Real-time data fault hour value processing method based on intelligent prediction algorithm

Abstract

The invention discloses a real-time data fault hour processing method based on an intelligent prediction algorithm. The method comprises the steps of real-time database modeling and acquisition point attribute configuration; real-time data acquisition and preprocessing; performing conventional processing on normal data; processing the fault time value through an intelligent prediction algorithm to obtain a correction value; updating the query mechanism after replacing the correction value; and dynamically optimizing the prediction model. According to the real-time data fault hour processing method based on the intelligent prediction algorithm, the prediction precision of fault hour correction is remarkably improved through a double-layer prediction architecture in which a first-order autoregression moving average model and a second-order long and short-term memory network are combined with an attention mechanism. The first-order model provides basic correction reference for the fault time value and effectively captures the linear change rule of the data, the second-order model further mines the long-term dependency relationship of the data on the basis and focuses on the key time step characteristics, the correction value is closer to the real data through the synergistic effect of the first-order model and the second-order model, and the influence of the fault time value on the data quality is greatly reduced.

Description

Technical Field

[0001] This invention belongs to the field of real-time data fault handling technology, and relates to a real-time data fault time value processing method based on intelligent prediction algorithm. Background Technology

[0002] With the rapid development of new energy power plants, the accuracy and completeness of real-time data are crucial for their stable operation. However, due to malfunctions in signal measurement, processing, and transmission equipment at measurement points, such as sensor failures, instrument failures, signal measurement and processing device failures, signal transmission device failures, and signal transmission channel failures, fault values ​​frequently appear in the collected real-time data, severely impacting the operational efficiency and safety of the power plant. Traditional fault handling methods mainly rely on manual judgment and simple statistical analysis, which are insufficient for effectively predicting and processing fault values ​​and cannot meet the high data processing requirements of new energy power plants. Summary of the Invention

[0003] The purpose of this invention is to provide a real-time data fault time value processing method based on intelligent prediction algorithms, which solves the problem that traditional fault processing methods in the prior art are difficult to effectively predict and process fault time values ​​and cannot meet the data processing requirements of new energy power plants.

[0004] The technical solution adopted in this invention is a real-time data fault time value processing method based on intelligent prediction algorithm, comprising: Step 1: Real-time database modeling and data collection point attribute configuration; Step 2: Real-time data acquisition and preprocessing; Step 3: Perform routine processing on normal data; Step 4: The fault time value is processed by an intelligent prediction algorithm to obtain the corrected value; Step 5: Update the query mechanism after replacing the corrected value; Step 6: Dynamically optimize the prediction model.

[0005] The invention is further characterized by: Step 1 includes collecting real-time data from new energy power plants to establish a real-time database and configuring core attributes for each data collection point; The core attributes include data type, normal value range, fault value handling strategy, correction value quality code identifier, data storage strategy, prediction algorithm statistical interval, fault quality code definition, and collection period.

[0006] Step 2 includes: After collecting real-time data from the power plant, preprocessing is performed, and then fault timing value is determined. Normal data is processed in step 3, and fault timing value is processed in step 4. Preprocessing includes data validity verification and quality code parsing.

[0007] Fault time value determination: Real-time data is determined to be fault time value if it meets any of the following conditions: The value exceeds the predefined normal range by ±10%; The real-time data is in an invalid floating-point format; The quality code indicates that it is invalid.

[0008] Step 3 involves performing routine processing on the normal data before proceeding to step 5; Standard processing includes feature extraction, data storage, and feature caching.

[0009] Step 4 includes: Step 4.1: After processing the data determined to be fault time values ​​using a first-order prediction model, the first-order prediction results and feature vectors are obtained; Step 4.2: Input the first-order prediction result and feature vector into the second-order prediction model to obtain the final fault time value prediction result, i.e., the corrected value, and then proceed to step 5.

[0010] The first-order prediction model uses an autoregressive moving average model, while the second-order prediction model uses a long short-term memory network combined with an attention mechanism.

[0011] Step 5 includes: replacing data with the correction value, updating the query mechanism after dual-track storage of the data, summarizing the corrected data and normal data, issuing a warning signal when the correction value obtained in step 4 meets the set warning conditions, and not issuing a warning signal and proceeding to step 6 when the correction value obtained in step 4 does not meet the set warning conditions.

[0012] The data is replaced with the corrected value obtained in step 4, and the quality code is marked with the special identifier configured in step 1. Dual-track storage simultaneously saves the original fault value and the corrected value. The query mechanism includes default query and raw data query; The default query returns the corrected value; the raw data query returns the original value at the time of the fault.

[0013] Step 6 involves updating and optimizing the first-order and second-order prediction models based on the set optimization cycle and the newly added running data, and then executing step 2.

[0014] The beneficial effects of this invention are: This invention significantly improves the prediction accuracy of fault time value correction through a two-layer prediction architecture combining a first-order autoregressive moving average model and a second-order long short-term memory network with an attention mechanism. The first-order model provides a basic correction reference for fault time values, effectively capturing the linear variation pattern of the data. The second-order model further mines the long-term dependencies of the data and focuses on key time step features. The synergistic effect of the two makes the corrected value closer to the real data, greatly reducing the impact of fault time values ​​on data quality. In terms of model adaptability, the design of the two-layer prediction architecture can flexibly cope with the complex characteristics of real-time data from new energy power plants. Whether it is stable linear data change or nonlinear data pattern with fluctuations, it can effectively adapt through the complementary advantages of different models, and meet the data processing needs of different equipment and different operating scenarios of power plants. The real-time performance of data processing is fully guaranteed. Through real-time database modeling, the integrated operation of data acquisition, storage and processing is realized, which can efficiently support the high-concurrency processing of massive sensor data, greatly shorten the operation delay from data acquisition to processing, ensure that fault values ​​can be quickly identified and corrected, and provide timely and reliable data support for the real-time monitoring and operation decision of power plants. Data reliability is significantly enhanced through a dual-track storage mechanism. This mechanism simultaneously stores both the original fault time values ​​and corrected values, ensuring data traceability while supporting flexible selection of query types based on actual needs. Whether it's corrected values ​​required for daily data analysis or original fault time values ​​needed for troubleshooting, they can be retrieved on demand, meeting the data usage needs of different scenarios. Furthermore, the dynamic optimization mechanism of the predictive model updates model parameters and structure periodically by incorporating newly added operational data, ensuring the model maintains excellent predictive performance and continuously adapts to changes in the operating status of power plant equipment, further improving the long-term effectiveness and stability of the entire data processing solution. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the real-time data fault time value processing method based on intelligent prediction algorithm of the present invention. Figure 2 This is a schematic diagram of the data processing flow of the first-order prediction model in this invention; Figure 3 This is a schematic diagram of the structure of the second-order prediction model in this invention. Detailed Implementation

[0016] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0017] The real-time data fault time-value processing method based on intelligent prediction algorithm includes the following steps: Step 1: Real-time database modeling and data collection point attribute configuration; Real-time data is collected from new energy power plants to build a real-time database, and core attributes are configured for each data collection point. Core attributes include data type (e.g., switch quantity / integer / floating point), normal value range (e.g., current: 200-250A), fault time value handling strategy (enable intelligent prediction correction value replacement), correction value quality code identifier (e.g., 0x0A), data storage strategy (dual-track storage of fault time value and correction value), prediction algorithm statistical interval (e.g., the first 5 minutes), fault quality code definition (e.g., 0x80), and acquisition cycle (e.g., 1 second). Step 2: Real-time data acquisition and preprocessing; After collecting real-time data from the power plant via industrial bus (such as Modbus TCP / IP, OPC DA / UA, etc.), the data is preprocessed and then fault time value is determined. Normal data is processed in step 3, and fault time value is processed in step 4. Preprocessing includes: Data validity verification: Based on the normal value range of the data collection point attributes, floating-point numbers are valid floating-point numbers that conform to the IEEE standard; Quality code parsing: Parses the quality code field in real-time data (such as OPC standard, IEC 61850 standard); Fault value determination: A fault value is determined to be a value that meets any of the following conditions: the value exceeds the predefined normal range by ±10%; the real-time data is an invalid floating-point number format; the quality code indication is invalid (e.g., 0x80). Step 3: Perform routine processing on normal data; After performing routine processing on the normal data, proceed to step 5; Routine procedures include: Feature extraction: Calculate and extract features (such as moving mean, moving variance, rate of change, peak factor, etc.); Data storage: Data is stored in the real-time database according to the attribute configuration of the collection point, and the quality code is marked as 0x00 (normal). Feature caching: Key features are stored in the model input cache pool for subsequent prediction model training; Step 4: The fault time value is processed by an intelligent prediction algorithm to obtain the corrected value; Step 4.1, First-order prediction: The first-order prediction model, namely the autoregressive moving average model (ARMA), is used to process the data value determined to be the fault time value to obtain the first-order prediction result and feature vector; The Autoregressive Moving Average (ARMA) model determines the order of the model through the Autocorrelation Function (ACF) and the Partial Autocorrelation Function (PACF). ARMA (Autoregressive Moving Average Model) is a time series data forecasting model that combines autoregressive (AR) and moving average (MA) to effectively capture linear relationships in time series data. The formula for the ARMA model is: , in, It is time The observed values, and These are the parameters for autoregression and moving average, respectively. It is an error term; Order representation: ARMA(p,q), where p is the AR order and q is the MA order; Order calculation: Autocorrelation function (ACF); Definition: Measuring a sequence With lagged k-period sequence The linear correlation is given by the formula: , Function: To identify the order q of MA(q), the ACF is truncated after lag of order q (suddenly approaches 0); Partial autocorrelation function (PACF); Definition: After controlling for the effects of intermediate lags, measuring and The direct correlation is expressed by the formula: , Function: Identifies the order p of AR(p), and PACF is truncated after a lag of p (suddenly approaches 0).

[0018] Predicted output: Based on the ARMA(p,q) formula after determining the order, the model parameters are fitted using historical data. and Preliminary fault time value prediction results were obtained.

[0019] Step 4.2, Second-order prediction: Input the first-order prediction result and feature vector into the second-order prediction model, namely the Long Short-Term Memory (LSTM) network combined with the attention mechanism, to make predictions and obtain the final fault time value prediction result, i.e., the corrected value, and then proceed to step 5. LSTM can effectively capture long-term dependencies in time series data, while the attention mechanism can enhance the model's focus on key time steps and improve prediction accuracy. The LSMT model formula is as follows: The LSTM update formula is as follows: , , , , , , in, , , These are the forget gate, input gate, and output gate. and These are candidate states and cell states. It is in a hidden state. It is the sigmoid activation function.

[0020] Attention mechanism: The formula for the attention mechanism is: , , in, It is the energy value at time step t. It is attention weight. It is a context vector.

[0021] The first-order prediction results and the extracted feature vectors are input into the LSTM model, and combined with the attention mechanism, the final fault time value prediction result is obtained. Through the attention mechanism, the model can automatically identify key time steps and improve the accuracy of prediction.

[0022] Step 5: Update the query mechanism after replacing the corrected value; The data is replaced using the correction value. The data is stored in a dual-track system and then updated and queried. The corrected data and normal data are summarized. When the correction value obtained in step 4 meets the set warning conditions, a warning signal is issued. When the correction value obtained in step 4 does not meet the set warning conditions, no warning signal is issued and step 6 is executed. Data replacement: Replace the fault value with the corrected value obtained in step 4, and mark the quality code with the special identifier (0x0A) configured in step 1. Dual-track storage: Simultaneously saves the original fault value (with 0x80 quality code) and the correction value (with 0x0A quality code); The query mechanism includes default query and raw data query; Default query: Returns corrected value (quality code 0x0A); Raw data query: Specify the original fault value to return via API extended parameters; Based on the second-order prediction results, when a serious anomaly is detected, an early warning signal is issued to remind operators to take measures, such as adjusting equipment parameters and performing equipment maintenance, in order to avoid more serious failures. Step 6: Dynamically optimize the prediction model; After updating and optimizing the first-order and second-order prediction models based on the set optimization cycle and the newly added running data, proceed to step 2; Regularly collect new operational data to update and optimize the first-order prediction model (ARMA) and the second-order prediction model (LSTM combined with attention mechanism) to improve the accuracy and adaptability of the model; for example, adjust the parameters of the ARMA model based on the new operational data, and optimize the structure of the LSTM model and the weight allocation of the attention mechanism.

[0023] This invention automates the processing of fault time values ​​through database modeling, intelligent prediction, and dynamic replacement mechanisms, thereby improving data quality and system reliability. It aims to solve the challenge of processing real-time data fault time values ​​(fault time values ​​refer to the values ​​collected at measurement points when sensor, instrument, signal measurement and processing device, signal transmission device, or signal transmission channel malfunctions occur) in power monitoring systems. This invention is applicable to industrial real-time database management systems used in monitoring scenarios of critical facilities such as generator sets and power transmission and transformation equipment. By acquiring and analyzing real-time data, this invention identifies fault time values ​​and uses intelligent prediction algorithms to obtain corrected values ​​that approximate the true values. These corrected values ​​mask the fault time values, thereby reducing their impact on power monitoring systems, power data analysis platforms, and equipment diagnostic analysis models, and improving the stability and data reliability of the power system. This invention configures attributes for each data acquisition point during real-time database modeling, including normal value range, fault handling strategy, and corrected value quality code identifier. It collects real-time data from power plants, determines whether the data represents a fault value through quality code parsing and predefined ranges. If it's a normal value, key features are extracted and stored in the database according to the acquisition point attributes. If it's a fault value, a two-order prediction algorithm is used: First-order prediction: an ARMA model is used, with the model order determined by ACF and PACF; Second-order prediction: the ARMA results and feature vectors are input into an LSTM+attention mechanism model. Then, the fault value is replaced with a second-order corrected value and marked with a special quality code, achieving dual-track storage of original and corrected values ​​and on-demand querying. Finally, new data is periodically used to incrementally update and fully optimize the prediction model. This invention enables rapid identification of fault values ​​and corrected value replacement, significantly reducing the impact of faulty system operation and data analysis platforms. Example 1 This embodiment proposes a real-time data fault time value processing method based on intelligent prediction algorithms, such as... Figure 1As shown, it includes: Step 1: Real-time database modeling and data collection point attribute configuration; Step 2: Real-time data acquisition and preprocessing; Step 3: Perform routine processing on normal data; Step 4: The fault time value is processed by an intelligent prediction algorithm to obtain the corrected value; Step 5: Update the query mechanism after replacing the corrected value; Step 6: Dynamically optimize the prediction model.

[0024] Example 2 This embodiment proposes a real-time data fault time value processing method based on intelligent prediction algorithms, such as... Figure 1 As shown, it includes: Step 1: Real-time database modeling and data collection point attribute configuration; Real-time data is collected from new energy power plants to build a real-time database, and core attributes are configured for each data collection point. The core attributes include data type, normal value range, fault value handling strategy, correction value quality code identifier, data storage strategy, prediction algorithm statistical interval, fault quality code definition, and collection period.

[0025] Step 2: Real-time data acquisition and preprocessing; Step 3: Perform routine processing on normal data; Step 4: The fault time value is processed by an intelligent prediction algorithm to obtain the corrected value; Step 5: Update the query mechanism after replacing the corrected value; Step 6: Dynamically optimize the prediction model.

[0026] Example 3 This embodiment proposes a real-time data fault time value processing method based on intelligent prediction algorithms, such as... Figure 1 As shown, it includes: Step 1: Real-time database modeling and data collection point attribute configuration; Step 2: Real-time data acquisition and preprocessing; After collecting real-time data from the power plant, preprocessing is performed, and then fault timing value is determined. Normal data is processed in step 3, and fault timing value is processed in step 4. Preprocessing includes data validity verification and quality code parsing.

[0027] Fault time value determination: Real-time data is determined to be fault time value if it meets any of the following conditions: The value exceeds the predefined normal range by ±10%; The real-time data is in an invalid floating-point format; The quality code indicates that it is invalid.

[0028] Step 3: Perform routine processing on normal data; Step 4: The fault time value is processed by an intelligent prediction algorithm to obtain the corrected value; Step 5: Update the query mechanism after replacing the corrected value; Step 6: Dynamically optimize the prediction model.

[0029] Example 4 This embodiment proposes a real-time data fault time value processing method based on intelligent prediction algorithms, such as... Figure 1 As shown, it includes: Step 1: Real-time database modeling and data collection point attribute configuration; Step 2: Real-time data acquisition and preprocessing; Step 3: Perform routine processing on normal data; Step 3 involves performing routine processing on the normal data before proceeding to step 5; Standard processing includes feature extraction, data storage, and feature caching.

[0030] Step 4: The fault time value is processed by an intelligent prediction algorithm to obtain the corrected value; Step 5: Update the query mechanism after replacing the corrected value; Step 6: Dynamically optimize the prediction model.

[0031] Example 5 This embodiment proposes a real-time data fault time value processing method based on intelligent prediction algorithms, such as... Figure 1 As shown, it includes: Step 1: Real-time database modeling and data collection point attribute configuration; Step 2: Real-time data acquisition and preprocessing; Step 3: Perform routine processing on normal data; Step 4: The fault time value is processed by an intelligent prediction algorithm to obtain the corrected value; Step 4.1: After processing the data determined to be fault time values ​​using a first-order prediction model, the first-order prediction results and feature vectors are obtained; Step 4.2: Input the first-order prediction result and feature vector into the second-order prediction model to obtain the final fault time value prediction result, i.e., the corrected value, and then proceed to step 5.

[0032] The first-order prediction model uses an autoregressive moving average model, while the second-order prediction model uses a long short-term memory network combined with an attention mechanism.

[0033] Step 5: Update the query mechanism after replacing the corrected value; Step 6: Dynamically optimize the prediction model.

[0034] Example 6 This embodiment proposes a real-time data fault time value processing method based on intelligent prediction algorithms, such as... Figure 1 As shown, it includes: Step 1: Real-time database modeling and data collection point attribute configuration; Step 2: Real-time data acquisition and preprocessing; Step 3: Perform routine processing on normal data; Step 4: The fault time value is processed by an intelligent prediction algorithm to obtain the corrected value; Step 5: Update the query mechanism after replacing the corrected value; The data is replaced using the correction value. The data is stored in a dual-track system and then updated and queried. The corrected data and normal data are summarized. When the correction value obtained in step 4 meets the set warning conditions, a warning signal is issued. When the correction value obtained in step 4 does not meet the set warning conditions, no warning signal is issued and step 6 is executed.

[0035] The data is replaced with the corrected value obtained in step 4, and the quality code is marked with the special identifier configured in step 1. Dual-track storage simultaneously saves the original fault value and the corrected value. The query mechanism includes default query and raw data query; The default query returns the corrected value; the raw data query returns the original value at the time of the fault.

[0036] Step 6: Dynamically optimize the prediction model.

[0037] Example 7 This embodiment proposes a real-time data fault time value processing method based on intelligent prediction algorithms, such as... Figure 1 As shown, it includes: Step 1: Real-time database modeling and data collection point attribute configuration; Step 2: Real-time data acquisition and preprocessing; Step 3: Perform routine processing on normal data; Step 4: The fault time value is processed by an intelligent prediction algorithm to obtain the corrected value; Step 5: Update the query mechanism after replacing the corrected value; Step 6: Dynamically optimize the prediction model.

[0038] After updating and optimizing the first-order and second-order prediction models based on the set optimization cycle and the newly added running data, proceed to step 2.

[0039] The relevant data of a certain embodiment of the present invention implemented in a 20MW photovoltaic power plant's #3 inverter (model: Sungrow PowerTitan 1500V) are as follows: The data acquisition dimensions and devices are shown in Table 1: Table 1

[0040] To focus on monitoring power fluctuation faults caused by abnormal DC-side current, the attributes for the DC-side current acquisition points are configured in the real-time database as shown in Table 2, and a real-time data fault time value processing method based on intelligent prediction algorithm is enabled: Table 2

[0041] The current time is 00:00:29 on June 1, 20XX. The real-time running data (sampling frequency 1Hz) from 00:00:00 to 00:00:29 has been collected and stored in the real-time database, as shown in Table 3.

[0042] Table 3. Original dataset (30 records)

[0043] The real-time data for the current time is shown in Table 4: Table 4

[0044] The value is valid and within the normal range, the quality code is normal, proceed to the next normal value processing step; The temporal characteristics are shown in Table 5 (sliding window = 5 seconds); Table 5

[0045] The environmental association characteristics are shown in Table 6: Table 6

[0046] The equipment status characteristics are shown in Table 7; Table 7

[0047] The above key feature values ​​are stored in the model input cache pool for subsequent prediction model training.

[0048] In step 4, when the real-time data for the next moment, as shown in Table 8, arrives: Table 8

[0049] The current real-time current data is shown in Table 9: Table 9

[0050] Based on the data preprocessing conditions in step 2: Values ​​exceeding the 200-250A range and with a quality code of 0x80 are marked as fault values.

[0051] Process the data that is determined to be faulty: 2.1, adopting as follows Figure 2 The Autoregressive Moving Average (ARMA) model shown is used for prediction, and the order of the model is determined by the autocorrelation function (ACF) and the partial autocorrelation function (PACF).

[0052] 2.1.1 Model order determination; Based on the preprocessed current series (239.05A~238.61A), ACF / PACF analysis was performed: ACF values ​​were calculated for lags 0 to 4, and truncated at lag 2 → moving average order q=2; PACF values ​​were calculated for lags 0 to 4, and truncated at lag 2 → autoregressive order p=2; the optimal ARMA model order was selected as ARMA(2,2). 2.1.2 Parameter estimation; The model parameters were fitted using the least squares method: , Wherein: Autoregressive coefficient: (Weight of the influence of the current in the first second). (Weight of the current in the first 2 seconds); Moving average coefficient: (The correction weight for the error in the first second). (Correction weights for the first 2 seconds of error); residual After iterative processing of the original data sequence on ARMA(2,2), the expected value can be predicted at 00:00:30.

[0053] 2.2.3 Correction value: 238.80A; 2.2 Second-order prediction: Input the first-order prediction result and feature vector as follows: Figure 3 The Long Short-Term Memory (LSTM) network shown is used in conjunction with an attention mechanism model for prediction.

[0054] 2.2.1 Construct the feature vectors as shown in Table 10; Table 10

[0055] The 12-dimensional vector input values ​​are shown in Table 11: Table 11

[0056] 2.2.2 Model Structure; Input layer: 12-dimensional feature vectors; LSTM layer: 2 layers × 64 neurons (ReLU activation); Attention layer: 4-head self-attention mechanism; Output layer: 1-dimensional failure probability (sigmoid activation); 2.2.3 The distribution of attention weights is shown in Table 12: Table 12

[0057] 2.2.4 Prediction result: 238.98A; The records stored in step 5 are shown in Table 13: Table 13

[0058] When querying historical data for data analysis, calling Get Archived Values ​​returns a dataset with corrected values, as shown in Table 14. Table 14

[0059] When you need to use the original dataset with fault values ​​as shown in Table 15, call GetArchivedValuesIncludeFault: Table 15

[0060] In step 6, the first-order prediction model (ARMA) and the second-order prediction model (LSTM combined with attention mechanism) are updated and optimized according to the optimization cycle set as shown in Table 16 and new running data are collected periodically to improve the accuracy and adaptability of the model.

[0061] Table 16

Claims

1. A real-time data fault time-value processing method based on intelligent prediction algorithm, characterized in that, include: Step 1: Real-time database modeling and data collection point attribute configuration; Step 2: Real-time data acquisition and preprocessing; Step 3: Perform routine processing on normal data; Step 4: The fault time value is processed by an intelligent prediction algorithm to obtain the corrected value; Step 5: Update the query mechanism after replacing the corrected value; Step 6: Dynamically optimize the prediction model.

2. The real-time data fault time value processing method based on intelligent prediction algorithm according to claim 1, characterized in that, Step 1 includes collecting real-time data from new energy power plants to establish a real-time database and configuring core attributes for each data collection point. The core attributes include data type, normal value range, fault value handling strategy, correction value quality code identifier, data storage strategy, prediction algorithm statistical interval, fault quality code definition, and collection period.

3. The real-time data fault time value processing method based on intelligent prediction algorithm according to claim 1, characterized in that, Step 2 includes: After collecting real-time data from the power plant, preprocessing is performed, and then fault timing value is determined. Normal data is processed in step 3, and fault timing value is processed in step 4. Preprocessing includes data validity verification and quality code parsing.

4. The real-time data fault time value processing method based on intelligent prediction algorithm according to claim 3, characterized in that, The fault time value is determined by real-time data satisfying any of the following conditions: The value exceeds the predefined normal range by ±10%; The real-time data is in an invalid floating-point format; The quality code indicates that it is invalid.

5. The real-time data fault time value processing method based on intelligent prediction algorithm according to claim 1, characterized in that, Step 3 includes performing routine processing on normal data before proceeding to step 5; Standard processing includes feature extraction, data storage, and feature caching.

6. The real-time data fault time value processing method based on intelligent prediction algorithm according to claim 1, characterized in that, Step 4 includes: Step 4.1: After processing the data determined to be fault time values ​​using a first-order prediction model, the first-order prediction results and feature vectors are obtained; Step 4.2: Input the first-order prediction result and feature vector into the second-order prediction model to obtain the final fault time value prediction result, i.e., the corrected value, and then proceed to step 5.

7. The real-time data fault time-value processing method based on intelligent prediction algorithm according to claim 6, characterized in that, The first-order prediction model uses an autoregressive moving average model, while the second-order prediction model uses a long short-term memory network combined with an attention mechanism.

8. The real-time data fault time value processing method based on intelligent prediction algorithm according to claim 1, characterized in that, Step 5 includes: replacing data with a correction value, updating the query mechanism after dual-track storage of the data, summarizing the corrected data and normal data, issuing a warning signal when the correction value obtained in step 4 meets the set warning conditions, and not issuing a warning signal and proceeding to step 6 when the correction value obtained in step 4 does not meet the set warning conditions.

9. The real-time data fault time value processing method based on intelligent prediction algorithm according to claim 8, characterized in that, The data is replaced with the fault value obtained in step 4, and the quality code is marked with the special identifier configured in step 1. The dual-track storage simultaneously saves the original fault time value and the correction value; The query mechanism includes default query and raw data query; The default query returns the corrected value; the raw data query returns the original value at the time of the fault.

10. The real-time data fault time value processing method based on intelligent prediction algorithm according to claim 1, characterized in that, Step 6 includes updating and optimizing the first-order prediction model and the second-order prediction model based on the set optimization cycle and the newly added running data, and then executing step 2.

Images