Quantum-classical machine learning hybrid wind power prediction method

Abstract

The invention discloses a quantum-classical machine learning hybrid wind power prediction method, which comprises the following steps of: acquiring historical wind power, a fan sensor and meteorological data, and setting the wind power as a learning target; the method comprises the following steps: decomposing formatted time characteristics, generating 6-dimensional periodic coding time characteristics through sine-cosine function mapping, splicing the 6-dimensional periodic coding time characteristics with meteorological characteristics, selecting first n characteristic vectors according to importance scores, coding the first n characteristic vectors into a quantum state, and coding a mapping angle by using a parameterized revolving door; and constructing a parameterized quantum circuit to capture high-dimensional nonlinear characteristics, obtaining an output expected value through terminal measurement, calculating a mean square error as a loss function, adjusting a parameter to minimize the loss function, and predicting wind power by using the trained quantum neural network. The technical problems that an existing model is large in parameter quantity, complex in hyper-parameter space, slow in training convergence, high in model black box performance and the like are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power prediction, and in particular to a quantum-classical machine learning hybrid wind power prediction method and system. Background Art

[0002] With the development of renewable energy, wind power is increasingly playing a larger role in the global energy mix. However, wind power output is subject to significant volatility and uncertainty. Accurately forecasting wind power is crucial for achieving secure grid dispatch and optimizing energy management.

[0003] Traditional wind power forecasting methods are mostly based on classic machine learning algorithms, such as support vector machines (SVM), multi-layer perceptrons (MLP), and long short-term memory networks (LSTM). Although these methods have achieved certain results, they face the following challenges: The large number of parameters and complex hyperparameter space lead to a significant increase in the computational overhead of model optimization analysis; the training convergence speed is slow, usually requiring a large number of training rounds (epochs); the model calculation process has a highly black-box nature, and the interpretable analysis of model behavior faces severe challenges.

[0004] In recent years, quantum neural networks (QNNs), an emerging quantum machine learning model, have demonstrated unique advantages in high-dimensional nonlinear modeling, small sample generalization, and feature mapping. Introducing quantum computing into wind power forecasting is a promising solution to address these challenges. Summary of the Invention

[0005] The purpose of the present invention is to overcome the above-mentioned technical deficiencies and provide a quantum-classical machine learning hybrid wind power prediction method to solve the technical problems in the existing technology, such as the large number of parameters and complex hyperparameter space, which lead to a significant increase in the computational overhead of model optimization analysis; the slow training convergence speed, which usually requires a large number of training rounds (epochs); the model calculation process has a high degree of black box characteristics, and the interpretability analysis of the model behavior faces severe challenges.

[0006] To achieve the above technical objectives, in a first aspect, the technical solution of the present invention provides a quantum-classical machine learning hybrid wind power prediction method, comprising the following steps: Collect historical wind power data, wind turbine sensor data, and meteorological data. Wind turbine sensor data includes wind speed, wind direction, and formatted time, and set wind power as the learning target. Perform feature decomposition on the formatted time to extract the month time component, date time component, and hour time component, and map them through the sine-cosine periodic function to generate a 6-dimensional periodic encoding time feature; concatenating the 6-dimensional periodic encoding time features and the meteorological features extracted from the meteorological data into feature vectors, sorting them by feature importance scores, and selecting the top n feature vectors with the highest correlation with wind power; The selected eigenvector is encoded into a quantum state, and a parameterized rotation gate encoding method is used to map the eigenvector to the rotation angle of the quantum bit; Parameterized quantum circuits are constructed through multiple trainable rotation gates and entanglement gates to capture high-dimensional nonlinear characteristics. The expected output value is obtained by measuring the quantum circuit end, and the mean square error between the predicted value and the actual wind power value is calculated as the loss function. Adjust circuit parameters, minimize the loss function, and use the trained quantum neural network to predict wind power.

[0007] Compared with the prior art, the present invention has the following beneficial effects: 1. Reduced parameter and hyperparameter complexity: Quantum embedding maps high-dimensional data into a low-dimensional quantum state space, significantly reducing the number of model parameters. Furthermore, quantum architecture eliminates the need to design hyperparameters such as classical neuron layers and activation functions, fundamentally simplifying the hyperparameter optimization problem.

[0008] 2. Significantly improved training efficiency: Thanks to the characteristics of quantum parallel computing, the model can quickly approach the optimal solution in the early stages of training, and the number of epochs required for convergence is reduced by more than 50% compared to classical methods.

[0009] 3. The computational process is transparent and analyzable: Unitary evolution based on quantum gate operations is naturally mathematically reversible, and its computational process can be visualized and traced through quantum circuits, providing a physical-level white-box explanation of the model behavior.

[0010] According to some embodiments of the present invention, after collecting historical wind power data, wind turbine sensor data, and meteorological data, the following steps are included: Preprocess the data, fill in missing values, remove outliers, and normalize numerical features.

[0011] According to some embodiments of the present invention, sorting by feature importance score and selecting the top n feature vectors with the highest correlation with wind power includes the following steps: After training with the random forest regressor, the top eight eigenvectors with the highest correlation with wind power are selected by sorting by feature importance score.

[0012] According to some embodiments of the present invention, after mapping the eigenvector to the rotation angle of the quantum bit, the following steps are included: Each eigenvector value {\mathrm{x}}_{\mathrm{i}}\in \left [ {-1,1} \right ] , as quantum gate parameters, quantum gate parameters include: , ,in 、 are the revolving door parameters.

[0013] According to some embodiments of the present invention, adjusting circuit parameters to minimize a loss function includes the steps of: Adopting the classic optimizer Adam, the circuit parameters are continuously adjusted to minimize the loss function; In each round of iteration, the quantum gate parameters are updated according to the calculated gradient information, and the updated quantum gate parameters are reset to the quantum circuit.

[0014] In a second aspect, the present invention provides a quantum-classical machine learning hybrid wind power prediction system, which applies the quantum-classical machine learning hybrid wind power prediction method as described in any one of the first aspects.

[0015] According to some embodiments of the present invention, a quantum-classical machine learning hybrid wind power prediction system includes: A data acquisition device collects historical wind power data, wind turbine sensor data, and meteorological data. The wind turbine sensor data includes wind speed, wind direction, and formatting time, and sets wind power as a learning target. The feature encoding module decomposes the formatted time into features, extracts the month time component, date time component, and hour time component, and maps them through the sine-cosine periodic function to generate a 6-dimensional periodic encoding time feature; A feature selection module concatenates the 6-dimensional periodic encoding time features and the meteorological features extracted from the meteorological data into feature vectors, sorts them by feature importance score, and selects the top n feature vectors with the highest correlation with wind power; The quantum coding module encodes the selected eigenvector into a quantum state and uses a parameterized rotation gate encoding method to map the eigenvector to the rotation angle of the quantum bit; Quantum circuits, including multiple trainable rotation gates and entanglement gates, are used to capture high-dimensional nonlinear characteristics; The loss function calculation module obtains the expected output value through quantum circuit terminal measurement, and calculates the mean square error between the predicted value and the actual wind power value as the loss function; The wind power prediction module adjusts circuit parameters, minimizes the loss function, and uses the trained quantum neural network to predict wind power.

[0016] In a third aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to enable a computer to execute the quantum-classical machine learning hybrid wind power prediction method as described in any one of the first aspects. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of a quantum-classical machine learning hybrid wind power prediction method provided by one embodiment of the present invention; Figure 2 This is a quantum circuit diagram of a wind power prediction method using a hybrid quantum-classical machine learning approach provided by one embodiment of the present invention; Figure 3 This is a comparison chart of the predicted results and true values ​​of the wind power prediction method using a hybrid quantum-classical machine learning method provided by one embodiment of the present invention; Figure 4 This is a comparison chart of a quantum-classical machine learning hybrid wind power prediction method provided by an embodiment of the present invention and a classical model. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] Reference Figure 1-4 , Figure 1 This is a flow chart of a quantum-classical machine learning hybrid wind power prediction method provided by one embodiment of the present invention; Figure 2 This is a quantum circuit diagram of a wind power prediction method using a hybrid quantum-classical machine learning approach provided by one embodiment of the present invention; Figure 3 This is a comparison chart of the predicted results and true values ​​of the wind power prediction method using a hybrid quantum-classical machine learning method provided by one embodiment of the present invention; Figure 4 This is a comparison chart of a quantum-classical machine learning hybrid wind power prediction method provided by an embodiment of the present invention and a classical model.

[0020] In one embodiment, a hybrid quantum-classical machine learning wind power prediction method includes the following steps: collecting historical wind power data, wind turbine sensor data, and meteorological data, where the wind turbine sensor data includes wind speed, wind direction, and formatted time, and setting wind power as a learning target; performing feature decomposition on the formatted time to extract the month time component, the date time component, and the hour time component, and mapping each of these components through a sine-cosine periodic function to generate a 6-dimensional periodically encoded time feature; concatenating the 6-dimensional periodically encoded time feature with the meteorological features extracted from the meteorological data into a feature vector, sorting the feature vectors by feature importance score, and selecting the top n feature vectors with the highest correlation with wind power; encoding the selected feature vectors into quantum states, and mapping the feature vectors to the rotation angles of quantum bits using a parameterized rotating gate encoding method; constructing a parameterized quantum circuit using multiple trainable rotating gates and entanglement gates, where the quantum circuit is used to capture high-dimensional nonlinear features; obtaining an output expected value through measurement at the end of the quantum circuit, and calculating the mean square error between the predicted value and the actual wind power value as a loss function; adjusting circuit parameters to minimize the loss function, and using the trained quantum neural network to predict wind power.

[0021] The present invention collects historical wind power data, wind turbine sensor data (including wind speed, wind direction, and formatted time), and meteorological data. The integration of various data can more comprehensively reflect the actual situation of wind power being affected by various factors, so that subsequent predictions can be based on a richer information foundation, thereby improving the accuracy and reliability of the predictions.

[0022] This method performs feature decomposition on formatted time, extracting the month, date, and hour components. It then generates a six-dimensional, periodically encoded time feature through sine-cosine periodic function mapping. This sophisticated temporal feature processing helps capture the cyclical variations in wind power across different time scales, such as seasonal and daily variations. This allows for better integration of time factors into forecasting models, enhancing understanding of wind power fluctuations over time.

[0023] This method combines temporal and meteorological features into feature vectors, sorts them by feature importance, and selects the top n feature vectors with the highest correlation with wind power. This feature screening mechanism removes redundant features that contribute little to the prediction results or may even interfere with them, focusing on key features. This reduces data dimensionality while improving model training and prediction efficiency, ultimately boosting prediction accuracy.

[0024] This method uses parameterized rotating gate encoding to encode feature vectors into quantum states, and constructs parameterized quantum circuits using multiple trainable rotating gates and entanglement gates. Quantum circuits can exploit the unique quantum properties of qubits, such as superposition and entanglement, to capture nonlinear features in high-dimensional space. Traditional machine learning methods often face limitations such as high computational complexity and poor performance when dealing with high-dimensional nonlinear problems. The use of quantum circuits enables models to better mine complex relationships in data, thereby improving the quality of wind power forecasts.

[0025] This method uses quantum circuit terminal measurements to determine the expected output value. The mean squared error between the predicted value and the actual wind power value is used as a loss function, and the circuit parameters are then adjusted to minimize this loss function. This feedback optimization approach based on quantum measurement results enables the entire model to continuously adjust itself, converging towards more accurate wind power predictions, ensuring the effectiveness and adaptability of the prediction method.

[0026] The present invention ultimately uses a trained quantum neural network to predict wind power. The quantum neural network combines the advantages of quantum computing and the structural characteristics of neural networks. Compared with simple classical machine learning models, it has stronger representation and computing capabilities, and is expected to show better performance in processing complex wind power prediction problems. For example, it can more accurately predict wind power output in the face of complex meteorological conditions and changeable wind turbine operating conditions.

[0027] The present invention may specifically include the following steps: Step 1: Data collection and preprocessing; Collect historical wind power data, wind turbine sensor data, and meteorological data, including wind speed, wind direction, formatting time, and other information, and set wind power as the learning target (label).

[0028] Preprocess the data, fill in missing values, remove outliers, and normalize numerical features.

[0029] Step 2: Periodic encoding of time information; In order to effectively express the periodicity of time characteristics, the formatted time Perform feature decomposition to extract the three time components of month, day, and hour, and map them through the sine-cosine periodic function (where is a period parameter, such as a daily period = 24 ): ; Finally, a 6-dimensional periodic encoding time feature is generated. This method can enhance the continuity and periodicity of time, thereby improving the model's learning efficiency for the periodic variation characteristics of wind power data.

[0030] Step 3: Feature selection; The time features obtained by period encoding and other meteorological features are concatenated into a feature vector. After training with a random forest regressor, the feature vectors are sorted by their importance scores, and the top 10 features with the highest correlation with wind power are selected for subsequent modeling.

[0031] Step 4: Quantum state encoding; The selected features Encoded as a quantum state, the parameterized rotation gate encoding method is used to map the eigenvalue to the rotation angle of the quantum bit, for example: For each eigenvalue {\mathrm{x}}_{\mathrm{i}}\in \left [ {-1,1} \right ] , which is used as the parameter of the quantum gate, such as , wait.

[0032] Thus, the conversion from classical vector to quantum bit state is realized for one feature, and then more features are adapted by increasing the number of quantum bits.

[0033] Step 5: Construct a parameterized quantum circuit (PQC); Pass multiple trainable turnstiles , and entanglement gates (such as CNOT) to construct parameterized quantum circuits to capture high-dimensional nonlinear characteristics.

[0034] Step 6: Measurement output and objective function construction; The output prediction value is obtained by measuring at the end of the quantum circuit, such as the measurement result in the Z direction (qubit 0): ; The loss function is calculated by calculating the mean square error (MSE) between the predicted value and the actual wind power value: ; Step 7: Optimize the training process; The classical optimizer Adam is used to continuously adjust the circuit parameters to minimize the loss function. In each iteration, the quantum gate parameters are updated based on the calculated gradient information and reset to the PQC.

[0035] Step 8: Model validation and prediction; The trained quantum neural network is used to predict wind power and the accuracy is compared with the actual value.

[0036] Through the refinement of data preprocessing, the targeting of feature selection, the strong characterization capability of quantum computing, and quantum-classical collaborative optimization, this invention is expected to achieve higher prediction accuracy and generalization capability than traditional machine learning in processing tasks such as wind power forecasting, which have high dimensions, strong nonlinearity, and significant periodic characteristics, and provide more reliable decision support for wind power grid scheduling and energy management.

[0037] Improved results: While achieving performance similar to that of classical neural networks, quantum neural networks exhibit multi-dimensional advantages: Parameter efficiency is significantly improved: e.g. Figure 2 As shown, a quantum neural network requires only 16 adjustable parameters (implemented by quantum rotating gates), while a classical neural network (with two hidden layers, 50 neurons in hidden layer 1 and 25 neurons in hidden layer 2) requires 1,751 parameters. This reduces the number of parameters in the quantum model to 0.91% of the classical model (a compression ratio of 110:1), achieving a parameter reduction of two orders of magnitude.

[0038] Training efficiency is significantly improved: Figure 4 As shown in the upper left figure (Comparison of training process (MSE curve)), the quantum neural network achieves exponential convergence in the early stages of training (< 5 epochs), and its convergence speed is 3 times faster than that of the classical neural network.

[0039] Generalization and robustness are significantly improved: Figure 4 The lower left figure (prediction error distribution (residual plot)) shows the distribution of the difference (residual) between the predicted value and the true value in the test sample, which is an important indicator to measure the generalization ability of the model.

[0040] As shown, the errors of the quantum neural network on the test data are more concentrated, indicating that it has strong generalization ability and robustness, low sensitivity to abnormal samples, and is more suitable for scenarios with strong uncertainty in actual wind power forecasting tasks.

[0041] The model performance is similar to that of a classic neural network: Figure 4 The upper right corner figure (comparison of the actual value and predicted value of the test set): The quantum neural network also performs very well in the medium and low value ranges, and there may be unstable predictions in the high value range.

[0042] Figure 4 Bottom right graph (RMSE (Root Mean Square Error) comparison): The quantum neural network and the classical neural network have similar data on the test set (RMSE 20 vs 18), and their overall performance is similar.

[0043] In one embodiment, a quantum-classical machine learning hybrid wind power prediction method includes the following steps: collecting historical wind power data, wind turbine sensor data, and meteorological data, where the wind turbine sensor data includes wind speed, wind direction, and formatted time, and setting wind power as a learning target; preprocessing the data to fill missing values, remove outliers, and normalize numerical features; performing feature decomposition on the formatted time to extract the month time component, date time component, and hour time component, and mapping them through a sine-cosine periodic function to generate a 6-dimensional periodic encoding time feature; and The dimensional periodic encoded time features and the meteorological features extracted from the meteorological data are spliced ​​into feature vectors, which are sorted by feature importance scores, and the top n eigenvectors with the highest correlation with wind power are selected; the selected eigenvectors are encoded as quantum states, and the eigenvectors are mapped to the rotation angles of quantum bits using parameterized revolving gate encoding; a parameterized quantum circuit is constructed using multiple trainable revolving gates and entanglement gates, and the quantum circuit is used to capture high-dimensional nonlinear features; the output expected value is obtained through measurement at the end of the quantum circuit, and the mean square error between the predicted value and the actual wind power value is calculated as the loss function; the circuit parameters are adjusted to minimize the loss function, and the wind power is predicted using the trained quantum neural network.

[0044] Missing Value Compensation: Wind power data collection may result in missing data due to sensor failures, transmission interruptions, and other factors. Using data with missing values ​​directly for modeling results in incomplete information and can lead to biased model training. Compensating for missing values ​​(e.g., using the mean, median, or interpolation) ensures data continuity and integrity, enabling the model to more comprehensively learn data patterns.

[0045] Outlier removal: Extreme weather and transient equipment failures can generate anomalous data (e.g., sudden increases in wind speed to physically unreasonable levels). If left unaddressed, these outliers can mislead the model into learning incorrect patterns and reduce prediction stability. Removing outliers purifies the data distribution, ensuring that the model is trained on realistic and reasonable samples.

[0046] Normalizing numerical features: Wind power-related features have widely varying dimensions and value ranges (e.g., wind speed in m / s, power in kW). Unnormalized features can cause the model to be overly sensitive to large numerical features (e.g., power values ​​can be much larger than wind speed values), affecting parameter optimization efficiency. Normalization (e.g., mapping to [0, 1] or a standard normal distribution) scales all features to a uniform scale, ensuring a more balanced impact on the model, accelerating the convergence of quantum circuit parameters, and avoiding training fluctuations caused by varying feature scales.

[0047] The preprocessing method of this invention reduces noise and erroneous signals in the raw data by standardizing the data format and correcting for data deviations. For quantum-classical hybrid models, quantum state encoding requires higher stability of the input data. If the input features contain anomalies or are missing, this will directly affect the angular mapping accuracy of the parameterized revolving gate and distort the feature representation of the quantum circuit. Preprocessing can provide cleaner input for quantum state encoding, ensuring that the quantum circuit accurately captures the correlation between features, ultimately improving the accuracy and robustness of wind power forecasting.

[0048] This method may have different data standards and formats. Preprocessing unifies the data format through standardization (such as normalization), allowing different types of features such as time features and meteorological features to be effectively spliced ​​into feature vectors. This lays the foundation for subsequent feature selection and quantum state encoding, ensuring that the synergy of multi-source data is fully utilized.

[0049] Furthermore, the feature importance scores are sorted and the top n feature vectors with the highest correlation with wind power are selected, including the steps of: after training with a random forest regressor, sorting by the feature importance scores and selecting the top 8 feature vectors with the highest correlation with wind power.

[0050] The random forest regressor objectively quantifies the strength of each feature's correlation with wind power by calculating its contribution during the decision tree split (e.g., the Gini index or mean squared error reduction). This data-driven scoring approach avoids the subjectivity of manual feature screening and ensures that the top eight features selected are truly core factors that significantly impact wind power forecasting (e.g., key meteorological characteristics, cyclical time characteristics, etc.). This allows the model to focus on valuable information and reduce interference from irrelevant features.

[0051] Wind power forecasting involves high-dimensional raw features, such as six-dimensional time-encoded features plus various meteorological features. This high-dimensional data increases the complexity of quantum state encoding, requiring more qubits or more complex revolving gate mappings, and can lead to model overfitting. Selecting the top eight highly important features effectively reduces data dimensionality, alleviating the computational burden of quantum circuits and accelerating model training. Furthermore, this streamlined feature set reduces the interference of noise features on the model, improving forecast stability.

[0052] Furthermore, the circuit parameters are adjusted to minimize the loss function, including the following steps: using the classical optimizer Adam to continuously adjust the circuit parameters to minimize the loss function; in each round of iteration, the quantum gate parameters are updated according to the calculated gradient information, and the updated quantum gate parameters are reset to the quantum circuit.

[0053] The parameters of quantum circuits (such as the angle of a revolving door) are essentially classical values, but parameter optimization requires the relationship between quantum measurement results (output expected values) and a loss function. The Adam optimizer, a mature optimizer in the classical domain, features adaptive learning rate adjustment and momentum-accelerated convergence. It effectively handles the closed-loop process of "quantum computation output → classical loss calculation → classical parameter update → quantum circuit reconstruction" in hybrid quantum-classical scenarios. This avoids the complexity of directly designing a quantum optimizer while leveraging the stability of classical optimization to ensure the training process.

[0054] The Adam optimizer calculates the gradient of the loss function with respect to quantum gate parameters, inferring them backwards from the measurement results of the quantum circuit to achieve targeted parameter updates. By adjusting the rotation gate and entanglement gate parameters based on the gradients in each iteration, the loss function (mean squared error) converges along the direction of fastest descent, significantly reducing the number of iterations compared to undirected random search or simple gradient descent. This is particularly important for quantum computing—quantum circuit measurements and parameter updates inherently incur computational overhead. The Adam optimizer's efficiency reduces overall training costs.

[0055] In one embodiment, a quantum-classical machine learning hybrid wind power prediction system includes: a data acquisition device for collecting historical wind power data, wind turbine sensor data, and meteorological data, wherein the wind turbine sensor data includes wind speed, wind direction, and formatted time, and wind power is set as a learning target; a feature encoding module for performing feature decomposition on the formatted time, extracting the month time component, the date time component, and the hour time component, and generating a 6-dimensional periodic encoding time feature through sine-cosine periodic function mapping respectively; a feature selection module for splicing the 6-dimensional periodic encoding time feature with the meteorological feature extracted from the meteorological data into a feature vector, and evaluating the feature importance. The system sorts the eigenvectors and selects the top n eigenvectors with the highest correlation with wind power. The quantum coding module encodes the selected eigenvectors into quantum states and uses parameterized rotating gate encoding to map the eigenvectors to the rotation angles of quantum bits. The quantum circuit includes multiple trainable rotating gates and entanglement gates, and the quantum circuit is used to capture high-dimensional nonlinear features. The loss function calculation module obtains the expected output value through the end measurement of the quantum circuit, and calculates the mean square error between the predicted value and the actual wind power value as the loss function. The wind power prediction module adjusts the circuit parameters, minimizes the loss function, and uses the trained quantum neural network to predict wind power.

[0056] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may include a memory remotely located relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0057] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate and may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0058] In addition, an embodiment of the present invention also provides a computer-readable storage medium, which stores computer-executable instructions, and the computer-executable instructions are executed by a processor or controller, for example, by a processor in the above-mentioned terminal embodiment, so that the above-mentioned processor can execute the quantum-classical machine learning hybrid wind power prediction method in the above-mentioned embodiment.

[0059] Those skilled in the art will appreciate that all or some of the steps and systems described above can be implemented as software, firmware, hardware, or any combination thereof. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on computer-readable media, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is well known to those skilled in the art, the term computer storage media encompasses volatile and non-volatile, removable, and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVDs) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. Furthermore, as is well known to those skilled in the art, communication media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0060] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above implementation. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present invention.

[0061] The above specific embodiments of the present invention do not constitute a limitation on the scope of protection of the present invention. Any other corresponding changes and modifications made based on the technical concept of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A wind power prediction method based on quantum-classical machine learning hybrid, characterized in that: Including steps: Collect historical wind power data, wind turbine sensor data, and meteorological data. Wind turbine sensor data includes wind speed, wind direction, and formatted time, and set wind power as the learning target. Perform feature decomposition on the formatted time to extract the month time component, date time component, and hour time component, and map them through the sine-cosine periodic function to generate a 6-dimensional periodic encoding time feature; concatenating the 6-dimensional periodic encoding time features and the meteorological features extracted from the meteorological data into feature vectors, sorting them by feature importance scores, and selecting the top n feature vectors with the highest correlation with wind power; The selected eigenvector is encoded into a quantum state, and a parameterized rotation gate encoding method is used to map the eigenvector to the rotation angle of the quantum bit; Parameterized quantum circuits are constructed through multiple trainable rotation gates and entanglement gates to capture high-dimensional nonlinear characteristics. The expected output value is obtained by measuring the quantum circuit end, and the mean square error between the predicted value and the actual wind power value is calculated as the loss function. Adjust circuit parameters, minimize the loss function, and use the trained quantum neural network to predict wind power.

2. The wind power prediction method based on quantum-classical machine learning hybrid according to claim 1 is characterized in that: After collecting historical wind power data, wind turbine sensor data, and meteorological data, the following steps are included: Preprocess the data, fill in missing values, remove outliers, and normalize numerical features.

3. The wind power prediction method based on quantum-classical machine learning hybrid according to claim 1 is characterized in that: Sort by feature importance score and select the top n feature vectors with the highest correlation with wind power, including the following steps: After training with the random forest regressor, the top eight eigenvectors with the highest correlation with wind power are selected by sorting them according to their importance scores.

4. The wind power prediction method based on quantum-classical machine learning hybrid according to claim 1 is characterized in that: After mapping the eigenvector to the rotation angle of the quantum bit, the following steps are included: Each eigenvector value {\mathrm{x}}_{\mathrm{i}}\in \left [ {-1,1} \right ] , as quantum gate parameters, quantum gate parameters include: , ,in 、 are the revolving door parameters.

5. The wind power prediction method based on quantum-classical machine learning hybrid according to claim 1, characterized in that: Adjust the circuit parameters to minimize the loss function, including the following steps: Adopting the classic optimizer Adam, the circuit parameters are continuously adjusted to minimize the loss function; In each round of iteration, the quantum gate parameters are updated according to the calculated gradient information, and the updated quantum gate parameters are reset to the quantum circuit.

6. A quantum-classical machine learning hybrid wind power prediction system, characterized by: The wind power prediction method using a hybrid quantum-classical machine learning method as described in any one of claims 1 to 5 is applied.

7. The quantum-classical machine learning hybrid wind power prediction system according to claim 6 is characterized in that: include: A data acquisition device collects historical wind power data, wind turbine sensor data, and meteorological data. The wind turbine sensor data includes wind speed, wind direction, and formatting time, and sets wind power as a learning target. The feature encoding module decomposes the formatted time into features, extracts the month time component, date time component, and hour time component, and maps them through the sine-cosine periodic function to generate a 6-dimensional periodic encoding time feature; A feature selection module concatenates the 6-dimensional periodic encoding time features and the meteorological features extracted from the meteorological data into feature vectors, sorts them by feature importance score, and selects the top n feature vectors with the highest correlation with wind power; The quantum coding module encodes the selected eigenvector into a quantum state and uses a parameterized rotation gate encoding method to map the eigenvector to the rotation angle of the quantum bit; Quantum circuits, including multiple trainable rotation gates and entanglement gates, are used to capture high-dimensional nonlinear characteristics; The loss function calculation module obtains the expected output value through quantum circuit terminal measurement, and calculates the mean square error between the predicted value and the actual wind power value as the loss function; The wind power prediction module adjusts circuit parameters, minimizes the loss function, and uses the trained quantum neural network to predict wind power.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the quantum-classical machine learning hybrid wind power prediction method according to any one of claims 1 to 5.

Images