Hydrogen clock error forecasting method, system and device based on improved recurrent neural network and storage medium

Abstract

According to the hydrogen clock error forecasting method, system and device based on the improved recurrent neural network and the storage medium, the memory characteristic of the recurrent neural network is used for representing the time relation, so that hydrogen clock error forecasting is carried out, and the accuracy of hydrogen clock error forecasting is improved. The method comprises the following steps: acquiring original clock error data of a hydrogen clock, and preprocessing the original clock error data to obtain preprocessed clock error data; constructing a hybrid recurrent neural network model, and initializing the hybrid recurrent neural network model; dividing the preprocessed clock error data into a training set, a verification set and a test set; based on the training set, the verification set and the test set, using an RAdam optimizer and MAPE as loss functions, and optimizing the hybrid recurrent neural network model until a preset precision is reached; and preprocessing the clock error data of the to-be-predicted time period, and inputting the preprocessed clock error data of the to-be-predicted time period into the optimized hybrid recurrent neural network model to obtain a predicted clock error value.

Description

Technical Field

[0001] The present application relates to the field of atomic clock time scales, and specifically to a hydrogen clock clock error prediction method, system, device and storage medium based on an improved recurrent neural network. Background Art

[0002] Hydrogen clocks are generally selected as the master clocks of the timekeeping clock group because of their good short-term stability. The Coordinated Universal Time announced by the International Bureau of Weights and Measures has a lag and cannot control local signals in real time. The timekeeping laboratory can only use the results of the hydrogen clock clock error prediction to provide an accurate and reliable time scale. Therefore, the accuracy of the clock error prediction directly affects the real-time control of the master clock and determines the generation of high-quality signals.

[0003] Due to the influence of random noise inside the atomic clock, the clock error data of the hydrogen clock is random and complex. In recent years, artificial intelligence, as an emerging and efficient field, can make more accurate predictions for this type of data through self-learning and optimization. In the prior art, an improved BP neural network is used to predict the clock error of the hydrogen clock. Subsequently, a recurrent neural network (RNN) with memory is further used to process sequence data. The current state is passed to the prediction of the next time step through the cyclic connection of the hidden layer. However, the traditional RNN will have problems with gradient explosion and gradient disappearance during training. Subsequently, a recurrent neural network LSTM with long short-term memory capability is also used for clock error prediction.

[0004] However, the single-layer LSTM model in the prior art cannot fully characterize the required features. At the same time, due to the complex structure of the LSTM model and the large number of gated units used, a large amount of memory will be occupied when performing multi-layer combinations of the LSTM model, resulting in low computational efficiency. Therefore, how to better achieve a systematic characterization of the clock error characteristics and use it to predict the hydrogen clock error is a problem that needs to be solved at present. Summary of the invention

[0005] The embodiments of the present application provide a hydrogen clock error prediction method, system, device and storage medium based on an improved recurrent neural network, which utilizes the memory characteristics of the recurrent neural network to characterize the time relationship, thereby performing hydrogen clock error prediction and improving the accuracy of the hydrogen clock error prediction.

[0006] In order to achieve the above objectives, this application adopts the following technical solutions:

[0007] In a first aspect, the present application provides a hydrogen clock error prediction method based on an improved recurrent neural network, the method comprising:

[0008] Obtaining the original clock error data of the hydrogen clock and performing preprocessing to obtain preprocessed clock error data;

[0009] Construct a hybrid recurrent neural network model and initialize the hybrid recurrent neural network model;

[0010] The preprocessed clock error data is divided into a training set, a validation set, and a test set;

[0011] Based on the training set, validation set, and test set, and using the RAdam optimizer and MAPE as the loss function, optimize the hybrid recurrent neural network model until the preset accuracy is achieved;

[0012] The clock difference data of the time period to be predicted is preprocessed, and the preprocessed clock difference data of the time period to be predicted is input into the optimized hybrid recurrent neural network model to obtain the predicted clock difference value.

[0013] In a possible design scheme, the method of the first aspect further includes that the preprocessing comprises:

[0014] By analyzing the characteristics of the original clock error data, the normal fluctuation range of the original clock error data is determined;

[0015] Based on the normal fluctuation range, the outliers of the original clock error data are eliminated, and then the original clock error data after eliminating the outliers are normalized.

[0016] A possible design scheme, the method of the first aspect also includes that the hybrid recurrent neural network model includes multiple layers, which are LSTM layer, Dropout layer, GRU layer and Dense layer in sequence,

[0017] LSTM layer, used to extract long-distance time-dependent features;

[0018] GRU layer, used to extract long-range temporal dependency features with less computational overhead;

[0019] A Dropout layer is added between the LSTM layer and the GRU layer to prevent overfitting of the original clock error data;

[0020] Dense layer is used to obtain the final prediction result.

[0021] A possible design scheme, the method of the first aspect also includes initializing the hybrid recurrent neural network model, including: initializing the connection weights of each layer of the hybrid recurrent neural network model, setting the number of neurons in the LSTM layer; setting the parameters of the Dropout layer; setting the number of neurons in the GRU layer; setting the number of neurons output by the Dense layer.

[0022] In a possible design scheme, the method of the first aspect further includes, after obtaining the predicted clock difference value, further including:

[0023] The predicted clock difference value is normalized and inverse transformed to obtain the actual predicted value.

[0024] In a second aspect, the present application provides a hydrogen clock clock error prediction system based on an improved recurrent neural network, the system comprising: a data collection and processing module, for obtaining the original clock error data of the hydrogen clock, and performing preprocessing to obtain the preprocessed clock error data; and for preprocessing the clock error data of the time period to be predicted;

[0025] A model building module is used to build a hybrid recurrent neural network model and initialize the hybrid recurrent neural network model;

[0026] A model optimization module is used to divide the preprocessed clock error data into a training set, a validation set, and a test set; it is also used to optimize the hybrid recurrent neural network model based on the training set, the validation set, and the test set, using the RAdam optimizer and MAPE as the loss function, until the preset accuracy is achieved;

[0027] The model prediction module is used to input the preprocessed clock difference data of the time period to be predicted into the optimized hybrid recurrent neural network model to obtain the predicted clock difference value.

[0028] In a possible design scheme, the system of the first aspect further includes: preprocessing includes determining a normal fluctuation range, eliminating outliers and normalizing,

[0029] By analyzing the characteristics of the original clock error data, the normal fluctuation range of the original clock error data is determined;

[0030] Based on the normal fluctuation range, the outliers of the original clock error data are eliminated, and then the original clock error data after eliminating the outliers are normalized.

[0031] In a possible design scheme, the system of the first aspect also includes that the data collection and processing module is also used to perform a normalized inverse transformation on the predicted clock difference value to obtain an actual predicted value.

[0032] In a third aspect, an electronic device is provided, comprising: a processor, and a memory for storing a program, wherein the program comprises instructions, which when executed by the processor cause the processor to execute the hydrogen clock error prediction method based on an improved recurrent neural network of the first aspect.

[0033] In a fourth aspect, a computer-readable storage medium is provided, which includes a computer program or instruction stored therein. When the computer program or instruction is executed, the hydrogen clock difference prediction method based on an improved recurrent neural network of the first aspect is executed.

[0034] In the embodiment of the present application, the original clock error data of the hydrogen clock is obtained and preprocessed to obtain the preprocessed clock error data; and a hybrid recurrent neural network model is constructed and initialized; the preprocessed clock error data is divided into a training set, a validation set and a test set, and the hybrid recurrent neural network model is optimized using the RAdam optimizer and MAPE as the loss function until the preset accuracy is reached; then the clock error data of the time period to be predicted is preprocessed, and the preprocessed clock error data of the time period to be predicted is input into the optimized hybrid recurrent neural network model to obtain the predicted clock error value. Thus, the memory characteristics of the recurrent neural network are used to characterize the time relationship, the hydrogen clock clock error is predicted, and the accuracy of the hydrogen clock clock error prediction is improved.

[0035] Other features and advantages of the present application will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0037] Figure 1 Schematic diagram of the process of hydrogen clock error prediction method based on improved recurrent neural network provided in the embodiment of the present application Figure 1 ;

[0038] Figure 2 Schematic diagram of the process of hydrogen clock error prediction method based on improved recurrent neural network provided in the embodiment of the present application Figure 2 ;

[0039] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application. At the same time, in the description of the embodiments of the present application, the terms "first", "second", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more features. In the description of the embodiments of the present application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0041] Figure 1 Schematic diagram of the process of hydrogen clock error prediction method based on improved recurrent neural network provided in the embodiment of the present application Figure 1 The process of the hydrogen clock error prediction method based on the improved recurrent neural network is as follows:

[0042] Step S101, obtaining the original clock error data of the hydrogen clock, and performing preprocessing to obtain the preprocessed clock error data.

[0043] It can be understood that the above pre-processing includes the following steps:

[0044] By analyzing the characteristics of the original clock error data, the normal fluctuation range of the original clock error data is determined;

[0045] Based on the normal fluctuation range, the outliers of the original clock error data are eliminated, and then the original clock error data after eliminating the outliers are normalized.

[0046] Exemplarily, the original clock error data of the hydrogen clock is obtained as X=(x1, x2, ..., x L ). By analyzing the clock error characteristics of the current hydrogen clock, for example, calculating the mean and standard deviation of the original clock error data, the normal fluctuation range M of the original clock error data is determined.

[0047] Outliers were eliminated and data were normalized according to the normal fluctuation range of clock error.

[0048] Outlier elimination can be understood as setting upper and lower thresholds based on the normal fluctuation range M. If the clock error exceeds the range M, it is considered an outlier, and can be corrected, discarded, or interpolated based on the error size of the abnormal data (using the normal data before and after the abnormal value to adopt the linear fitting value). Interpolation is performed to replace abnormal data points).

[0049] Data normalization can be understood as using Min-Max Normalization on the original clock error data after outliers are eliminated to map the clock error data to the range of [-1,1]. The calculation method is as follows: Among them, x is the original data, min(x) and max(x) are the minimum and maximum values ​​in the data respectively, and x′ is the normalized data; you can also use Z-score standardization: standardize the data by mean and standard deviation so that its mean is 0 and standard deviation is 1. The calculation method is as follows: Wherein, μ is the mean of the data, σ is the standard deviation, and x′ is the standardized data; or other methods are not limited here.

[0050] Step S102, constructing a hybrid recurrent neural network model and initializing the hybrid recurrent neural network model.

[0051] The hybrid recurrent neural network model includes multiple layers, namely LSTM layer, Dropout layer, GRU layer and Dense layer. Among them, the LSTM layer is used to extract long-distance time-dependent features; the GRU layer is used to extract long-distance time-dependent features with less computational overhead; the Dropout layer is added between the LSTM layer and the GRU layer to prevent overfitting of the original clock error data; the Dense layer is used to obtain the final prediction result.

[0052] In specific applications, it is necessary to initialize the connection weights of each layer of the hybrid recurrent neural network model, set the number of neurons in the LSTM layer, set the parameters of the Dropout layer, set the number of neurons in the GRU layer, and set the number of neurons output by the Dense layer.

[0053] Exemplarily, in the hybrid recurrent neural network model, the number of neurons in the LSTM layer is set to 1, the parameter of the Dropout layer is set to 0.25, the number of neurons in the GRU layer is set to 1, and the number of neurons output by the Dense layer is set to 0. It should be noted that the confirmation of the above-mentioned values ​​of each layer is based on the computing power of the engineering environment.

[0054] The following describes the LSTM layer, Dropout layer, GRU layer, and Dense layer in detail.

[0055] The LSTM layer introduces a "gate" mechanism based on the traditional RNN, including a forget gate, an input gate, and an output gate to solve the problems of gradient disappearance and gradient explosion.

[0056] The GRU layer simplifies the structure based on the LSTM layer, reduces the number of "gates", simplifies it to reset gates and update gates, and improves the calculation efficiency. Here, the GRU layer and the LSTM layer are combined to utilize more features in the model without reducing the calculation speed.

[0057] The five gates involved in the GRU layer and the LSTM layer use two activation functions, tanh and sigmoid, and the corresponding expressions are

[0058] The LSTM layer calculates the forget gate, input gate, and output gate in sequence. The corresponding process is as follows:

[0059] f t =sigmoid(W f ·[h t-1 , d t_in ]+b f ),

[0060] i t =sigmoid(W i ·[h t-1 , d t_in ]+b i ),

[0061] C t =f t *C t-1 +i t *tanh(W C ·[h t-1 , d t_in ]+b C ),

[0062] h t =sigmoid(W o ·[h t-1 , d t_in ]+b o )*tanh(C t ).

[0063] Among them, b f is the bias of the forget gate, d t_in is the input at the current moment, h t-1 is the memory of the previous moment, W f is the weight of the forget gate, f t is the output corresponding to the forget gate. For sigmoid and tanh, please refer to the previous text for understanding. i is the bias of the input gate, W i is the weight of the forget gate, i t is the output of the input gate. C t-1 is the output of the control item in the input gate at the previous moment, W C is the weight of the control item, b C is the deviation of the control item. o is the weight of the output gate, b o is the deviation of the output gate, h t is the output of the output gate, i.e. the memory of the current neuron.

[0064] The GRU layer calculates the reset gate and the update gate in sequence. The corresponding calculation process is as follows:

[0065] r t =sigmoid(W r ·[h t-1 , d t_in ]+b r ),

[0066] z t =sigmoid(W z ·[h t-1 , dt_in ]+b z ),

[0067] C t =(1-z t )*C t-1 +z t *tanh(W c ·[r t *h t-1 , d t_in ]+b C ),

[0068] h t =C t .

[0069] Among them, b r is the bias of the reset gate, d t_in is the input at the current moment, h t-1 is the memory of the previous moment, W r To reset the gate weight, r t To reset the output of the gate, sigmoid and tanh can be understood by referring to the previous text. z is the bias of the update gate, W z is the weight of the update gate, z t is the output of the update gate. C t-1 is the output of the control item at the previous moment, W C is the weight of the control item, b C is the deviation of the control item.

[0070] In addition, in order to connect the LSTM layer and the GRU layer and to avoid overfitting, a Dropout layer is used as a transition in the middle. Specifically, the Dropout layer randomly "discards" a part of the neurons in the neural network (that is, the output weights of some neurons are set to 0), thereby reducing the model's dependence on specific neurons and enhancing the generalization ability of the model. It should also be noted that, generally, the parameters of the Dropout layer are generally set between 0.2 and 0.5. A higher value may cause the model to be too "sparse" during training, affecting the model from learning useful features. For example, the embodiment of the present application sets the parameter of the Dropout layer to 0.25.

[0071] The number of neurons in the Dense layer can be determined according to the expected prediction range without any restriction.

[0072] Step S103, dividing the preprocessed clock error data into a training set, a validation set and a test set.

[0073] That is to say, the preprocessed clock error data is divided into training set, validation set and test set according to a certain ratio. The training set is used to train the model. This part of the data accounts for about 70%-80% of the total data set. The model will learn the relationship between features and labels through this data. The validation set is used to adjust the hyperparameters of the model to prevent overfitting. The validation set generally accounts for 10%-15% of the data. During the model training process, the performance will be regularly evaluated on the validation set to adjust the hyperparameters. The test set is used to finally evaluate the generalization ability of the model. The test set usually accounts for 10%-15% of the data. It does not participate in training and parameter adjustment. It is only used to evaluate the performance of the model on unseen data after the model training is completed.

[0074] Exemplarily, in the embodiment of the present application, X=(x1, x2, ..., x L ), split into input and output pairs that can be used for model training, as follows:

[0075] ([x1, x2, …, x I ],[x I+1 , x I+2 , …, x i+o ]),([x2,…,x I , x I+1 ],[x I+2 , …, x i+O , x i+O+1 ]),

[0076] ([x1, x2, …, x I ],[x I+1 , x I+2 , …, x i+O ]),([x2,…,x I , x I+1 ],[x I+2 , …, x i+O , x i+O+1 ]), …,

[0077] ([x L-I-O+1 , x L-I-O+2 , …, x L-O ],[x L-O+1 , x L-O+2 , …, x L ]), and then divide the LI-O+1 data pairs into training set, evaluation set and test set in proportional order.

[0078] Step S104, based on the training set, the validation set and the test set, and using the RAdam optimizer and MAPE as the loss function, optimize the hybrid recurrent neural network model until a preset accuracy is reached.

[0079] When using the training set, validation set, and test set to optimize the model, you can also use the RAdam optimizer to optimize the model. This optimizer can automatically adjust the learning rate, and has fast convergence speed and strong robustness.

[0080] The RAdam optimizer optimization method improves the performance in the early stages of training by correcting the adaptive learning rate. Specifically:

[0081] Use the modified learning rate to update the parameters. The parameter update rule of RAdam is as follows: Among them, η is the learning rate, r t is the variance cutoff term, in, β2=0.99, which is the attenuation coefficient of the adaptive learning rate; ε is used to prevent the divisor from being 0, and is set to ε=10 -8 , is the momentum part, β1=0.9 is the momentum attenuation coefficient, To increase the adaptive part,

[0082] In short, the RAdam optimizer makes the model training process smoother by correcting the instability that the Adam optimizer may cause in the early stages of training, which is especially effective in the case of small batch training.

[0083] At the same time, we also consider that the training data values ​​are small, and the mean square error, absolute error and other precisions are not high. Using MAPE as the loss function during the training process can directly optimize the relative error, ensure that the model prediction results are more accurate, and can reflect the proportion of the prediction error to the true value, which is especially suitable for time series and regression tasks. The specific calculation method is: Among them, yi is the true value of the i-th sample, is the predicted value of the ith sample, and n is the total number of samples.

[0084] Step S105, preprocessing the clock difference data of the time period to be predicted, inputting the preprocessed clock difference data of the time period to be predicted into the optimized hybrid recurrent neural network model to obtain a predicted clock difference value.

[0085] For the optimized hybrid recurrent neural network model, the clock difference data of the time period to be predicted is preprocessed to obtain X p_in =(x1, x2, ..., x i ), and input the optimized hybrid recurrent neural network model, we can get Y p_out =(x1, x2, ..., x o ).

[0086] In addition, after step S105, the predicted clock difference value is subjected to normalized inverse transformation to obtain an actual predicted value.

[0087] Corresponding to the normalization step of the above step S101 example, the predicted clock difference value is normalized and inversely transformed. For example, for the minimum-maximum normalization, the corresponding inverse transformation is x=x′(max(x)-min(x))+min(x). For another example, for the standardized normalization, the corresponding inverse transformation is x=x′×σ+μ, and the actual predicted value has been obtained.

[0088] In summary, the embodiment of the present application obtains the original clock error data of the hydrogen clock and performs preprocessing to obtain the preprocessed clock error data; and constructs a hybrid recurrent neural network model and initializes it; the preprocessed clock error data is divided into a training set, a validation set and a test set, and the hybrid recurrent neural network model is optimized using the RAdam optimizer and MAPE as the loss function until the preset accuracy is reached; then the clock error data of the time period to be predicted is preprocessed, and the preprocessed clock error data of the time period to be predicted is input into the optimized hybrid recurrent neural network model to obtain the predicted clock error value. Thus, the memory characteristics of the recurrent neural network are used to characterize the time relationship, predict the hydrogen clock error, and improve the accuracy of the hydrogen clock error prediction.

[0089] Combination of the above Figure 1-Figure 2 The hydrogen clock clock error prediction method based on the improved recurrent neural network provided in the embodiment of the present application is described in detail, and the hydrogen clock clock error prediction system based on the improved recurrent neural network provided in the embodiment of the present application is described in detail below. The system specifically includes: a data collection and processing module, a model construction module, a model optimization module and a model prediction module, as shown below:

[0090] The data collection and processing module is used to obtain the original clock error data of the hydrogen clock and perform preprocessing to obtain the preprocessed clock error data; it is also used to preprocess the clock error data of the time period to be predicted;

[0091] A model building module is used to build a hybrid recurrent neural network model and initialize the hybrid recurrent neural network model;

[0092] A model optimization module is used to divide the preprocessed clock error data into a training set, a validation set, and a test set; it is also used to optimize the hybrid recurrent neural network model based on the training set, the validation set, and the test set, using the RAdam optimizer and MAPE as the loss function, until the preset accuracy is achieved;

[0093] The model prediction module is used to input the preprocessed clock difference data of the time period to be predicted into the optimized hybrid recurrent neural network model to obtain the predicted clock difference value.

[0094] In addition, for the above-mentioned specific implementation of the system, since it is basically similar to the method implementation, the description is relatively simple, and the relevant parts can be referred to the partial description of the method implementation. Moreover, it should be noted that in each module of the system of the present application, the components therein are logically divided according to the functions to be implemented, but the present application is not limited thereto, and the components can be re-divided or combined as needed.

[0095] The present application embodiment also provides an electronic device 300, such as Figure 3 As shown, it includes a memory 301, a processor 302, and a computer program 303 stored in the memory 301 and executable on the processor 302. When the processor 302 executes the computer program 303, the steps of the hydrogen clock error prediction method based on the improved recurrent neural network provided in the first aspect are implemented.

[0096] In applications, the electronic device may include, but is not limited to, a processor and a memory. Figure 3 This is only an example of an electronic device and does not constitute a limitation on the electronic device. It may include more or fewer components than shown in the figure, or combine certain components, or different components, such as input and output devices, network access devices, etc. Input and output devices may include cameras, audio acquisition / playback devices, display screens, etc. Network access devices may include a network module for wirelessly communicating with external devices.

[0097] In applications, the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0098] In applications, the memory may be an internal storage unit of an electronic device in some embodiments, such as a hard disk or memory of the electronic device. The memory may also be an external storage device of the electronic device in other embodiments, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. The memory may also include both an internal storage unit of the electronic device and an external storage device. The memory is used to store operating systems, applications, boot loaders (BootLoader), data, and other programs, such as program codes of computer programs, etc. The memory may also be used to temporarily store data that has been output or is to be output.

[0099] The embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.

[0100] The present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of each of the above-mentioned method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to an electronic device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk.

[0101] Those of ordinary skill in the art will appreciate that the devices and algorithm 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 this application. In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic, and another point is that the coupling or direct coupling or communication connection between each other shown or discussed can be through some interfaces, indirect coupling or communication connection of devices, which can be electrical, mechanical or other forms.

[0102] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such 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 application, and should all be included in the protection scope of the present application.

Claims

1. A hydrogen clock error prediction method based on an improved recurrent neural network, characterized in that: The method comprises: Obtaining the original clock error data of the hydrogen clock and performing preprocessing to obtain preprocessed clock error data; Constructing a hybrid recurrent neural network model and initializing the hybrid recurrent neural network model; Dividing the preprocessed clock error data into a training set, a validation set and a test set; Based on the training set, the validation set and the test set, and using the RAdam optimizer and MAPE as the loss function, optimizing the hybrid recurrent neural network model until a preset accuracy is achieved; The clock difference data of the time period to be predicted is preprocessed, and the preprocessed clock difference data of the time period to be predicted is input into the optimized hybrid recurrent neural network model to obtain a predicted clock difference value.

2. The hydrogen clock error prediction method based on improved recurrent neural network according to claim 1 is characterized in that: The pre-processing comprises: Determining a normal fluctuation range of the original clock error data by performing characteristic analysis on the original clock error data; Based on the normal fluctuation range, outliers are eliminated from the original clock error data, and then the original clock error data after outliers are eliminated is normalized.

3. The hydrogen clock error prediction method based on improved recurrent neural network according to claim 1 is characterized in that: The hybrid recurrent neural network model includes multiple layers, namely, LSTM layer, Dropout layer, GRU layer and Dense layer. The LSTM layer is used to extract long-distance time-dependent features; The GRU layer is used to extract long-distance time-dependent features with less computational overhead; The Dropout layer is added between the LSTM layer and the GRU layer to prevent overfitting of the original clock error data; The Dense layer is used to obtain the final prediction result.

4. The hydrogen clock error prediction method based on improved recurrent neural network according to claim 3 is characterized in that: The initialization of the hybrid recurrent neural network model includes: initializing the connection weights of each layer of the hybrid recurrent neural network model, setting the number of neurons in the LSTM layer; setting the parameters of the Dropout layer; setting the number of neurons in the GRU layer; and setting the number of neurons output by the Dense layer.

5. The hydrogen clock error prediction method based on improved recurrent neural network according to claim 1 is characterized in that: After obtaining the predicted clock difference value, the method further includes: The predicted clock difference value is normalized and inversely transformed to obtain the actual predicted value.

6. A hydrogen clock error prediction system based on an improved recurrent neural network, characterized in that: The system comprises: The data collection and processing module is used to obtain the original clock error data of the hydrogen clock and perform preprocessing to obtain the preprocessed clock error data; it is also used to preprocess the clock error data of the time period to be predicted; A model building module, used to build a hybrid recurrent neural network model and initialize the hybrid recurrent neural network model; A model optimization module, used to divide the preprocessed clock error data into a training set, a validation set and a test set; and further used to optimize the hybrid recurrent neural network model based on the training set, the validation set and the test set, using a RAdam optimizer and MAPE as a loss function, until a preset accuracy is achieved; The model prediction module is used to input the preprocessed clock difference data of the time period to be predicted into the optimized hybrid recurrent neural network model to obtain the predicted clock difference value.

7. The hydrogen clock error prediction system based on improved recurrent neural network according to claim 6 is characterized in that: The pre-processing comprises: Determining a normal fluctuation range of the original clock error data by performing characteristic analysis on the original clock error data; Based on the normal fluctuation range, outliers are eliminated from the original clock error data, and then the original clock error data after outliers are eliminated is normalized.

8. The hydrogen clock error prediction system based on improved recurrent neural network according to claim 6 is characterized in that: The data collection and processing module is also used to perform a normalized inverse transformation on the predicted clock difference value to obtain an actual predicted value.

9. An electronic device, characterized in that: The electronic device comprises: a processor, and a memory storing a program, wherein the program comprises instructions, and when the instructions are executed by the processor, the processor is caused to perform the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a computer program or instruction stored therein, and when the computer program or instruction is executed, the method according to any one of claims 1 to 7 is performed.

Images