A model and data double-driven CVT error measurement method and system

Abstract

This invention provides a CVT error measurement method based on a model and data dual-driven approach. It acquires real-time CVT environmental factor monitoring data and real-time CVT metrological monitoring data; constructs a metrological error mechanism model; and analyzes the real-time CVT metrological monitoring data based on the metrological error mechanism model to determine the ideal metrological error. It then acquires a fully trained error ensemble model and predicts additional metrological errors based on the real-time CVT environmental factor monitoring data to determine the target additional metrological error. Finally, it determines the actual CVT metrological error based on the target additional metrological error and the mechanistic metrological error. This invention measures the actual CVT metrological error through a metrological error mechanism model and an error ensemble model, enhancing the accuracy of the measured values ​​through model- and data-driven measurement, thus achieving precise measurement of the actual CVT metrological error.

Description

Technical Field

[0001] This invention proposes a CVT error measurement method based on a dual-drive approach of model and data, which belongs to the field of data analysis in the power industry. Background Technology

[0002] The power grid is growing rapidly. CVT (Continuous Voltage Meter), as a conventional electricity metering device, is widely used in the State Grid due to its excellent insulation performance and cost advantages. However, its operating conditions are highly variable, and metering errors are affected by various factors, including the external environment and its own internal components. Over long-term operation, problems such as decreased insulation performance, excessive error, and high failure rates can arise, directly impacting the accuracy of electricity metering and the fairness of power system transactions. Therefore, achieving accurate real-time online measurement of CVT metering errors, improving the system's accurate metering and maintenance early warning capabilities, and thus protecting the long-term stable operation of the power system, is a major challenge that urgently needs to be addressed in the smart grid field. CVT metering errors are affected by environmental temperature, humidity, environmental pollution, electromagnetic fields, and secondary loads. Traditional model-driven solutions to this problem require analyzing multiple factors to construct a mathematical model for linear superposition. This not only ignores the coupling information between various factors and fails to consider the underlying physical mechanisms, leading to inaccurate results, but also makes the multi-factor model difficult to solve due to its complexity.

[0003] Due to limitations in model-driven approaches, and with the rapid development of artificial intelligence, historical CVT operation environmental sample data can be extracted and used for data mining. Machine learning and deep learning can be employed to build a real-time calculation model for additional errors based on data. When the data sample size is large and the overall data quality is high, the results obtained using data-driven approaches are significantly better. However, the power grid's handling of relevant anomaly data is still in its initial stages, with issues such as insufficient overall sample size and low overall data quality. Regarding the real-time calculation of CVT metering errors, considering the combined influence of multiple factors, relying solely on model-driven approaches for overall metering error correction presents challenges such as the difficulty in constructing a model that considers multiple factors and inaccurate results. Conversely, choosing only data-driven approaches for operational status assessment results in low accuracy and poor robustness for a single data-driven model.

[0004] Therefore, this invention provides a CVT error measurement method based on both model and data. It uses a dual-drive approach of data and model to measure CVT error, solving the problems of model-driven measurement, which is difficult to construct, and the low accuracy and poor robustness of single data-driven model measurement. Summary of the Invention

[0005] In view of this, in order to solve the problems of existing model-driven measurement, which is difficult to construct, and data-driven measurement, which suffers from low model accuracy and poor robustness due to insufficient data sample quality, it is necessary to provide an environmentally friendly and insulating gas discharge decomposition characteristic detection and evaluation method to simplify the detection scheme and form regular conclusions for practical application.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention provides a CVT error measurement method based on a model and data dual-drive approach, comprising:

[0008] Acquire real-time CVT environmental factor monitoring data and real-time CVT metering monitoring data;

[0009] A measurement error mechanism model is constructed, and the ideal measurement error is determined by performing measurement error analysis on the real-time CVT measurement monitoring data based on the measurement error mechanism model.

[0010] Obtain a fully trained error ensemble model, and based on the fully trained error ensemble model, perform additional measurement error prediction on the real-time CVT environmental factor monitoring data to determine the target additional measurement error;

[0011] The actual measurement error of CVT is determined based on the target measurement additional error and the mechanism measurement error.

[0012] In some possible implementations, obtaining the fully trained error ensemble model includes:

[0013] Collect CVT environmental factor monitoring data, and construct an environmental factor data sample set based on the CVT environmental factor monitoring data;

[0014] An error ensemble model is constructed, and the error ensemble model is trained based on the environmental factor data sample set to obtain a fully trained error ensemble model.

[0015] In some possible implementations, the error ensemble model includes a base learner layer and a meta-learner layer;

[0016] The base learner layer includes at least three algorithm models as base learners for the base learner layer.

[0017] In some possible implementation methods, an error ensemble model is constructed, and the error ensemble model is trained based on the environmental factor data sample set to obtain a fully trained error ensemble model, including:

[0018] Based on the environmental factor data sample set, the base learner of the base learner layer is trained by K-fold cross-validation to obtain a fully trained base learner layer.

[0019] The environmental factor data sample set is input into the fully trained base learner layer to obtain the first meta-dataset;

[0020] Based on the weight allocation model, the data in the first meta-dataset are weighted according to the time dimension and precision to obtain the second meta-dataset;

[0021] The algorithm model of the meta-learner layer is trained based on the second meta-dataset to obtain a fully trained meta-learner layer, that is, a fully trained error ensemble model.

[0022] In some possible implementations, the base learner layer further includes a weight allocation model; the weight allocation model includes a time weight function and a precision weight function; the method further includes:

[0023] The time weight of each base learner's prediction result is determined according to the time weighting function.

[0024] The accuracy weight of each base learner is determined based on the accuracy weight function and the prediction accuracy of each base learner.

[0025] In some possible implementations, the acquisition of real-time CVT environmental factor monitoring data and real-time CVT metering monitoring data includes:

[0026] The CVT metering and monitoring data are the amplitude and phase data of the CVT collected by the operation and maintenance sensors;

[0027] The CVT environmental factor monitoring data includes data collected by the operating parameter monitoring system, including climate and environmental data and operating condition data;

[0028] The climate and environmental data include the temperature and humidity of the CVT operating environment;

[0029] Operating condition data includes CVT surface contamination, CVT ambient electric field, and CVT secondary load.

[0030] In some possible implementations, the target additional measurement error is determined by predicting additional measurement errors based on the fully trained error ensemble model of the real-time CVT environmental factor monitoring data, including:

[0031] The real-time CVT environmental factor monitoring data is input into the base learner layer, and the measurement additional error is predicted based on the base learner layer to obtain the first target meta dataset;

[0032] The weights of the first target metadata dataset are assigned based on the weight assignment model to obtain the second target metadata dataset.

[0033] The second target meta-dataset is input into the meta-learner layer to predict the target measurement additional error.

[0034] In some possible implementations, the construction of a measurement error mechanism model and the determination of the ideal measurement error based on the measurement error mechanism model of the real-time CVT measurement monitoring data include:

[0035] The amplitude and phase data of the CVT are input into the measurement error mechanism model, and the ideal measurement error is determined by the formula of the measurement error mechanism model.

[0036] In some possible implementations, the weights of each base learner are determined based on the accuracy weight function and the prediction accuracy of each base learner, including:

[0037] The mean absolute percentage error of the prediction results of each base learner is calculated based on the prediction results of each base learner.

[0038] The weights of each base learner's results are determined based on the mean absolute percentage error of each base learner and the accuracy weighting function.

[0039] On the other hand, the present invention also provides a CVT error measurement system based on a model and data dual-drive approach, comprising:

[0040] The data acquisition unit is used to acquire real-time CVT environmental factor monitoring data and real-time CVT metering monitoring data.

[0041] The mechanism model construction unit is used to construct a measurement error mechanism model and, based on the measurement error mechanism model, perform measurement error analysis on the real-time CVT measurement monitoring data to determine the ideal measurement error.

[0042] The error ensemble model acquisition unit is used to acquire a fully trained error ensemble model, and based on the fully trained error ensemble model, to perform additional measurement error prediction on the real-time CVT environmental factor monitoring data to determine the target additional measurement error.

[0043] The actual measurement error prediction unit determines the actual measurement error of the CVT based on the target measurement additional error and the mechanism measurement error.

[0044] Compared with existing technologies, the beneficial effects of this invention include: a CVT error measurement method based on model and data dual-drive. The data drive adopts the measurement error ensemble model proposed in this invention, which is an improved Stacking ensemble model. By adding weights to the ensemble model, the prediction results of the base learner layer are weighted to obtain the second meta-dataset, which makes the data quality provided to the meta-learner layer better and the prediction results of the meta-learner more accurate. At the same time, multiple algorithm models are used in the base learner layer to improve the generalization ability and robustness of the base learner through the ensemble model.

[0045] Furthermore, this invention uses a measurement error mechanism model to predict rational measurement errors, solving the problem that the model needs to consider multiple factors and is difficult to construct, and the results obtained by the model are inaccurate. By using a model-driven measurement error mechanism model and a data-driven measurement error integration model, the accuracy of CVT error measurement is improved through the dual drive of data and model, and the precise measurement of CVT measurement error is realized. Attached Figure Description

[0046] Figure 1 A flowchart of an embodiment of the CVT error measurement method based on model and data dual-drive provided by the present invention;

[0047] Figure 2 This is a structural diagram of an embodiment of the CVT error measurement system based on model and data dual-drive provided by the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0049] It should be understood that the illustrative drawings are not drawn to scale. The flowcharts used in this invention illustrate operations implemented according to some embodiments of the invention. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or performed simultaneously. Furthermore, those skilled in the art, guided by the content of this invention, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0050] Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor systems and / or microcontroller systems.

[0051] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0052] This invention provides a CVT error measurement method based on a dual-drive approach of model and data, which will be described below.

[0053] Figure 1 A flowchart of an embodiment of the CVT error measurement method based on model and data dual-drive provided by the present invention includes the following steps:

[0054] S101. Obtain real-time CVT environmental factor monitoring data and real-time CVT metering monitoring data;

[0055] S102. Construct a measurement error mechanism model, and perform measurement error analysis on the real-time CVT measurement monitoring data based on the measurement error mechanism model to determine the ideal measurement error;

[0056] S103. Obtain a fully trained error ensemble model, and based on the fully trained error ensemble model, predict the additional measurement error of the real-time CVT environmental factor monitoring data to determine the target additional measurement error.

[0057] S104. Determine the actual measurement error of CVT based on the target measurement additional error and the mechanism measurement error.

[0058] It should be noted that the CVT (Capacitance Type Voltage Transformer) is a type of voltage transformer that uses series capacitors to divide the voltage, and then electromagnetic transformers to reduce the voltage and isolate it. It is used as a voltage transformer for meters, relay protection, etc.

[0059] It should be further noted that the error ensemble model is an improved Stacking ensemble model.

[0060] Compared with existing technologies, a CVT error measurement method based on model and data dual-drive is proposed in this invention. The data-driven approach adopts the measurement error ensemble model, which is an improved Stacking ensemble model. By adding weights to the ensemble model, the prediction results of the base learner layer are weighted to obtain the second meta-dataset, which improves the data quality provided to the meta-learner layer and makes the prediction results of the meta-learner more accurate. At the same time, multiple algorithm models are used in the base learner layer to improve the generalization ability and robustness of the base learner through the ensemble model.

[0061] Furthermore, this invention uses a measurement error mechanism model to predict rational measurement errors, solving the problem that the model needs to consider multiple factors and is difficult to construct, and the results obtained by the model are inaccurate. By using a model-driven measurement error mechanism model and a data-driven measurement error integration model, the accuracy of CVT error measurement is improved through the dual drive of data and model, and the precise measurement of CVT measurement error is realized.

[0062] In this embodiment of the invention, obtaining a fully trained error ensemble model includes:

[0063] Collect CVT environmental factor monitoring data, and construct an environmental factor data sample set based on the CVT environmental factor monitoring data;

[0064] An error ensemble model is constructed, and the error ensemble model is trained based on the environmental factor data sample set to obtain a fully trained error ensemble model.

[0065] It should be noted that the CVT environmental factor monitoring data is data collected by the operating parameter monitoring system. One part of it is climate environmental data, including the temperature and humidity of the CVT operating environment; the other part is operating condition data, including surface contamination, environmental electric field and secondary load, etc.

[0066] In this embodiment of the invention, the error ensemble model includes a base learner layer and a meta learner layer;

[0067] The base learner layer includes at least three algorithm models as base learners for the base learner layer.

[0068] In specific embodiments, the model algorithm can be XGBoost, Ridge Regression, RF, SVR, GBDT, or LSTM.

[0069] After decomposition, CVT measurement error data can be categorized into three frequency bands: low-frequency, mid-frequency, and high-frequency, exhibiting complex overall characteristics. The low-frequency signal components show relatively clear regularity, changing slowly and with smooth waveforms. Gradient Boosted Decision Tree (GBDT) and its improved algorithm, Extreme Gradient Boosting (XGBoost), demonstrate strong generalization ability while maintaining sufficient accuracy for application requirements, making them suitable for real-time calculation of low-frequency signals in CVT measurement errors. The high-frequency signal component exhibits some randomness and fluctuation. The Long Short-Term Memory (LSTM) algorithm can uncover more data fluctuation patterns in long-term time-series measurement error real-time calculations, and model training for the high-frequency signal component can be completed through nonlinear mapping. Support Vector Regression (SVR) and Random Forest (RF) algorithms can effectively mine high-frequency signals after data decomposition. SVM demonstrates strong generalization and fitting capabilities, effectively addressing regression problems of high-dimensional features, while RF is less prone to overfitting and exhibits strong noise resistance. The Ridge Regression Algorithm can process the low-frequency component signals after decomposition, and it has a fast learning speed and high computational efficiency.

[0070] In this embodiment of the invention, the step of constructing an error ensemble model and training the error ensemble model based on the environmental factor data sample set to obtain a fully trained error ensemble model includes:

[0071] Based on the environmental factor data sample set, the base learner of the base learner layer is trained by K-fold cross-validation to obtain a fully trained base learner layer.

[0072] The environmental factor data sample set is input into the fully trained base learner layer to obtain the first meta-dataset;

[0073] Based on the weight allocation model, the data in the first meta-dataset are weighted according to the time dimension and precision to obtain the second meta-dataset;

[0074] The algorithm model of the meta-learner layer is trained based on the second meta-dataset to obtain a fully trained meta-learner layer, that is, a fully trained error ensemble model.

[0075] In a specific embodiment, the environmental factor data sample set is divided into a training set and a test set. The training set is then divided into K parts by K folding, which serve as the base data for the K base learners of the primitive learner layer. K-fold cross-validation training is then performed to obtain a fully trained base learner layer. Based on the base data and the fully trained base learner layer, prediction is performed to obtain a first meta-dataset. Based on the weight allocation model, the data in the meta-dataset is weighted according to the time dimension and accuracy to obtain a second meta-dataset.

[0076] The second meta-dataset is used as the training set for the meta-learner layer to train the meta-learner layer. The trained meta-learner layer is then tested using a test set. If the test result of the meta-learner layer meets the error threshold, a fully trained error ensemble model is obtained. Otherwise, training continues until a fully trained error ensemble model is obtained.

[0077] It should be noted that the error thresholds include mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE). The specific error threshold calculation formulas are as follows:

[0078]

[0079]

[0080]

[0081] In the formula, y i For the actual quantity, The actual measurement error is represented by n, where n is the number of samples.

[0082] In this embodiment of the invention, the base learner layer further includes a weight allocation model; the weight allocation model includes a time weight function and a precision weight function; the method further includes:

[0083] The time weight of each base learner's prediction result is determined according to the time weighting function.

[0084] The accuracy weight of each base learner is determined based on the accuracy weight function and the prediction accuracy of each base learner.

[0085] In a specific embodiment, the time weight calculation process for the prediction results of each base learner is as follows:

[0086]

[0087] The time weights are determined by the optimal time scaling factor. A weighted prediction is obtained by weighting the K predictions of the u-th base learner. This weighted prediction is then compared with the prediction of the additional measurement error by the meta-learner layer, and finally with the prediction of the additional measurement error by the unweighted meta-learner layer. If the prediction improves, the iteration continues in the direction of increasing T until the error no longer decreases; otherwise, the iteration continues in the direction of decreasing T until the optimal time scaling factor T is found.

[0088] The precision weights are determined based on the average absolute percentage error of each base learner, and the specific calculation formula is as follows:

[0089]

[0090] in, This represents the mean absolute percentage error of the u-th base learner.

[0091] In this embodiment of the invention, acquiring real-time CVT environmental factor monitoring data and real-time CVT metering monitoring data includes:

[0092] The CVT metering and monitoring data are the amplitude and phase data of the CVT collected by the operation and maintenance sensors;

[0093] The CVT environmental factor monitoring data includes data collected by the operating parameter monitoring system, including climate and environmental data and operating condition data;

[0094] The climate and environmental data include the temperature and humidity of the CVT operating environment;

[0095] Operating condition data includes CVT surface contamination, CVT ambient electric field, and CVT secondary load.

[0096] It should be noted that a timestamp contains 12 sets of amplitude and phase data.

[0097] In this embodiment of the invention, the additional measurement error prediction based on the fully trained error ensemble model to determine the target additional measurement error of the real-time CVT environmental factor monitoring data includes:

[0098] The real-time CVT environmental factor monitoring data is input into the base learner layer, and the measurement additional error is predicted based on the base learner layer to obtain the first target meta dataset;

[0099] The weights of the first target metadata dataset are assigned based on the weight assignment model to obtain the second target metadata dataset.

[0100] The second target meta-dataset is input into the meta-learner layer to predict the target measurement additional error.

[0101] In a specific embodiment, the meta-learner layer uses the Extra-Trees algorithm model as the meta-learner to predict the target measurement additional error.

[0102] In this embodiment of the invention, the step of constructing a measurement error mechanism model and determining the ideal measurement error by performing measurement error analysis on the real-time CVT measurement monitoring data based on the measurement error mechanism model includes:

[0103] The amplitude and phase data of the CVT are input into the measurement error mechanism model to determine the ideal measurement error.

[0104] In a specific embodiment, the formula for the measurement error mechanism model is as follows:

[0105]

[0106] Wherein, U1 and U2 are the first and second voltage values ​​of the capacitive voltage transformer, respectively. and These are the first and second phase values ​​of the capacitive voltage transformer, respectively, k n This represents the voltage division ratio of a capacitive voltage transformer.

[0107] This invention also provides a CVT error measurement system based on a dual-drive model and data approach, such as... Figure 2 The CVT error measurement system 200 based on model and data dual-drive provided by the present invention includes:

[0108] 201. Data acquisition unit, used to acquire real-time CVT environmental factor monitoring data and real-time CVT metering monitoring data;

[0109] 202. Mechanism model construction unit, used to construct a measurement error mechanism model, and to perform measurement error analysis on the real-time CVT measurement monitoring data based on the measurement error mechanism model to determine the ideal measurement error;

[0110] 203. Error ensemble model acquisition unit, used to acquire a fully trained error ensemble model, and based on the fully trained error ensemble model, to predict additional measurement errors in the real-time CVT environmental factor monitoring data to determine the target additional measurement error;

[0111] 204. Actual measurement error prediction unit: determines the actual measurement error of CVT based on the target measurement additional error and the mechanism measurement error.

[0112] This invention provides a CVT error measurement method based on a dual-driven model and data approach. The data-driven approach employs the measurement error ensemble model proposed in this invention, which is an improved Stacking ensemble model. By adding weights to the ensemble model, the prediction results of the base learner layer are weighted to obtain a second meta-dataset, resulting in better data quality for the meta-learner layer and more accurate prediction results. At the same time, multiple algorithm models are used in the base learner layer to improve the generalization ability and robustness of the base learner through the ensemble model.

[0113] Furthermore, this invention uses a measurement error mechanism model to predict rational measurement errors, solving the problem that the model needs to consider multiple factors and is difficult to construct, and the results obtained by the model are inaccurate. By using a model-driven measurement error mechanism model and a data-driven measurement error integration model, the accuracy of CVT error measurement is improved through the dual drive of data and model, and the precise measurement of CVT measurement error is realized.

[0114] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0115] The foregoing has provided a detailed description of the CVT error measurement method and system based on model and data dual-drive provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. The above description is only a preferred embodiment of this invention, but the protection scope of this invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this invention should be included within the protection scope of this invention.

Claims

1. A CVT error measurement method based on model and data dual-drive, characterized in that, include: Acquire real-time CVT environmental factor monitoring data and real-time CVT metering monitoring data; A measurement error mechanism model is constructed, and the ideal measurement error is determined by performing measurement error analysis on the real-time CVT measurement monitoring data based on the measurement error mechanism model. Obtain a fully trained error ensemble model, and based on the fully trained error ensemble model, perform additional measurement error prediction on the real-time CVT environmental factor monitoring data to determine the target additional measurement error; The actual measurement error of CVT is determined based on the target measurement additional error and the mechanism measurement error. The step of obtaining a fully trained error ensemble model includes: Collect CVT environmental factor monitoring data, and construct an environmental factor data sample set based on the CVT environmental factor monitoring data; An error ensemble model is constructed, and the error ensemble model is trained based on the environmental factor data sample set to obtain a fully trained error ensemble model. The error ensemble model includes a base learner layer and a meta learner layer. The base learner layer includes at least three algorithm models as base learners of the base learner layer. The construction of the error ensemble model, and the training of the error ensemble model based on the environmental factor data sample set to obtain a fully trained error ensemble model, includes: Based on the environmental factor data sample set, the base learner of the base learner layer is trained by K-fold cross-validation to obtain a fully trained base learner layer. The environmental factor data sample set is input into the fully trained base learner layer to obtain the first meta-dataset; The second metadata is obtained by weighting the data in the first metadata dataset according to the time dimension and precision based on the weighting model. The algorithm model of the meta-learner layer is trained based on the second meta-dataset to obtain a fully trained meta-learner layer, that is, a fully trained error ensemble model. The base learner layer further includes a weight allocation model; the weight allocation model includes a time weight function and a precision weight function; the method further includes: The time weight of the prediction result of each base learner is determined according to the time weight function. The accuracy weight of each base learner is determined based on the accuracy weight function and the prediction accuracy of each base learner.

2. The CVT error measurement method based on model and data dual-drive as described in claim 1, characterized in that, The CVT metering and monitoring data are the amplitude and phase data of the CVT collected by the operation and maintenance sensors; The CVT environmental factor monitoring data includes data collected by the operating parameter monitoring system, including climate and environmental data and operating condition data; The climate and environmental data include the temperature and humidity of the CVT operating environment; Operating condition data includes CVT surface contamination, CVT ambient electric field, and CVT secondary load.

3. The CVT error measurement method based on model and data dual-drive as described in claim 1, characterized in that, Based on the fully trained error ensemble model, additional measurement error prediction is performed on the real-time CVT environmental factor monitoring data to determine the target additional measurement error, including: The real-time CVT environmental factor monitoring data is input into the base learner layer, and the measurement additional error is predicted based on the base learner layer to obtain the first target meta dataset; The weights of the first target metadata dataset are assigned based on the weight assignment model to obtain the second target metadata dataset. The second target meta-dataset is input into the meta-learner layer to predict the target measurement additional error.

4. The CVT error measurement method based on model and data dual-drive as described in claim 1, characterized in that, The construction of a measurement error mechanism model, and the determination of the ideal measurement error based on the measurement error mechanism model of the real-time CVT measurement monitoring data, includes: The amplitude and phase data of the CVT are input into the measurement error mechanism model, and the ideal measurement error is determined by the calculation formula of the measurement error mechanism model.

5. The CVT error measurement method based on model and data dual-drive as described in claim 1, characterized in that, The weights of each base learner are determined based on the accuracy weight function and the prediction accuracy of each base learner, including: The mean absolute percentage error of the prediction results of each base learner is calculated based on the prediction results of each base learner. The weights of each base learner's results are determined based on the mean absolute percentage error of each base learner and the accuracy weighting function.

6. A CVT error measurement system based on model and data dual-drive, characterized in that, include: The data acquisition unit is used to acquire real-time CVT environmental factor monitoring data and real-time CVT metering monitoring data. The mechanism model construction unit is used to construct a measurement error mechanism model and, based on the measurement error mechanism model, perform measurement error analysis on the real-time CVT measurement monitoring data to determine the ideal measurement error. The error ensemble model acquisition unit is used to acquire a fully trained error ensemble model, and based on the fully trained error ensemble model, to perform additional measurement error prediction on the real-time CVT environmental factor monitoring data to determine the target additional measurement error. The actual measurement error prediction unit determines the actual measurement error of the CVT based on the target measurement additional error and the mechanism measurement error. The error ensemble model acquisition unit is further configured to collect CVT environmental factor monitoring data, construct an environmental factor data sample set based on the CVT environmental factor monitoring data, construct an error ensemble model, and train the error ensemble model based on the environmental factor data sample set to obtain a fully trained error ensemble model. The error ensemble model includes a base learner layer and a meta learner layer. The base learner layer includes at least three algorithm models as base learners of the base learner layer. The error ensemble model acquisition unit is further configured to perform K-fold cross-validation training on the base learner of the base learner layer based on the environmental factor data sample set to obtain a fully trained base learner layer; input the environmental factor data sample set data into the fully trained base learner layer to obtain a first meta-dataset; assign weights to the data in the first meta-dataset based on the weight allocation model from the time dimension and precision to obtain a second meta-dataset; and train the algorithm model of the meta-learner layer based on the second meta-dataset to obtain a fully trained meta-learner layer, i.e., obtain a fully trained error ensemble model. The base learner layer further includes a weight allocation model, which includes a time weight function and a precision weight function. The error ensemble model acquisition unit is also used to determine the time weight of the prediction result of each base learner according to the time weight function; and to determine the precision weight of each base learner according to the precision weight function and the prediction precision of each base learner.

Images