Hydropower station Beidou PPP-B2b time service error intelligent dynamic correction method

Abstract

The invention discloses an intelligent dynamic correction method for a Beidou PPP-B2b time service error of a hydropower station, and the method comprises the steps: marking timestamps on collected TEC time sequence data, dam body vibration data and environmental meteorological data, and synchronizing heterogeneous data streams to a unified time sequence; processing the data to obtain TEC time sequence features, dam body vibration feature vectors and meteorological change trend features; the TEC time sequence features and the meteorological change trend features are input into an LSTM network for processing, then output codes and dam body vibration feature vectors are spliced and then input into a Transform encoder, and a self-attention mechanism is adopted for processing and analysis to obtain a time service deviation prediction value; and taking a time service deviation predicted value as a compensation amount, dynamically correcting a receiver clock error obtained by resolving the Beidou PPP-B2b signal, and then outputting accurate time information. According to the invention, dynamic compensation of time service errors can be realized, high-precision time service is provided for operation of the hydropower station, and the hydropower station can operate efficiently and stably.

Description

Technical Field

[0001] This invention belongs to the field of BeiDou timing technology, specifically relating to an intelligent dynamic correction method for BeiDou PPP-B2b timing errors in hydropower stations. Background Technology

[0002] Time synchronization technology refers to the technology of providing time information to other devices or systems through standard interfaces or protocols. Its core purpose is to ensure time consistency among multiple devices or systems, providing a reliable time reference for data acquisition, transmission, and processing. Time synchronization technology is widely used in fields such as power, communications, water conservancy, and finance, playing a particularly important role in scenarios requiring high-precision time synchronization.

[0003] For example, high-precision time synchronization is one of the core foundations for the safe, stable, and efficient operation of hydropower stations. From active / reactive power regulation and grid synchronization of generator units to condition monitoring, fault recording, automatic control, and relay protection systems, all rely on high-precision time references at the nanosecond to microsecond level. Global Navigation Satellite Systems (GNSS), especially my country's BeiDou Navigation Satellite System, have become the mainstream time synchronization technology for hydropower stations due to their wide coverage, high accuracy, and relatively low cost. BeiDou Precision Point Positioning (PPP-B2b) service can further improve the time synchronization accuracy to the sub-nanosecond level, providing strong technical support for hydropower stations. However, in practical engineering applications, especially in the special industrial environment of hydropower stations, high-precision time synchronization based on BeiDou PPP-B2b faces a series of severe challenges, making it difficult to maintain its theoretical accuracy continuously in real-world scenarios. These challenges are mainly manifested in the following aspects:

[0004] (1) Significant impact of ionospheric disturbances: As an essential medium for signal propagation, the dynamic changes in the total electron content (TEC) of the ionosphere are one of the main sources of timing errors. Hydropower stations are mostly located in mountainous valleys, where ionospheric activity itself is uncertain. In addition, the complex path of satellite signals through the atmosphere results in large residuals in the correction of traditional ionospheric models. Although PPP-B2b broadcasts precise orbits and clock errors, its ability to model and correct instantaneous, regional drastic ionospheric disturbances (such as scintillation and abrupt changes) is limited. These disturbances can directly cause jumps or slow drifts in receiver clock error calculations.

[0005] (2) Vibration interference in complex environments: The dam body and powerhouse structure of hydropower stations will generate continuous environmental vibrations under the action of unit operation and water flow impact. This vibration will be directly transmitted to the GNSS receiver antenna, causing the antenna phase center to shift by micrometers to millimeters, which is the so-called "multipath effect" and "antenna phase center disturbance". This error caused by vibration is difficult to model and separate in traditional timing calculation, which seriously damages the stability of carrier phase observation values, thus introducing timing noise and bias that cannot be ignored.

[0006] (3) The combined effect of local meteorological conditions: The special microclimate environment formed in the reservoir area of ​​the hydropower station leads to frequent and drastic changes in meteorological elements such as temperature, air pressure, and humidity. The fluctuation of meteorological parameters not only affects the tropospheric delay of the signal, but its rapid changes also have a subtle impact on the hardware circuit of the receiver itself (such as temperature-compensated crystal oscillators), resulting in additional frequency drift of the receiver's internal time base. Existing technologies usually only focus on the static correction of tropospheric delay, while ignoring this "endogenous" error of the receiver induced by environmental meteorological changes.

[0007] Currently, mainstream methods for addressing these issues often employ single approaches or simple combinations. For example, filtering algorithms are used to smooth clock error results, or post-hoc compensation is achieved by establishing statistical models of a single factor (such as ionospheric TEC) and historical deviations. These methods have significant limitations. They often target only a specific type of error source, failing to comprehensively consider the coupled effects of multiple sources such as the ionosphere, mechanical vibration, and meteorological environment at the system level. Furthermore, traditional time-series models (such as ARIMA and Kalman filtering) struggle to effectively capture the complex nonlinear temporal dependencies in multi-source heterogeneous data. In addition, simple feature splicing and linear regression cannot learn the deep modulation effects of vibration, meteorology, and other factors on ionospheric time-series error patterns, resulting in insufficient generalization ability and prediction accuracy of the compensation model. Therefore, a smart time synchronization method that can deeply integrate multi-source environmental sensing data from hydropower stations and accurately model the complex nonlinear mapping relationship between this data and time synchronization deviations is still needed. Summary of the Invention

[0008] To address the aforementioned shortcomings, this invention discloses an intelligent dynamic correction method for BeiDou PPP-B2b timing errors in hydropower stations. This method enables dynamic compensation for timing errors, providing high-precision timing services for hydropower station operation and ensuring efficient and stable operation of the hydropower station.

[0009] This invention is achieved using the following technical solution:

[0010] A method for intelligent dynamic correction of BeiDou PPP-B2b timing error in hydropower stations includes the following steps:

[0011] (1) Receive BeiDou PPP-B2b signals in real time and analyze the BeiDou PPP-B2b signals to obtain the time series data of total electron content (TEC) in the ionosphere; collect dam vibration data by sensors deployed on the dam body of the hydropower station, and collect environmental meteorological data of the hydropower station at the same time;

[0012] The time series of total electron content in the ionosphere is expressed as follows: ,in This indicates the received BeiDou PPP-B2b raw signal. This represents the PPP-B2b signal parsing algorithm;

[0013] (2) Timestamp the obtained time series data of total ionospheric electron content, dam vibration data, and environmental meteorological data, i.e. Furthermore, heterogeneous data streams are synchronized to a unified time series using an interpolation algorithm and Formula I, where Formula I is... , where the coefficient Determined by boundary conditions;

[0014] (3) Process the time series data of total electron content in the ionosphere, calculate the rate of change, the mean and standard deviation of the sliding window to obtain the TEC time series characteristics; extract the energy value, main frequency and spectral entropy of the preset frequency band from the dam vibration data as the dam vibration feature vector characterizing the vibration characteristics. The preset frequency band can be set to 0-10Hz or 10-50Hz. Low frequency vibration (0-10Hz) is related to the overall deformation of the dam or low frequency water flow pulsation, while high frequency vibration (10-50Hz) is related to the operation of mechanical equipment or local turbulence; perform first-order difference processing on the environmental meteorological data to obtain the meteorological change trend characteristics.

[0015] The rate of change R(t) represents the change in TEC value between adjacent time points, and is calculated using the formula: R(t) = T(t) - T(t-1), where t≥2;

[0016] The mean of the sliding window is the sliding mean μ(t) over time t with respect to window size ω, where ω is a positive integer. Its calculation formula is as follows: ,

[0017] Where t≥ω, and i is the time index within the window;

[0018] The standard deviation of the sliding window is the sliding standard deviation σ(t) over time t, and its calculation formula is:

[0019] , where t≥ω;

[0020] The timing characteristics of TEC are represented as follows: ;

[0021] The dam vibration data is denoted as V(t). Features are extracted through frequency domain analysis. It is assumed that the power spectral density is obtained by performing a Fourier transform on the signal. ,in For frequency, the formula for calculating power spectral density is: ;

[0022] Let the lower and upper limits of the preset frequency band be respectively and The energy value E is calculated using the integral power spectral density formula: For discrete data, the summation formula is: ,in Frequency resolution; main frequency It is the frequency corresponding to the maximum power in the power spectrum, and the calculation formula is: That is, to search The frequency of the global maximum point;

[0023] Extracting spectral entropy involves normalizing the power spectrum into a probability distribution: Then calculate the entropy: ,if =0, then define ;

[0024] The vibration characteristic vector of the dam body is represented as follows: ;

[0025] The environmental meteorological data is denoted as The meteorological change trend characteristics are obtained through first-order difference processing, specifically the change trend. The calculation formula is: , where t≥2, represents the change in meteorological values ​​between adjacent time points, used to capture short-term trends;

[0026] The aforementioned meteorological change trend characteristics are expressed as follows: ;

[0027] (4) Input the TEC time series features and meteorological change trend features obtained in step (3) into the LSTM network for processing, establish the long-term dependency relationship between the TEC time series features and meteorological change trend features, then concatenate the output encoding of the LSTM network with the dam vibration feature vector obtained in step (3) to obtain the concatenated features, then input the concatenated features into the Transformer encoder, use the self-attention mechanism to learn the modulation effect of vibration and meteorological factors on the time series error pattern, and finally regress the real-time timing deviation prediction value through the fully connected layer;

[0028] (5) Use the predicted time deviation value obtained in step (4) as the compensation amount to dynamically correct the receiver clock error obtained by solving the Beidou PPP-B2b signal, and then output accurate time information.

[0029] Furthermore, in step (1), vibration signals are collected using MEMS acceleration sensors deployed on the dam body of the hydropower station, specifically vibration acceleration signals in three orthogonal directions (X, Y, Z), denoted as the time-domain vector a(t) = [a x (t), a y (t),a z[(t)], then power spectral density (PSD) estimation based on fast Fourier transform (FFT) is performed on the acceleration signal in each direction to obtain the corresponding vibration spectrum data, and the obtained vibration spectrum data is used as dam vibration data for subsequent processing.

[0030] Furthermore, the environmental meteorological data mentioned in step (1) includes data on temperature, humidity, and air pressure of the hydropower station environment, and the environmental meteorological data is expressed as follows: ,in This represents the temperature, air pressure, and humidity at time t.

[0031] Furthermore, in step (3), the time series data of total electron content in the ionosphere are processed by moving average, and then the rate of change is calculated by first-order difference.

[0032] Furthermore, the LSTM network in step (4) includes a forget gate, an input gate, a cell state update gate, and an output gate; the forget gate is used to discard which information; the input gate is used to store which new information; the cell state update gate is used to update the old cell state to a new cell state; and the output gate is used to determine which parts of the cell state will be output.

[0033] Among them, input feature concatenation: ;

[0034] Input gates in an LSTM cell: ,

[0035] Forgotten Gate: ,

[0036] Candidate memory cells: ,

[0037] Memory cell renewal: ,

[0038] Output gate: ,

[0039] Hidden state output: ,

[0040] LSTM output encoding: ;

[0041] The splicing features: ,

[0042] Location coding: ;

[0043] Multi-head self-attention mechanism (the first) The layer includes the following formula:

[0044] Query, key-value projection: ,

[0045] Scaling Dot Product Attention: ,

[0046] Multi-head attention splicing: ,

[0047] Feedforward networks and layer normalization: ;

[0048] The fully connected layer regression: .

[0049] Furthermore, in step (4), the loss function used in the fully connected layer regression is:

[0050] .

[0051] Furthermore, in step (5), the observation equation for solving the clock bias of the PPP-B2b receiver is: , ;

[0052] Receiver clock bias estimation: ;

[0053] Receiver clock error after dynamic clock compensation: Where the adaptive gain factor is:

[0054] ;

[0055] System time estimation: ;

[0056] Final output time: ,in .

[0057] Furthermore, in step (5), a time accuracy index is used for performance evaluation. The formula for the time accuracy index is:

[0058] .

[0059] Compared with existing technologies, this technical solution has the following advantages:

[0060] 1. This invention aims to dynamically sense, accurately predict, and compensate in real time for timekeeping deviations introduced by environmental factors. This fully leverages the theoretical accuracy advantages of BeiDou PPP-B2b technology in complex industrial environments, providing a stable, reliable, and secure high-precision time reference for critical hydropower station systems. Notably, it is the first to introduce the dam vibration spectrum as a key input into the timekeeping error model, pioneering a quantitative compensation relationship between mechanical vibration and timekeeping deviation, thus solving the failure problem of traditional models under vibration conditions. Simultaneously, the proposed LSTM-Transformer serial hybrid model fully utilizes the advantages of LSTM in handling time-series dependencies and Transformer in capturing cross-modal correlations, achieving deeper information fusion of multi-source heterogeneous data and high prediction accuracy.

[0061] 2. This invention, through modular design and a deployment strategy of offline training and online inference, meets the real-time and reliability requirements of complex industrial environments while ensuring nanosecond-level timing accuracy. Attached Figure Description

[0062] Figure 1 This is a flowchart of the intelligent dynamic correction method for Beidou PPP-B2b timing error in hydropower stations described in Example 1.

[0063] Figure 2 This is a network structure diagram of the intelligent dynamic correction method for Beidou PPP-B2b timing error in hydropower stations described in this invention.

[0064] Figure 3 The graph shows the time difference results obtained by different methods described in Example 2. Curve 1 represents the control method, and curve 2 represents the intelligent dynamic correction method for Beidou PPP-B2b time synchronization error of hydropower stations described in Example 2. Detailed Implementation

[0065] The present invention is further illustrated by the following examples, but these are not intended to limit the invention. Specific experimental conditions and methods not specified in the following examples are generally conventional methods well known to those skilled in the art.

[0066] Example 1: As Figure 1 As shown, a method for intelligent dynamic correction of BeiDou PPP-B2b timing errors in hydropower stations includes the following steps:

[0067] (1) Receive BeiDou PPP-B2b signals in real time and analyze the BeiDou PPP-B2b signals to obtain the time series data of total electron content (TEC) in the ionosphere; collect dam vibration data by sensors deployed on the dam body of the hydropower station, and collect environmental meteorological data of the hydropower station at the same time;

[0068] The time series of total electron content in the ionosphere is expressed as follows: ,in This indicates the received BeiDou PPP-B2b raw signal. This represents the PPP-B2b signal parsing algorithm;

[0069] Vibration signals are collected using MEMS acceleration sensors deployed on the dam body of the hydropower station, specifically vibration acceleration signals in three orthogonal directions (X, Y, Z), denoted as the time-domain vector a(t) = [a x (t), a y (t), a z [(t)], and then the power spectral density (PSD) of the acceleration signal in each direction is estimated based on fast Fourier transform (FFT) to obtain the corresponding vibration spectrum data;

[0070] The environmental meteorological data includes temperature, humidity, and air pressure data for the hydropower station environment, and the environmental meteorological data is expressed as follows: ,in This represents the temperature, air pressure, and humidity at time t.

[0071] (2) Timestamp the obtained time series data of total ionospheric electron content, dam vibration data, and environmental meteorological data, i.e. Furthermore, heterogeneous data streams are synchronized to a unified time series using an interpolation algorithm and Formula I, where Formula I is... , where the coefficient Determined by boundary conditions;

[0072] (3) Process the time series data of total ionospheric electron content, calculate the rate of change, the mean and standard deviation of the sliding window to obtain the TEC time series characteristics; extract the energy value, main frequency and spectral entropy of the preset frequency band from the dam vibration data as the dam vibration feature vector characterizing the vibration characteristics; perform first-order difference processing on the environmental meteorological data to obtain the meteorological change trend characteristics; perform moving average processing on the time series data of total ionospheric electron content, and then perform first-order difference calculation to obtain the rate of change;

[0073] The rate of change R(t) represents the change in TEC value between adjacent time points, and is calculated using the formula: R(t) = T(t) - T(t-1), where t≥2;

[0074] The mean of the sliding window is the sliding mean μ(t) over time t with respect to window size ω, where ω is a positive integer. Its calculation formula is as follows: ,

[0075] Where t≥ω, and i is the time index within the window;

[0076] The standard deviation of the sliding window is the sliding standard deviation σ(t) over time t, and its calculation formula is:

[0077] , where t≥ω;

[0078] The timing characteristics of TEC are represented as follows: ;

[0079] The dam vibration data is denoted as V(t). Features are extracted through frequency domain analysis. It is assumed that the power spectral density is obtained by performing a Fourier transform on the signal. ,in For frequency, the formula for calculating power spectral density is: ;

[0080] Let the lower and upper limits of the preset frequency band be respectively and The energy value E is calculated using the integral power spectral density formula: For discrete data, the summation formula is: ,in Frequency resolution; main frequency It is the frequency corresponding to the maximum power in the power spectrum, and the calculation formula is: That is, to search The frequency of the global maximum point;

[0081] Extracting spectral entropy involves normalizing the power spectrum into a probability distribution: Then calculate the entropy: ,if =0, then define ;

[0082] The vibration characteristic vector of the dam body is represented as follows: ;

[0083] The environmental meteorological data is denoted as The meteorological change trend characteristics are obtained through first-order difference processing, specifically the change trend. The calculation formula is: , where t≥2, represents the change in meteorological values ​​between adjacent time points, used to capture short-term trends;

[0084] The aforementioned meteorological change trend characteristics are expressed as follows: ;

[0085] (4) Input the TEC time series features and meteorological change trend features obtained in step (3) into the LSTM network for processing, establish the long-term dependency relationship between the TEC time series features and meteorological change trend features, then concatenate the output encoding of the LSTM network with the dam vibration feature vector obtained in step (3) to obtain the concatenated features, then input the concatenated features into the Transformer encoder, use the self-attention mechanism to learn the modulation effect of vibration and meteorological factors on the time series error pattern, and finally regress the real-time timing deviation prediction value through the fully connected layer;

[0086] (5) Use the predicted time deviation value obtained in step (4) as the compensation amount to dynamically correct the receiver clock error obtained by solving the Beidou PPP-B2b signal, and then output accurate time information.

[0087] Example 2: The difference between the intelligent dynamic correction method for Beidou PPP-B2b timing error of hydropower stations described in this example and the method described in Example 1 is that the LSTM network in step (4) includes a forget gate, an input gate, a cell state update, and an output gate; the forget gate is used to discard which information; the input gate is used to store which new information; the cell state update is used to update the old cell state to a new cell state; and the output gate is used to determine which parts of the cell state will be output.

[0088] Among them, input feature concatenation: ;

[0089] Input gates in an LSTM cell: ,

[0090] Forgotten Gate: ,

[0091] Candidate memory cells: ,

[0092] Memory cell renewal: ,

[0093] Output gate: ,

[0094] Hidden state output: ,

[0095] LSTM output encoding: ;

[0096] The splicing features: ,

[0097] Location coding: ;

[0098] Multi-head self-attention mechanism (the first) The layer includes the following formula:

[0099] Query, key-value projection: ,

[0100] Scaling Dot Product Attention: ,

[0101] Multi-head attention splicing: ,

[0102] Feedforward networks and layer normalization: ;

[0103] The fully connected layer regression: ;

[0104] The loss function used in the fully connected layer regression is: ;

[0105] In step (5), the observation equation for solving the clock bias of the PPP-B2b receiver is: , ;

[0106] Receiver clock bias estimation: ;

[0107] Receiver clock error after dynamic clock compensation: Where the adaptive gain factor is:

[0108] ;

[0109] System time estimation: ;

[0110] Final output time: ,in .

[0111] Furthermore, performance is evaluated using a time precision index, the formula for which is:

[0112] .

[0113] The time synchronization of the hydropower station was performed according to the method described in Example 2, and compared with the original BeiDou PPP-B2b time synchronization method. The obtained time difference results are as follows: Figure 3 As shown, the method of the present invention can effectively obtain dynamic compensation for timing errors, providing higher precision timing services for hydropower station operation.

[0114] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for intelligent dynamic correction of BeiDou PPP-B2b timing error in hydropower stations, characterized in that: Includes the following steps: (1) Receive BeiDou PPP-B2b signals in real time and analyze the BeiDou PPP-B2b signals to obtain the time series of total ionospheric electron content; collect dam vibration data through sensors deployed on the dam body of the hydropower station, and simultaneously collect environmental meteorological data of the hydropower station; the time series of total ionospheric electron content is expressed as: ,in This indicates the received BeiDou PPP-B2b raw signal. This represents the PPP-B2b signal parsing algorithm; (2) Timestamp the obtained time series data of total ionospheric electron content, dam vibration data, and environmental meteorological data, i.e. Furthermore, heterogeneous data streams are synchronized to a unified time series using an interpolation algorithm and Formula I, where Formula I is... , where the coefficient Determined by boundary conditions; (3) Process the time series data of total electron content in the ionosphere, calculate the rate of change, the mean and standard deviation of the sliding window to obtain the TEC time series characteristics; extract the energy value, main frequency and spectral entropy of the preset frequency band from the dam vibration data as the dam vibration feature vector characterizing the vibration characteristics; perform first-order difference processing on the environmental meteorological data to obtain the meteorological change trend characteristics; The rate of change R(t) represents the change in TEC value between adjacent time points, and is calculated using the formula: R(t) = T(t) - T(t-1), where t≥2; The mean of the sliding window is the sliding mean μ(t) over time t with respect to window size ω, where ω is a positive integer. Its calculation formula is as follows: , Where t≥ω, and i is the time index within the window; The standard deviation of the sliding window is the sliding standard deviation σ(t) over time t, and its calculation formula is: , where t≥ω; The timing characteristics of TEC are represented as follows: ; The dam vibration data is denoted as V(t). Features are extracted through frequency domain analysis. It is assumed that the power spectral density is obtained by performing a Fourier transform on the signal. ,in For frequency, the formula for calculating power spectral density is: ; Let the lower and upper limits of the preset frequency band be respectively and The energy value E is calculated using the integral power spectral density formula: For discrete data, the summation formula is: ,in Frequency resolution; main frequency It is the frequency corresponding to the maximum power in the power spectrum, and the calculation formula is: That is, to search The frequency of the global maximum point; Extracting spectral entropy involves normalizing the power spectrum into a probability distribution: Then calculate the entropy: ,if =0, then define ; The vibration characteristic vector of the dam body is represented as follows: ; The environmental meteorological data is denoted as The meteorological change trend characteristics are obtained through first-order difference processing, specifically the change trend. The calculation formula is: , where t≥2; The aforementioned meteorological change trend characteristics are expressed as follows: ; (4) Input the TEC time series features and meteorological change trend features obtained in step (3) into the LSTM network for processing, establish the long-term dependency relationship between the TEC time series features and meteorological change trend features, then concatenate the output encoding of the LSTM network with the dam vibration feature vector obtained in step (3) to obtain the concatenated features, then input the concatenated features into the Transformer encoder, use the self-attention mechanism to learn the modulation effect of vibration and meteorological factors on the time series error pattern, and finally regress the real-time timing deviation prediction value through the fully connected layer; (5) Use the predicted time deviation value obtained in step (4) as the compensation amount to dynamically correct the receiver clock error obtained by solving the Beidou PPP-B2b signal, and then output accurate time information.

2. The intelligent dynamic correction method for BeiDou PPP-B2b timing error in hydropower stations according to claim 1, characterized in that: In step (1), vibration signals are collected using MEMS acceleration sensors deployed on the dam body of the hydropower station, specifically, vibration acceleration signals in three orthogonal directions (X, Y, Z), denoted as the time-domain vector a(t) = [a x (t), a y (t), a z [(t)], and then the power spectral density (PSD) of the acceleration signal in each direction is estimated based on fast Fourier transform (FFT) to obtain the corresponding vibration spectrum data.

3. The intelligent dynamic correction method for BeiDou PPP-B2b timing error in hydropower stations according to claim 1, characterized in that: The environmental meteorological data mentioned in step (1) includes data on temperature, humidity, and air pressure of the hydropower station environment, and the environmental meteorological data is expressed as follows: ,in This represents the temperature, air pressure, and humidity at time t.

4. The intelligent dynamic correction method for BeiDou PPP-B2b timing error in hydropower stations according to claim 1, characterized in that: In step (3), the time series data of total electron content in the ionosphere are processed by moving average, and then the rate of change is calculated by first-order difference.

5. The intelligent dynamic correction method for BeiDou PPP-B2b timing error in hydropower stations according to claim 1, characterized in that: The LSTM network described in step (4) includes a forget gate, an input gate, a cell state update gate, and an output gate; the forget gate is used to discard which information; the input gate is used to store which new information; the cell state update gate is used to update the old cell state to a new cell state; and the output gate is used to determine which parts of the cell state will be output. Among them, input feature concatenation: ; Input gates in an LSTM cell: , Forgotten Gate: , Candidate memory cells: , Memory cell renewal: , Output gate: , Hidden state output: , LSTM output encoding: ; The splicing features: , Location coding: ; Multi-head self-attention mechanism (the first) The layer includes the following formula: Query, key-value projection: , Scaling Dot Product Attention: , Multi-head attention splicing: , Feedforward networks and layer normalization: ; The fully connected layer regression: .

6. The intelligent dynamic correction method for BeiDou PPP-B2b timing error in hydropower stations according to claim 5, characterized in that: In step (4), the loss function used in the fully connected layer regression is: 。 7. The intelligent dynamic correction method for BeiDou PPP-B2b timing error in hydropower stations according to claim 1, characterized in that: In step (5), the observation equation for solving the clock bias of the PPP-B2b receiver is: 、 ； Receiver clock bias estimation: ; Receiver clock error after dynamic clock compensation: Where the adaptive gain factor is: ； System time estimation: ; Final output time: ,in .

8. The intelligent dynamic correction method for BeiDou PPP-B2b timing error in hydropower stations according to claim 1, characterized in that: In step (5), the performance is evaluated using a time accuracy index, the formula for which is: 。

Images