodiments of the present invention, the timing error and the timing result are substituted into a first target timing result formula to obtain the first target timing result.

[0082] The formula for the first target timing result is as follows:

[0083] ;

[0084] In the above formula, The timing result for the first target is: for t The timing result corresponding to the low-frequency time code signal at the moment, for t Timing error at the moment.

[0085] In some embodiments of the present invention, S104 can be implemented through S1041, which is explained through the following steps.

[0086] S1041. Obtain a second target timing result through the timing error, the timing result, the window sliding average, and the second target timing result formula.

[0087] In some embodiments of the present invention, the timing error and the window sliding average are substituted into the second target timing result formula to obtain the second target timing result.

[0088] The formula for the second target timing result is as follows:

[0089] ;

[0090] In the above formula, Timing result for the second target, is the average value of the error prediction value in the sliding window corresponding to time t.

[0091] In an embodiment of the present invention, Figure 2 As shown, Figure 2 The framework diagram of the LSTM model with attention provided by the embodiment of the present invention. Figure 2 The LSTM model with attention includes the input layer, LSTM layer, attention layer and output layer. The LSTM layer (long short-term memory) includes the LSTM layer and the first dropout layer, and the attention layer includes the attention layer, the second dropout layer and the fully connected layer. , amplitude modulation m, and signal-to-noise ratio (SNR). The first and second dropout layers are dropout layers, which are activated when training the LSTM model with attention. Specifically, when training the LSTM model with attention, input data is fed into the input layer, processed by the LSTM layer, the first dropout layer, the attention layer, the second dropout layer, and the fully connected layer, before the timing error is obtained at the output layer. When applying the LSTM model with attention, input data is fed into the input layer, processed by the LSTM layer, the attention layer, the fully connected layer, and the timing error is obtained at the output layer. The LSTM layer is a three-layer stacked LSTM layer, meaning there are three sequentially connected LSTM layers.

[0092] In an embodiment of the present invention, Figure 3 As shown, Figure 3 This is an architecture diagram of an adaptive high-precision timing system based on deep learning provided by an embodiment of the present invention. Figure 3 The adaptive high-precision timing system based on deep learning includes:

[0093] The signal acquisition and monitoring module is used to monitor the low-frequency time code signal and Beidou signal collected in real time by the user equipment, and save the key parameters of the low-frequency time code signal during the monitoring process; the key parameters include signal field strength, carrier phase, amplitude modulation, and signal-to-noise ratio;

[0094] a low-frequency time code signal quality assessment module, configured to score the quality of the low-frequency time code signal based on a signal quality assessment model and the key parameters to obtain a first scoring result when both the monitored low-frequency time code signal and the Beidou signal are normal, or when the monitored low-frequency time code signal is normal and the monitored Beidou signal is abnormal;

[0095] a timing error prediction and compensation module, configured to, when the first scoring result is not less than a preset value, compensate the timing result corresponding to the low-frequency time code signal based on the timing error predicted by the long short-term memory model with attention and the key parameters to obtain a first target timing result;

[0096] The timing error prediction and compensation module is further configured to, when the monitored low-frequency time code signal is normal but the Beidou signal is abnormal and the first scoring result is less than a preset value, compensate the timing result corresponding to the low-frequency time code signal by using the timing error and the window sliding average to obtain a second target timing result;

[0097] A dynamic timing decision and execution module is used to complete the timing of the user equipment through the low-frequency time code timing mode using the first target timing result or to complete the timing of the user equipment through the low-frequency time code timing mode using the second target timing result.

[0098] The system also includes a pre-processing module for normalizing the collected signals.

[0099] Reference Figure 4 , shows a schematic structural diagram of an electronic device according to an embodiment of the present invention. The specific embodiment of the present invention does not limit the specific implementation of the electronic device.

[0100] like Figure 4 As shown, the electronic device may include: a processor (processor) 502 , a communication interface (Communications Interface) 504 , a memory (memory) 506 , and a communication bus 508 .

[0101] in:

[0102] The processor 502 , the communication interface 504 , and the memory 506 communicate with each other via a communication bus 508 .

[0103] The communication interface 504 is used to communicate with other electronic devices or servers.

[0104] The processor 502 is configured to execute the program 510 , and specifically may execute the relevant steps in the above method embodiment.

[0105] Specifically, the program 510 may include program codes, which include computer operation instructions.

[0106] Processor 502 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in a smart device may be of the same type, such as one or more CPUs, or different types, such as one or more CPUs and one or more ASICs.

[0107] The memory 506 is used to store the program 510. The memory 506 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage.

[0108] The program 510 may be specifically configured to enable the processor 502 to execute operations corresponding to the methods described in the above method embodiments.

[0109] The specific implementation of each step in program 510 can be found in the corresponding descriptions of the corresponding steps and units in the above-mentioned method embodiments, and will not be repeated here. Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working processes of the above-mentioned devices and modules can refer to the corresponding process descriptions in the above-mentioned method embodiments, and will not be repeated here.

[0110] It should be pointed out that, according to the needs of implementation, the various components / steps described in the embodiments of the present invention can be split into more components / steps, or two or more components / steps or partial operations of components / steps can be combined into new components / steps to achieve the purpose of the embodiments of the present invention.

[0111] The methods according to the embodiments of the present invention described above can be implemented in hardware, firmware, or as software or computer code that can be stored on a recording medium (such as a CD ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code originally stored on a remote recording medium or non-transitory machine-readable medium downloaded over a network and then stored on a local recording medium. Thus, the methods described herein can be processed by such software stored on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It will be understood that a computer, processor, microprocessor controller, or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods described herein are implemented. Furthermore, when a general-purpose computer accesses the code for implementing the methods described herein, the execution of the code transforms the general-purpose computer into a dedicated computer for performing the methods described herein.

[0112] Those skilled in the art will appreciate that the units and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the embodiments of the present invention.

[0113] The above implementation methods are only used to illustrate the embodiments of the present invention, and are not intended to limit the embodiments of the present invention. Ordinary technicians in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the embodiments of the present invention. Therefore, all equivalent technical solutions also fall within the scope of the embodiments of the present invention, and the scope of patent protection of the embodiments of the present invention should be defined by the claims.

Claims

1. An adaptive high-precision timing method based on deep learning, characterized in that: include: Monitor low-frequency time code signals and BeiDou signals collected in real time by user equipment, and save key parameters of the low-frequency time code signals during the monitoring process; the key parameters include signal strength, carrier phase, amplitude modulation, and signal-to-noise ratio; When the monitored low-frequency time code signal and the Beidou signal are both normal, or when the monitored low-frequency time code signal is normal and the Beidou signal is abnormal, scoring the quality of the low-frequency time code signal based on the signal quality assessment model and the key parameters to obtain a first scoring result; When the first scoring result is not less than a preset value, compensating the timing result corresponding to the low-frequency time code signal based on the long short-term memory model with attention and the timing error predicted by the key parameters to obtain a first target timing result, and using the first target timing result to complete timing of the user equipment through a low-frequency time code timing mode; When the monitored low-frequency time code signal is normal and the Beidou signal is abnormal and the first scoring result is less than a preset value, compensating the timing result corresponding to the low-frequency time code signal by using the timing error and the window sliding average to obtain a second target timing result, and using the second target timing result to complete timing of the user equipment through the low-frequency time code timing mode; When the monitored low-frequency time code signal and the Beidou signal are both normal and the first scoring result is less than the preset value, completing timing for the user equipment through the Beidou signal timing mode; When the monitored low-frequency time code signal is abnormal and the Beidou signal is normal, scoring the quality of the low-frequency time code signal based on the signal quality assessment model and key parameters to obtain a second scoring result; When the second scoring result is less than the preset value, completing timing for the user equipment through the Beidou signal timing mode; When the monitored low-frequency time code signal and the Beidou signal are both abnormal, alarm information is continuously issued.

2. The method according to claim 1, characterized in that The signal quality assessment model is a bidirectional gated recurrent model; the bidirectional gated recurrent model includes an input layer, a bidirectional gated recurrent layer, and a fully connected layer; Scoring the quality of the low-frequency time code signal based on the signal quality assessment model and the key parameters to obtain a first scoring result includes: The key parameters are input into the bidirectional gated recurrent layer using the input layer to obtain a merged feature vector; and the merged feature vector is input into the fully connected layer to obtain the first scoring result.

3. The method according to claim 1, characterized in that The attention-based long short-term memory model includes an input layer, three long short-term memory layers, an attention layer, a fully connected layer, and an output layer; the three long short-term memory layers are cascaded; Before compensating the timing result corresponding to the low-frequency time code signal based on the timing error predicted by the long short-term memory model with attention and the key parameters to obtain the first target timing result, the method further includes: Normalizing the key parameters to obtain normalized data; and inputting the normalized data into the three-layer long short-term memory layer through the input layer to obtain a first latent feature vector; Inputting the first latent feature vector into the attention layer to obtain a second latent feature vector; The second latent feature vector is extracted and combined using the fully connected layer to obtain a third latent feature vector, and the third latent feature vector is input into the output layer to obtain the timing error.

4. The method according to claim 1, wherein The timing error predicted based on the long short-term memory model with attention and the key parameters is used to compensate the timing result corresponding to the low-frequency time code signal to obtain a first target timing result, including: The first target timing result is obtained by using the timing error, the timing result and the first target timing result formula; the first target timing result formula is as follows: ； In the above formula, is the timing result of the first target, for t The timing result corresponding to the low-frequency time code signal at the moment, for t The timing error at the moment, BPC is the low-frequency time code signal.

5. The method according to claim 4, characterized in that The compensating the timing result corresponding to the low-frequency time code signal by using the timing error and the window sliding average to obtain a second target timing result includes: The second target timing result is obtained by using the timing error, the timing result, the window sliding average and the second target timing result formula; the second target timing result formula is as follows: ； In the above formula, is the second target timing result, for t The window sliding average corresponding to the moment, the window sliding average is the average value of the timing error prediction value within the sliding window.

6. An adaptive high-precision timing system based on deep learning, characterized in that: include: The signal acquisition and monitoring module is used to monitor the low-frequency time code signal and Beidou signal collected in real time by the user equipment, and save the key parameters of the low-frequency time code signal during the monitoring process; the key parameters include signal field strength, carrier phase, amplitude modulation, and signal-to-noise ratio; a low-frequency time code signal quality assessment module, configured to score the quality of the low-frequency time code signal based on a signal quality assessment model and the key parameters to obtain a first scoring result when both the monitored low-frequency time code signal and the Beidou signal are normal, or when the monitored low-frequency time code signal is normal and the monitored Beidou signal is abnormal; a timing error prediction and compensation module, configured to, when the first scoring result is not less than a preset value, compensate the timing result corresponding to the low-frequency time code signal based on the timing error predicted by the long short-term memory model with attention and the key parameters to obtain a first target timing result; The timing error prediction and compensation module is further configured to, when the monitored low-frequency time code signal is normal but the Beidou signal is abnormal and the first scoring result is less than a preset value, compensate the timing result corresponding to the low-frequency time code signal by using the timing error and the window sliding average to obtain a second target timing result; A dynamic timing decision and execution module is used to complete the timing of the user equipment through the low-frequency time code timing mode using the first target timing result or to complete the timing of the user equipment through the low-frequency time code timing mode using the second target timing result.

Images