A wind power prediction method and system based on quantum computing

Abstract

This application discloses a wind power prediction method and system based on quantum computing. The prediction method includes acquiring historical and current wind power datasets, normalizing the data to construct a supervised learning training sample set, training the system using the training sample set, acquiring wind power-related data for the time point to be predicted, preprocessing the data again to obtain the feature vector of the sample to be predicted, inputting it into the trained quantum support vector machine model, obtaining the wind power prediction value at the normalized scale through quantum inner product operations, and performing inverse normalization on the normalized wind power prediction value to obtain the wind power prediction result in physical units and outputting it. This application adopts a classical-quantum hybrid computing architecture, using quantum computing to realize the core computational part of wind power prediction, achieving an exponential acceleration of wind power prediction.

Description

Technical Field

[0001] This application belongs to the field of power load forecasting, and in particular relates to a wind power forecasting method and forecasting system. Background Technology

[0002] The safe, stable, and economical operation of a power system depends on the real-time dynamic balance between power generation and consumption. Since electrical energy is difficult to store on a large scale, achieving this balance requires accurate forecasting of future loads, which in turn guides the formulation and adjustment of power generation plans. Short-term load forecasting (typically referring to forecasts for the next few hours to days) is a crucial foundation for optimizing power generation capacity scheduling, reducing operating costs, and ensuring power supply reliability. Wind power, a distributed renewable energy generation mode, exhibits significant intermittency, randomness, and volatility. Its large-scale integration not only alters the traditional unidirectional power flow pattern of the power grid but also injects substantial uncertainty into the system, dramatically increasing the challenge of maintaining supply-demand balance. High-precision short-term forecasting of renewable energy generation capacity, combined with traditional load forecasting, forms an accurate assessment of the system's net load, which is of paramount importance for improving the grid's absorption capacity, ensuring stable system operation, and optimizing energy storage configuration.

[0003] To address these forecasting needs, academia and industry have developed various forecasting methods. Traditional methods, such as time series analysis and regression analysis, have limited capabilities when handling nonlinear and non-stationary load and renewable energy data. Machine learning methods, represented by artificial neural networks, have good nonlinear fitting capabilities, but suffer from drawbacks such as susceptibility to local optima, reliance on experience in model structure design, and long training times. Support Vector Machines (SVMs), based on statistical learning theory and minimizing structural risk, have demonstrated advantages in solving small-sample, nonlinear, and high-dimensional pattern recognition problems, and have been successfully applied to load and power forecasting. However, the high computational complexity of solving quadratic programming problems when training standard SVMs on large-scale datasets limits their application in real-world massive data scenarios. Summary of the Invention

[0004] To address the aforementioned problems, this application discloses a wind power prediction method based on quantum computing, comprising the following steps: S1: Obtain historical wind power dataset and current wind power dataset. The historical wind power dataset includes historical actual wind power data and historical environmental factor data for a consecutive (M-1) time points arranged in time series. The current wind power dataset includes real-time wind power data and real-time environmental factor data for the current time point. S2: Normalize each data point in the historical wind power dataset and the current wind power dataset, and use these normalized values ​​as dimensional feature values ​​to construct the input feature vector. The normalized actual power data at the next time point is used as the target variable. Construct a supervised learning training sample set { ( , )}, where i=1, 2, ... M, and M is the total number of training samples.

[0005] S3: Training is performed using a training sample set. The training process specifically includes the following quantum computing sub-steps: S31: Encode the feature vector set of the training sample set into a first quantum state. Encode the target variable set as a second quantum state | >; S32: Select the preset kernel function type and kernel parameters, calculate the kernel function value between any two training sample feature vectors, and form the kernel matrix K; utilize quantum states | >, prepare the quantum superposition state |χ> of all training data feature vectors, and obtain the ratio information of the kernel matrix K and its trace tr(K) by calculating the reduced density matrix of a specified subsystem in the superposition state; S33: Set the regularization parameter γ, and construct a system of linear equations for solving the least squares support vector machine (LSSVM): F·[b;α]=[0;y], where F is the quantum support vector machine matrix, b is the bias term, α is the Lagrange multiplier vector, and y is the vector composed of the target values ​​of all training samples; perform Hamiltonian decomposition and simulation on the quantum support vector machine matrix F, and construct a quantum circuit to realize the unitary evolution operator exp(-iFτ); use quantum state | As input, a quantum phase estimation algorithm is executed to obtain the phase-encoded state; S34: Add an auxiliary qubit, perform a controlled rotation operation on the auxiliary qubit to encode the reciprocal information of the eigenvalues ​​into the quantum state amplitude; then perform measurement and post-selection operations to obtain the parametric quantum state |b,α> representing the LSSVM model, and use the parametric quantum state |b,α> to control the training data quantum state |b,α>. The preparation of > generates a support vector quantum state |ũ> that encodes the parameter α into the amplitude; S4: Obtain wind power-related data for the time point to be predicted. The wind power-related data for the time point to be predicted includes the predicted wind speed, predicted wind direction, predicted temperature, and historical wind power datasets for multiple consecutive time points. Perform the same preprocessing as in step S2 to obtain the feature vector X0 of the sample to be predicted. S5: Encode the feature vector of the sample to be predicted as a quantum state |ṽ0>, input it into the quantum support vector machine model trained in step S3, and calculate the inner product of the support vector quantum state |ũ> and the test sample quantum state |ṽ0> through quantum inner product operation. The inner product value is the wind power prediction value y(X0) under the normalized scale. S6: Output Results: Perform inverse normalization on the normalized wind power prediction values ​​to obtain the wind power prediction results in physical units and output them.

[0006] Preferably, a data cleaning step is included between step 1 and step 2: performing at least one of the following processing on the historical wind power dataset and the current wind power dataset: outlier detection and removal, missing value imputation, and noise filtering.

[0007] Furthermore, in step 2, the normalization processing of the historical wind power dataset and the current wind power dataset adopts the min-max normalization method to linearly transform the data to the [0, 1] interval; the normalization processing of the actual power data adopts the logarithmic normalization method.

[0008] Furthermore, steps S3 and S5 are executed on the quantum processor, while the remaining steps are executed on the classical processor, which is integrated with the quantum processor in the same quantum computer.

[0009] Furthermore, the periodic encoding process in step S2 specifically involves converting the wind direction WD into a two-dimensional vector in the form of an angle. To maintain the periodic continuity between 0° and 360°, Let t be the wind direction at time t.

[0010] Furthermore, in step S3, the kernel function type is a Gaussian radial basis function, and its expression is: Where σ is the kernel width parameter, Let be the feature vectors of any two training samples.

[0011] Furthermore, in step S3, the feature vector set of the training sample set is encoded into a first quantum state using a quantum random access memory and an Oracle operator module. The specific expression is: , where x j Let y represent the feature vector of the j-th sample in the training dataset, and let y be the corresponding target variable. j ,|x j | represents the eigenvector x j The length of the mold, For the j-th eigenvector, the k-th value is... It is a set of orthogonal bases in the quantum system space.

[0012] Furthermore, in step S3, the expression for encoding vector y into the quantum state |y> is: ,in Let y be the Euclidean norm of vector y.

[0013] Further, in step S3, the quantum phase estimation algorithm includes: applying a Hadamard gate to the first register to generate a superposition state; using the first register qubit as the control bit, sequentially executing controlled exp(-iF2)... j The operation is performed with τ as a preset time parameter; the inverse quantum Fourier transform is performed on the first register and measured, and the result is stored in the second register to obtain the phase-encoded state.

[0014] This application also discloses a wind power prediction system based on quantum computing, used to execute the wind power prediction method based on quantum computing according to any one of claims 1-9, characterized in that it includes a historical data acquisition module, a data preprocessing module, a quantum computing module, and a data postprocessing module; the historical data acquisition module is used to execute step S1; the data preprocessing module is used to execute step S2; the quantum computing module includes a quantum processor, a quantum storage unit, and a quantum measurement and control platform, used to run steps S3, S4, and S5; and the data postprocessing module is used to run step S6.

[0015] In summary, compared with the prior art, the above-described technical solutions conceived in this application can achieve the following beneficial effects: This application employs a classical-quantum hybrid computing architecture, delegating the core computational part of wind power prediction—solving large-scale linear equations (HHL algorithm)—to a quantum computing module. By utilizing quantum phase estimation and the HHL algorithm, the training complexity of LSSVM is reduced from... Down to The prediction complexity is from Down to Quantum inner product calculation This results in exponential speedup. The amplitude encoding method of quantum states in the quantum computing part enables... One qubit can represent This method effectively handles high-dimensional features and massive samples. Based on quantum acceleration, it can incorporate more historical data and meteorological features to construct more complex models, thereby improving prediction accuracy. Furthermore, its rapid computing power enables faster model updates, meeting the real-time requirements of power grid dispatch for ultra-short-term wind power forecasting. In the overall methodology and system, the classical component handles data preprocessing and post-processing, while the quantum component handles model training and prediction computation. This method helps overcome the computational bottleneck of classical LSSVM under large data volumes, providing a feasible solution for real-time ultra-short-term wind power forecasting. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings used in the description of the embodiment or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a quantum computing-based wind power prediction method provided for embodiments of this application; Figure 2 This is a schematic diagram of the circuit for preparing eigenvector quantum states provided in an embodiment of this application; Figure 3 A schematic diagram of the quantum circuit for calculating the inner product of eigenvector quantum states provided in this application embodiment; Figure 4 A schematic diagram of the module composition of the wind power prediction system provided in this application; Figure 5 This is a schematic diagram illustrating the collaborative operation of the wind power prediction system provided in this application embodiment with external sites and equipment. Detailed Implementation

[0018] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the embodiments of this application will be further described in detail below with reference to the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0020] To address the uncertainty of grid-connected power caused by the intermittency, randomness, and volatility of new energy power generation systems, and to improve the high computational complexity, high computing power, time, and energy consumption requirements of existing power generation prediction methods when training large-scale datasets, this application proposes a power generation prediction method and system based on quantum computing. The technical solution of this invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] Figure 1 The flowchart of a wind power prediction method based on quantum computing provided in this application includes the following steps.

[0022] S1: Obtain historical wind power datasets and current wind power datasets.

[0023] The historical wind power dataset includes historical actual wind power data and historical environmental factor data for a consecutive (M-1) time point arranged in a time series; the current wind power dataset includes real-time wind power data and real-time environmental factor data for the current time point (which can be regarded as the Mth time point).

[0024] Historical and real-time environmental data can include environmental factors that have a significant and direct impact on wind power generation, such as wind speed, wind direction, temperature, humidity, and precipitation. It can also include periodic factors such as season and date. Wind resources exhibit periodic changes throughout the four quarters or twelve months of a year, and even on different days within a month. Preliminary tests can be conducted by increasing the weight of historical wind power data from previous months, quarters, and years, based on the current date, month, and season.

[0025] For data acquisition, power output data of wind farms or wind turbines are collected over historical time periods (e.g., every 10 or 15 minutes), along with relevant meteorological data such as wind speed, wind direction, and temperature. Measured data from the wind farm is preferred for this meteorological data. If measured data is unavailable, historical weather forecast data is used.

[0026] After data acquisition, the following steps may also be included: performing at least one of the following processing on the historical wind power dataset and the current wind power dataset: outlier detection and removal, missing value imputation, and noise filtering. Outlier detection and removal includes removing obviously erroneous or abnormal records from the collected data, such as negative power values ​​or exceeding the rated capacity, and then aligning the data to ensure that the power data and the timestamps of the meteorological data correspond perfectly.

[0027] S2: Normalize each data point in the historical wind power dataset and the current wind power dataset, and use these normalized values ​​as dimensional feature values ​​to construct the input feature vector. The actual power data corresponding to the next time point (referring to the time point after the i-th time point, i.e., the (i+1)-th time point) is normalized and then used as the target variable. Construct a supervised learning training sample set { ( , )}, where i=1, 2, ...,M, and M is the total number of training samples.

[0028] In terms of data planning, the wind power (in kW or MW) and wind speed (for each sample in the dataset at each time point) ),wind direction( A feature vector is composed of N data points, including temperature (Tt), humidity (Dt), precipitation (Ht), and other environmental factors. For example, using data from the first m time points to predict the next time point: Here It is the input feature vector. It is the output value. This represents the actual power at the first m time points. For wind speed at time t, Let t be the wind direction at time t.

[0029] The wind direction data needs to undergo periodic encoding, specifically by converting the wind direction (WD) into a two-dimensional vector in the form of an angle. To maintain the periodic continuity between 0° and 360°.

[0030] The normalization of feature vectors can be achieved using the min-max normalization method, i.e.: Indicates the first The first sample dimensional features, and These are the minimum and maximum values ​​of the feature over the entire training set, respectively. This is the normalized value.

[0031] The normalization of the target vector can be achieved using the min-max normalization method, i.e.: The target vector can also be normalized using logarithmic normalization: Similarly, and The minimum and maximum values ​​of this feature over the entire training set. These are the normalized values. To represent gradients, the normalized values ​​are uniformly expressed as... , .

[0032] The amplitude encoding method of quantum states in the quantum computing part enables One qubit can represent Dimensional data effectively processes high-dimensional features and massive samples. Based on quantum acceleration, it can incorporate more historical data and meteorological features to build more complex models, thereby improving prediction accuracy.

[0033] S3: Input the training sample set into the quantum support vector machine model for training. The training process specifically includes the following quantum computing sub-steps: S31: Encode the feature vector set of the training sample set into a first quantum state. >(each wind power feature vector) Encoding the target variable set as a second quantum state (encoding it as a high-dimensional quantum state) >, quantum state| > are all target variables A superposition state. This operation only requires... A qubit can represent an N-dimensional eigenvector and simultaneously represent M target variables. Compress to Quantum bits.

[0034] The training sample set in step S2 { ( , The expanded representation is as follows: {(x1, y1),(x2, y2)……(x)} j , y j ) …… (x M , y m )},x j Representing an N-dimensional vector, the classical data of the training set is represented using quantum states using quantum random access memory and an oracle operator, such as... Figure 2 As shown.

[0035] In quantum circuits, the Oracle is a module with known functionality that works in conjunction with quantum random access memory (QRAM) to... The quantum state preparation of the data is completed within the time complexity, and the high-dimensional eigenvectors are efficiently loaded into the quantum system for subsequent operations such as quantum phase estimation and inner product calculation.

[0036] The feature vector x is obtained through the Oracle module. j The corresponding quantum state: Where x j Let |x| represent the feature vector of the j-th sample in the training dataset. j | represents the eigenvector x j The length of the mold, For the j-th eigenvector, the k-th value is... It is a set of orthogonal bases in the quantum system space.

[0037] Encode the vector y into a quantum state | The specific expression for > is: ,in Let be the Euclidean norm of vector y. Where M is the total number of training samples. For the first The normalized target wind power value (real number) of each training sample. The ground state for calculating the sample index (usually using) Encoding by 1 qubit. Preparation | The essence of > is amplitude encoding, which is to encode an amplitude encoding signal. Classical vector The normalized component is directly mapped to the quantum state amplitude, and can also be prepared using quantum random access memory and the Oracle operator.

[0038] S32: Select the preset kernel function type and kernel parameters, calculate the kernel function value between any two training sample feature vectors, and form the kernel matrix K; utilize quantum states | > Prepare the quantum superposition state |χ> of all training data feature vectors, and implicitly obtain the ratio information of the kernel matrix K and its trace tr(K) by calculating the reduced density matrix of a specified subsystem in the superposition state.

[0039] Quantum superposition state |χ> specifically is , This is the normalization coefficient.

[0040] Calculate the kernel matrix The ratio of its trace Its purpose is to provide a basis for subsequent simulation of the Hamiltonian. Prepare key parameters. This is achieved implicitly and rapidly by preparing the quantum superposition states |χ> of all training data and calculating the partial trace (reduced density matrix) over the second system. Instead of explicitly calculating the entire matrix.

[0041] The role of kernel functions is to map feature vectors into a high-dimensional space, replacing complex nonlinear mappings, making feature vectors linearly separable in high-dimensional space, and solving nonlinear problems and the curse of dimensionality. Each kernel function has its advantages and limitations, and the choice of kernel function has a significant impact on the speed and performance of the support vector machine algorithm.

[0042] In this application, the kernel function is preferably a Gaussian radial basis function (RBF). Because the Gaussian RBF is radially symmetric, has strong generalization ability, and only has one parameter σ, it is chosen as the kernel function for the prediction model. Its expression is: Where σ is the kernel parameter (also known as the kernel width parameter), Let be the feature vectors of any two training samples.

[0043] S33: Set the regularization parameter γ, and construct the linear equation system for solving the least squares support vector machine (LSSVM): F·[b;α]=[0;y], where F is The matrix of a quantum support vector machine is given by , where b is the bias term and α is the Lagrange multiplier vector (α=[α1 ,α2 ,…,αM )). T ), where y is a vector consisting of the normalized target values ​​of all training samples (y=[y1,y2,…,yM]). T Hamiltonian decomposition and simulation of the quantum support vector machine matrix F are performed to construct a quantum circuit that realizes the unitary evolution operator exp(-iFτ), where τ is an artificially chosen time parameter. Represented in the quantum support vector machine matrix Under the influence of Hamiltonian, evolution time The corresponding You operator, the entire operator It refers to a subroutine that needs to be implemented via quantum circuitry on a quantum computer; in quantum states | As input, a quantum phase estimation algorithm is executed to estimate and obtain the phase-encoded state. The system of linear equations can be expanded as follows: in, K is the kernel matrix, γ is the regularization parameter, I is the identity matrix, and the superscript T represents the transpose of the matrix.

[0044] The steps involved in executing the quantum phase estimation algorithm are as follows.

[0045] A Hadamard gate is applied to the first register to generate a superposition state; using the first register qubit as the control bit, the controlled exp(-iF2) is executed sequentially. j τ) operation; perform inverse quantum Fourier transform on the first register and measure it, then store the result in the second register to obtain the phase-encoded state.

[0046] First, initialize the registers, setting the first register (used to store eigenvalues) to... The second register is | The number of qubits in the first register. This determines the accuracy of the eigenvalue estimation. Then, a Hadamard gate is applied to the first register to generate a uniform superposition state: Then, using the qubits of the first register as control bits, controlled execution is performed. Operation, among which Specifically, for the first One control bit (corresponding to) (power of power), apply controlled In fact, due to | >Can be decomposed into After controlled rotation, each eigenstate component accumulates phase. In this step, The implementation depends on the Hamiltonian simulation. Finally, an inverse quantum Fourier transform is performed on the first register to encode the phase information into the computational ground state, obtaining the phase-encoded state: in For quantum support vector machine matrices F eigenstates For the corresponding eigenvalues, yes of Bit-based binary approximation for The associated weights, when we measure the first register to get the first individual eigenvalues At that time, the second register is in the corresponding eigenstate. The probability amplitude is (probability is) ).

[0047] S34: Add auxiliary bits to... As control bits, a controlled rotation operation is performed on the auxiliary qubit to encode the reciprocal information of the eigenvalues ​​into the quantum state amplitude; subsequently, a measurement and post-selection operation is performed to obtain the parametric quantum state |b, α> representing the LSSVM model; the parametric quantum state |b, α> is then used to control the training data quantum state |b, α>. >Preparation (here referring to | >Re-preparation), generating a support vector quantum state |ũ> that encodes the parameter α into the amplitude: ,in, Referring to the previous text, For vectors The Euclidean norm.

[0048] In this step, the quantum state output from the previous step and the auxiliary qubit initialized to |0> are used as inputs. The calculated ground state encoded with phase information is used as the control bit. A controlled rotation operation (Ry(θj)) is performed on the auxiliary qubit, thereby encoding the reciprocal information of the eigenvalues ​​into the quantum state amplitude. Inverse quantum phase estimation is performed, followed by measurement and post-selection operations. The auxiliary qubit is measured, and if the measurement result is... The parameter quantum state |b,α> that characterizes the LSSVM model can be obtained if the measurement result is If the process fails, the above process is repeated. Quantum phase estimation and controlled rotation together form the core of HHL (solving linear equations).

[0049] use As a control mechanism, the quantum states of the training samples are selectively invoked. This is equivalent to adjusting the parameters... By combining with the corresponding eigenvector quantum state, a new superposition state |ũ> is formed. The quantum circuit that achieves this synthesis typically requires multiple controlled rotations and QRAM readouts. Its effect is equivalent to combining the parameter αi with the corresponding eigenvector quantum state, thereby forming a new superposition state.

[0050] LSSVM requires solving a large system of linear equations during training. When the number of training samples M is large, the time complexity of the classical algorithm reaches O(M³), and it also requires storing an O(M²) kernel matrix, making it difficult to meet the real-time and big data processing requirements of ultra-short-term rolling prediction. By employing quantum phase estimation, the HHL algorithm, and quantum state inner product calculation, the training complexity of LSSVM is reduced from... Down to The prediction complexity is from Down to This results in an exponential speedup. At this point, the quantum computing sub-steps already performed have completed the training of the quantum support vector machine model.

[0051] S4: Obtain wind power-related data for the time point to be predicted, including predicted wind speed, predicted wind direction, predicted temperature, and historical wind power datasets for multiple consecutive time points (including previous current time points). Perform the same preprocessing as in step S2 to obtain the feature vector X0 (normalized) of the sample to be predicted. The predicted wind speed, predicted wind direction, and predicted temperature for the time point to be predicted can be obtained from meteorological forecast data or predicted independently.

[0052] S5: Encode the feature vector of the sample to be predicted as the quantum state |ṽ0> of the test sample, and input it into the quantum support vector machine model trained in step S3. Calculate the inner product between the support vector quantum state |ũ> and the test sample quantum state |ṽ0> through quantum inner product operation. The inner product value is the wind power prediction value y(X0) under the normalized scale.

[0053] Figure 3This is a schematic diagram of a quantum circuit for calculating the inner product of eigenvector quantum states. The time complexity of calculating the inner product of two eigenvectors using quantum computing methods is O(logN). The SWAP module is a quantum controlled swap gate that operates on two qubits, allowing them to exchange qubits. Its logical structure can be composed of three NOT gates. The logic is relatively simple: after SWAP(A,B), if we define the qubit of A as 0 and the qubit of B as 1, after the logic gate operation, the observed result is that the qubit of A is 1 and the qubit of B is 0.

[0054] Based on the feature vector X0 of the test sample (which is constructed in the same way as the training sample, but the power part is the current real-time measurement value, and the meteorological part is the real-time value or forecast value), prepare the quantum state of the test sample. : = , .

[0055] Measured to obtain the expected value This value is the predicted value under normalized scaling. The calculation of quantum inner products utilizes the superposition and interference effects of quantum states, and can be performed... Parallel kernel function computation for all training samples is completed within a given time.

[0056] S6: Output Results: Perform inverse normalization on the normalized wind power prediction values ​​to obtain the wind power prediction results in physical units and output them.

[0057] If rolling prediction is performed, the newly obtained actual power is added to the historical data, the training set is updated, and the above process is repeated.

[0058] In some embodiments of this application, steps S3 and S5 above can be executed on a quantum processor, while the remaining steps are executed on a classical processor, with the classical processor and the quantum processor integrated in the same quantum computer.

[0059] The technical solution presented in this application can be viewed as a classical-quantum hybrid computing process. Core computational steps for wind power prediction, such as solving large-scale linear equations and quantum inner products, are performed by a quantum computer, resulting in exponential speedup. The classical component handles data preprocessing and post-processing, while the quantum component handles model training and prediction computation. This method overcomes the computational bottleneck of classical LSSVM under large data volumes, providing a feasible solution for real-time prediction of ultra-short-term wind power.

[0060] This application also provides a wind power prediction system, including a historical data acquisition module, a data preprocessing module, a quantum computing module, and a data post-processing module, as shown in the schematic diagram below. Figure 4The historical data acquisition module is used to execute step S1, acquiring historical wind power datasets and current wind power datasets. This historical data acquisition module can also be a signal transceiver module, used to send signal acquisition request signals to the outside world. When external data arrives via wired or wireless means, the signal transceiver module receives the signals sent by the specific object, acquires its data, and prepares to send it to the data preprocessing module.

[0061] The data preprocessing module executes step S2, normalizing each data point in the historical and current wind power datasets and using them as dimensional feature values ​​to construct the input feature vector. It then normalizes the actual power data corresponding to the next time point and uses it as the target variable to construct a supervised learning training sample set. Furthermore, the data preprocessing module also constructs the LSSVM linear equation system. Based on the normalized training sample set, it selects a preset kernel function type and kernel parameters, calculates the kernel function value between any two training sample feature vectors, forming the kernel matrix K; sets the regularization parameter γ, and constructs the linear equation system for solving the Least Squares Support Vector Machine (LSSVM). The data preprocessing module can be a classic computer motherboard module, integrating a CPU chip, memory, and a built-in preprocessing application program.

[0062] The quantum computing module includes a quantum processor, a quantum storage unit, and a quantum measurement and control platform, used to execute steps S3, S4, and S5. The quantum measurement and control platform includes an industrial control computer platform or a motherboard with integrated control chips (CPU, MCU). After receiving data from the data preprocessing module, the quantum measurement and control platform parses its signals into signals that the quantum processor can process. The quantum processor receives data signals and control signals from the measurement and control platform, prepares the corresponding quantum states, and performs unitary transformations on the quantum states according to the control signals to obtain the final quantum state. Statistical results are obtained by projecting measurements onto the final quantum state. The quantum measurement and control platform then transforms these statistical results to obtain the quantum computing results. The linear equation solving and quantum inner product calculation steps of the quantum support vector machine model in the wind power prediction method are performed within the quantum computing module.

[0063] For example, the quantum computing unit can be an optical quantum computer based on an optical quantum computing chip, which is its quantum processor. Data and control signals during the quantum computing process are converted into optical signals transmitted and evolved within the chip through the cooperation of a quantum measurement and control platform and the optical quantum computing chip.

[0064] The data post-processing module is used to execute step S6, performing de-normalization on the normalized wind power prediction value to obtain the wind power prediction result in physical units and output it. The data post-processing module can be a CPU chip or MCU chip of a classical computer, with a built-in post-processing application program. The data post-processing module can be integrated with the data post-processing module on the same CPU chip or MCU chip. The historical data acquisition module can be integrated with the data post-processing module and the data post-processing module on the same motherboard. In this scenario, the entire wind power prediction system can be built using a quantum computer. The historical data acquisition module, the data post-processing module, and the data post-processing module constitute the classical computing module. The classical computing module and the quantum computing module work together to complete the real-time prediction of wind power and feed the prediction data back to the power grid system, providing a basis for power regulation in the power grid system.

[0065] Figure 5 This diagram illustrates the collaborative operation of a wind power forecasting system with external sites and equipment. Historical wind power data is stored in a wind farm database along the route. Power data from each wind turbine in the wind farm is transmitted to the database via wired or wireless means for storage and immediate retrieval by the wind power forecasting system. Simultaneously, real-time power data from the turbines is also directly transmitted to the wind power forecasting system via wired or wireless means. Both real-time and historical power data are provided to the system for real-time forecasting. Historical or real-time meteorological data, including wind speed, wind direction, temperature, and humidity, is collected through data acquisition stations. For meteorological data, especially future weather forecasts, forecast data from weather stations can be used, stored and transmitted via the cloud or the internet.

[0066] In some other embodiments of this application, before step S3, a hyperparameter optimization step is also included: using time-series cross-validation, the regularization parameter γ and kernel width parameter σ of the quantum support vector machine model are optimized on a classical computer, and the validation window is strictly divided in time order to prevent future information leakage.

[0067] The regularization parameter γ and the kernel width parameter σ can also be optimized using quantum particle swarm optimization (PSO) algorithms, quantum genetic algorithms, and other similar algorithms. These algorithms can also run on quantum processors.

[0068] The various embodiments in this specification are described in a progressive, parallel, or combined manner. Each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other.

[0069] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or apparatus comprising a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or apparatus that includes the aforementioned element.

Claims

1. A wind power prediction method based on quantum computing, characterized in that, Includes the following steps: S1: Obtain historical wind power dataset and current wind power dataset. The historical wind power dataset includes historical actual wind power data and historical environmental factor data arranged in time series. The current wind power dataset includes real-time wind power data and real-time environmental factor data at the current time point. S2: Normalize each data point in the historical wind power dataset and the current wind power dataset, and use them as dimensional feature values ​​to construct the input feature vector. The normalized actual power data at the next time point is used as the target variable. Construct a training sample set {( , )}, where i=1, 2, ... M, and M is the total number of training samples; S3: Training is performed using the aforementioned training sample set. The training process specifically includes the following quantum computing sub-steps: S31: Encode the feature vector set of the training sample set into a first quantum state. Encode the target variable set as a second quantum state. >; S32: Select a preset kernel function type and kernel parameters, calculate the kernel function value between any two training sample feature vectors, and form a kernel matrix K; utilize the quantum state | >, prepare the quantum superposition state |χ> of all training data feature vectors, and obtain the ratio information of the kernel matrix K and its trace tr(K) by calculating the reduced density matrix of the specified subsystem in the superposition state; S33: Set the regularization parameter γ, and construct a system of linear equations for solving the least squares support vector machine (LSSVM): F·[b;α]=[0;y], where F is the quantum support vector machine matrix, b is the bias term, α is the Lagrange multiplier vector, and y is the vector composed of the target values ​​of all training samples; perform Hamiltonian decomposition and simulation on the quantum support vector machine matrix F, and construct a quantum circuit to realize the unitary evolution operator exp(-iFτ); using quantum state | As input, a quantum phase estimation algorithm is executed to obtain the phase-encoded state; S34: Add an auxiliary bit, perform a controlled rotation operation on the auxiliary bit, and encode the reciprocal information of the eigenvalues ​​into the quantum state amplitude; then perform measurement and post-selection operations to obtain the parametric quantum state |b,α> representing the LSSVM model, and use the parametric quantum state |b,α> to control the training data quantum state |b,α>. The preparation of > generates a support vector quantum state |ũ> that encodes the parameter α into the amplitude; S4: Obtain wind power-related data for the time point to be predicted. The wind power-related data for the time point to be predicted includes the predicted wind speed, predicted wind direction, predicted temperature, and historical wind power datasets for multiple consecutive time points. Perform the same preprocessing as in step S2 to obtain the feature vector X0 of the sample to be predicted. S5: Encode the feature vector of the sample to be predicted as the quantum state |ṽ0> of the test sample, input it into the quantum support vector machine model trained in step S3, and calculate the inner product of the support vector quantum state |ũ> and the quantum state |ṽ0> of the test sample through quantum inner product operation. The inner product value is the wind power prediction value y(X0) under the normalized scale. S6: Output Results: Perform inverse normalization on the normalized wind power prediction value to obtain the wind power prediction result in physical units and output it.

2. The wind power prediction method based on quantum computing according to claim 1, characterized in that, The step between step 1 and step 2 also includes a data cleaning step: performing at least one of the following processing on the historical wind power dataset and the current wind power dataset: outlier detection and removal, missing value imputation, and noise filtering.

3. The wind power prediction method based on quantum computing according to claim 1, characterized in that, In step 2, the normalization of the historical wind power dataset and the current wind power dataset adopts the min-max normalization method to linearly transform the data to the [0, 1] interval; the normalization of the actual power data adopts the logarithmic normalization method.

4. The wind power prediction method based on quantum computing according to claim 1, characterized in that, Steps S3 and S5 are executed on the quantum processor, while the remaining steps are executed on the classical processor, which is integrated with the quantum processor in the same quantum computer.

5. The wind power prediction method based on quantum computing according to claim 1, characterized in that, The periodic encoding process in step S2 specifically involves converting the wind direction WD into a two-dimensional vector in the form of an angle. To maintain the periodic continuity between 0° and 360°, Let t be the wind direction at time t.

6. The wind power prediction method based on quantum computing according to claim 1, characterized in that, In step S3, the kernel function type is a Gaussian radial basis function, and its expression is: Where σ is the kernel width parameter, Let be the feature vectors of any two training samples.

7. The wind power prediction method based on quantum computing according to claim 1, characterized in that, In step S3, the feature vector set of the training sample set is encoded into a first quantum state using a quantum random access memory and an Oracle operator module. The specific expression is: , where x j Let |x| represent the feature vector of the j-th sample in the training dataset. j | represents the eigenvector x j The length of the mold, For the j-th eigenvector, the k-th value is... It is a set of orthogonal bases in the quantum system space.

8. The wind power prediction method based on quantum computing according to claim 1, characterized in that, In step S3, the vector y is encoded as a quantum state |y>: ,in Let y be the Euclidean norm of vector y.

9. The wind power prediction method based on quantum computing according to claim 1, characterized in that, In step S3, the quantum phase estimation algorithm includes: applying a Hadamard gate to the first register to generate a superposition state; using the first register qubit as the control bit, sequentially executing controlled exp(-iF2)... j The operation is performed with τ as a preset time parameter; the inverse quantum Fourier transform is performed on the first register and measured, and the result is stored in the second register to obtain the phase-encoded state.

10. A quantum computing-based wind power prediction system, used to execute the quantum computing-based wind power prediction method according to any one of claims 1-9, characterized in that, It includes a historical data acquisition module, a data preprocessing module, a quantum computing module, and a data postprocessing module; The historical data acquisition module is used to execute step S1; The data preprocessing module is used to perform step S2; The quantum computing module includes a quantum processor, a quantum storage unit, and a quantum measurement and control platform, which are used to run steps S3, S4, and S5. The data post-processing module is used to run step S6.

Images