A photovoltaic power prediction method and system based on an improved RFNN-QLSTM

Abstract

The application discloses a photovoltaic power prediction method and system based on an improved RFNN-QLSTM of a VECM, relates to the photovoltaic power prediction field, and comprises the following steps: preprocessing original photovoltaic data; a vector error correction model is constructed to capture the balanced regression relationship between the photovoltaic data and meteorological factors and between the meteorological factors, an error correction term is introduced on the vector autoregressive model to identify the co-integration relationship and short-term deviation between variables in the photovoltaic data, and a new photovoltaic data set is screened and constructed through the vector error correction model; a photovoltaic power short-term prediction model is constructed, the new photovoltaic data is used to train the photovoltaic power short-term prediction model, and photovoltaic output prediction results are predicted and output. The application effectively solves the problem that the balanced relationship and the co-integration relationship between photovoltaic data and meteorological factors are not considered, and provides a new research idea for photovoltaic power prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of photovoltaic power prediction, and more particularly to a photovoltaic power prediction method and system based on an improved RFNN-QLSTM VECM. BACKGROUND

[0002] Photovoltaic power prediction is a key technology to reduce the adverse effects of photovoltaic output randomness. Generally, based on dynamic data such as meteorological forecast data, photovoltaic power data, and meteorological observation data, combined with information such as photovoltaic station location and equipment parameters, a model is established to predict the output trend of photovoltaic power in the future. This can effectively support power grid dispatching, maintenance planning, security and stability analysis, new energy consumption analysis, and other businesses.

[0003] Photovoltaic power prediction methods are generally divided into two categories: physical models and data-driven models. The prior art proposes an improved deep learning algorithm that applies long short-term memory networks and artificial neural networks to the prediction model. This research mainly focuses on parameter optimization and model optimization, and shows better performance compared to a single deep learning algorithm.

[0004] The fluctuation characteristics of the original data caused by changes in weather conditions can affect the accuracy of the prediction. Analyzing and reducing the fluctuation characteristics of the original data can further improve the accuracy of short-term photovoltaic power prediction methods. The prior art also uses wavelet packet decomposition technology to analyze photovoltaic power time series. Compared with traditional methods, this method integrates the decomposed time series into the prediction model, reducing the prediction error. However, this method focuses on capturing fluctuation characteristics at a small scale, ignoring the fact that photovoltaic power fluctuations are actually caused by changes in weather conditions. In the prior art, clustering analysis is performed on meteorological factors to select similar days in the initial data, significantly reducing the photovoltaic output prediction error. However, this method ignores the degree of influence of meteorological factors on photovoltaic power fluctuations and only clusters with one meteorological factor or simply clusters with multiple meteorological factors, which is not conducive to the extraction of fluctuation characteristics.

[0005] Quantum computing can use quantum mechanics laws to process information types that are difficult to implement with traditional methods. Considering the characteristics of photovoltaic power prediction around continuous value regression, the advantage of quantum support vector machine is mainly to solve discrete classification problems, and the adaptability and prediction ability of photovoltaic power time series prediction are not strong. The prior art proposes to improve existing artificial intelligence prediction methods in the field of quantum computing, but the adaptability and prediction ability of existing quantum improvement methods for photovoltaic power prediction still have room for improvement. Therefore, an improved RFNN-QLSTM photovoltaic power short-term prediction model based on a vector error correction model (VECM) is provided for photovoltaic power prediction, which improves the adaptability and accurate prediction ability of photovoltaic power prediction, and has certain engineering significance. SUMMARY

[0006] Therefore, the purpose of the present application is to provide an improved RFNN-QLSTM photovoltaic power prediction method and system based on a VECM, which improves the adaptability and accurate prediction ability of photovoltaic power prediction.

[0007] To achieve the above purpose, the present application adopts the following technical solutions:

[0008] An improved RFNN-QLSTM photovoltaic power prediction method based on a VECM, comprising:

[0009] Pretreating the original photovoltaic data for outliers and missing values;

[0010] Constructing a vector error correction model to capture the equilibrium regression relationship between photovoltaic data and meteorological factors, introducing an error correction term based on the vector autoregressive model to identify the cointegration relationship and short-term deviation between variables in the photovoltaic data, and screening relevant features in the photovoltaic data through the vector error correction model and constructing a new photovoltaic data set;

[0011] Constructing an improved RFNN-QLSTM photovoltaic power short-term prediction model based on a vector error correction model VECM.

[0012] Training the photovoltaic power short-term prediction model using the new photovoltaic data set, predicting and outputting the photovoltaic power prediction result.

[0013] Preferably, an error correction term is introduced based on the vector autoregressive model to obtain a vector error correction model, and the formula is as follows:

[0014]

[0015] where ΔY t is the difference of the time series variable at time t, representing the short-term dynamic relationship, and Γ iis the coefficient matrix of the lagged period i, ΔY t represents the time series under study, p is the lag length, Π is the impact matrix containing information about long-run relationships, and ε t is the n×1 error vector, which is used to estimate the likelihood ratio (LR) to determine the cointegrating vectors in the system ΔY t .

[0016] Preferably, the completed vector error correction model is subjected to Johansen cointegration relationship test and model stability test in turn, and cointegration result analysis is performed to identify the variable relationship in the photovoltaic data.

[0017] Preferably, the method for constructing the quantum long short-term memory network model comprises:

[0018] The quantum long short-term memory network model replaces the long short-term memory network unit with a variational quantum circuit to form a quantum long short-term memory network unit;

[0019] The quantum long short-term memory network model architecture is defined as:

[0020] f t = σ(VQC1(v t )),

[0021] i t = σ(VQC2(v t )),

[0022]

[0023] o t = σ(VQC4(v t )),

[0024] h t = o t × tanh(c t ),

[0025] Preferably, the quantum long short-term memory network further comprises:

[0026] Input layer: responsible for receiving and processing original time series data;

[0027] Output layer: converts the processing result of the quantum long short-term memory network model into the final output for the next step of time series prediction.

[0028] Preferably, the improved recurrent fuzzy neural network model comprises: replacing the recurrent layer in the recurrent fuzzy neural network with a quantum long short-term memory network to obtain the improved recurrent fuzzy neural network model.

[0029] The application also provides a photovoltaic power prediction method based on the improved RFNN-QLSTM of the VECM, which comprises the following modules:

[0030] A data preprocessing module is used for preprocessing outliers and missing values of original photovoltaic data.

[0031] A vector error correction module is used for capturing the balanced regression relationship between photovoltaic data and meteorological factors and meteorological factors, and introducing an error correction term based on a vector autoregressive model to identify the cointegration relationship and short-term deviation between variables in the photovoltaic data.

[0032] A photovoltaic power short-term prediction module is used for constructing an improved recurrent fuzzy neural network model in which a recurrent layer in a recurrent fuzzy neural network is replaced and improved by a quantum long short-term memory network, training a photovoltaic power short-term prediction model by using a new photovoltaic data set constructed by screening relevant features by the vector error correction model, and predicting and outputting photovoltaic output prediction results.

[0033] Compared with the prior art, the application discloses a photovoltaic power prediction method based on the improved RFNN-QLSTM of the VECM, which has the following advantages. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description only constitute a part of the embodiments of the present application, and all other drawings obtained by those of ordinary skill in the art without creative effort based on the provided drawings also belong to the protection scope of the present application.

[0035] Figure 1 A flowchart of a photovoltaic power prediction method based on an improved RFNN-QLSTM VECM provided by the present application.

[0036] Figure 2 A vector error correction model modeling flowchart provided by the present application;

[0037] Figure 3 A variational quantum circuit schematic diagram provided by the present application;

[0038] Figure 4 A quantum long and short term memory network structure diagram provided by the present application;

[0039] Figure 5 An improved recurrent fuzzy neural network structure diagram provided by the present application;

[0040] Figure 6 An AR root diagram provided by the present application;

[0041] Figure 7 A photovoltaic power prediction result diagram from 00:00 on August 22, 2020 to 23:55 on August 28, 2020 provided by the embodiment of the present application;

[0042] Figure 8 A photovoltaic power prediction result diagram on August 25, 26 and 27, 2020 provided by the embodiment of the present application. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort belong to the protection scope of the present application.

[0044] Embodiment 1

[0045] As shown in Figure 1 The present application discloses a photovoltaic power prediction method based on an improved RFNN-QLSTM VECM, which comprises the following steps:

[0046] Anomaly value and missing value preprocessing is performed on original photovoltaic data;

[0047] A vector error correction model (VECM) is constructed to capture the balanced regression relationship between photovoltaic data and meteorological factors, an error correction term is introduced into the vector autoregressive model to identify the cointegration relationship and short-term deviation between variables in the photovoltaic data, relevant features are screened out through the vector error correction model, and a new photovoltaic data set is constructed;

[0048] In the embodiment of the application, the relevant features refer to other meteorological feature statistical values in the original photovoltaic data set except for the power statistical value, and specific reference can be made to Table 1 (the features on the left side in Table 1), six features with high correlation with power are screened out through the error correction model, and the new photovoltaic data set is a data set containing the six features and the power statistical value.

[0049] An improved recurrent fuzzy neural network-quantum long short-term memory network (RFNN-QLSTM) photovoltaic power short-term prediction model based on the vector error correction model (VECM) is constructed.

[0050] Further, the photovoltaic power short-term prediction model is an improved recurrent fuzzy neural network model in which the recurrent layer in the recurrent fuzzy neural network is replaced and improved by the quantum long short-term memory network.

[0051] The photovoltaic power short-term prediction model is trained by using the new photovoltaic data set constructed by screening the relevant features through the vector error correction model, and the photovoltaic output prediction result is predicted and output.

[0052] Further, the error correction term is introduced into the vector autoregressive model to obtain the vector error correction model, and the formula is as follows:

[0053]

[0054] where ΔY t is the difference of the time series variable at time t, representing the short-term dynamic relationship, Γ i is the coefficient matrix of lag i, ΔY t represents the time series under study, p is the lag length, Π is the influence matrix containing information about the long-term relationship, and ε t is an n×1 error vector; in Johansen cointegration test, the error term is needed to estimate the likelihood ratio (LR) to determine the cointegration vector of ΔY t in the system.

[0055] Specifically, Figure 2A flowchart of the vector error correction model (VECM) modeling process is shown in FIG. 1. The VECM requires that all variables be stationary time series, so the Augment DuckyFuller (ADF) unit root test is first performed to test the stationarity of the variables, i.e., whether there is a unit root. Then, the lag length is determined based on the Akaike information criterion (AIC), the Schwarz information criterion (SIC), and the likelihood ratio (LR) comprehensive evaluation.

[0056] Further, the completed vector error correction model is subjected to Johansen cointegration test and model stability test in sequence, and the cointegration results are analyzed to identify the variable relationships in the photovoltaic data.

[0057] The variable relationships include long-term equilibrium relationships and short-term deviations. The long-term equilibrium relationship refers to a relationship in which variables in photovoltaic data (such as photovoltaic power generation, meteorological factors, etc.) tend to be stable in the long term. This relationship can be determined by statistical methods such as cointegration analysis. Cointegration indicates that although these variables may be non-stationary in the short term, there is a long-term stable equilibrium relationship between them.

[0058] Short-term deviations refer to the actual values of these variables deviating from their long-term equilibrium values in the short term due to various random factors (such as weather changes, equipment failures, etc.). This deviation is short-term and usually returns to equilibrium through other mechanisms in the long term.

[0059] The vector error correction model (VECM) captures this short-term deviation by introducing an error correction term (ECT). The error correction term is a function of the difference between the long-term equilibrium relationship and the actual value, which measures the degree of deviation of the system from the long-term equilibrium state; in the equation of VECM, this correction term usually appears with a negative coefficient, indicating that the system will adjust towards the long-term equilibrium relationship, i.e., if there is a positive deviation, the system will tend to reduce this deviation in the future, and vice versa.

[0060] In photovoltaic power prediction, identifying and correcting these short-term deviations is crucial to improving the accuracy of the prediction; through the VECM model, the short-term dynamic relationship between photovoltaic power generation and meteorological factors can be analyzed, and the long-term equilibrium relationship between them is considered, thus providing more accurate short-term power prediction.

[0061] Further, the present application adopts a vector error correction model, which assumes that all variables have the same integral order; the analysis of the vector error correction model is carried out under the framework of the vector autoregressive model.

[0062] Vector autoregression model is a econometric model that uses all current variables in the model to regress on several lagged variables, which is usually used to analyze the dynamic relationship between jointly endogenous variables and does not require any prior constraints. The principle of vector autoregression model is to construct a model for each endogenous variable as a function of lagged values, and to extend the single variable autoregression model to a vector autoregression model composed of multiple time series variables. According to the statistical characteristics of photovoltaic data, a vector autoregression model is established.

[0063] The vector autoregression model is represented as:

[0064] y t = c + A1y t-1 + A2y t-2 + … + A p y t-p + ξ t ;

[0065] where y t is the vector of endogenous variables, p is the lag length, c is the n x 1 constant vector, and A is the n x n matrix, ξ t is the n x 1 error vector. In practical applications, the vector autoregression model is usually converted to a matrix form. For example, the above vector autoregression model can be represented as follows (lag length is 2):

[0066]

[0067] In the modeling process, two indicators need to be determined: one is the number of variables, because the vector autoregression model needs to identify which characteristic variables in the given data are related to each other, and these variables should be included in the vector autoregression model; the second is the maximum lag length of the autoregression model, and a reasonable lag length can better reflect the relationship between variables. In the vector autoregression model, variables with correlation should be selected as explanatory variables, and Granger causality test is used to determine whether there is correlation between these variables. The mechanism of Granger causality test is that for two given time series (such as X, Y) that satisfy the stationary random process, using the historical data of X and Y to predict Y, if the prediction result is better than using the historical data of Y alone, it is considered that X helps to improve the prediction accuracy of Y, and there is a causal relationship between X and Y.

[0068] Since the long-term historical photovoltaic data is used in the embodiment, the unit root property of the variables in the data needs to be tested by unit root test, which is used to determine whether the characteristic variable has a unit root and is stationary. In a time series, if the variance and mean of a variable remain unchanged over time, the sequence is called a stationary sequence, i.e., the sequence has no unit root. In a vector autoregressive model, Augment DuckyFuller (ADF) unit root test and Ducky Fuller (DF) test methods are usually used to test the unit root property and the stationary state of the characteristic variable, where the ADF unit root test is an improved form of the DF test.

[0069] The formula of the ADF unit root test is as follows:

[0070] ΔY t = (p - 1) Y t-1 + y1ΔY t-1 + y2ΔY t-2 +... + y p ΔY t-p + ε t ;

[0071] where Δ represents difference, Yt represents the time series under examination, ε t represents an error term, p is the parameter to be tested, the value of p is determined by the ADF unit root statistic, if p = 1, the time series has a unit root, and the sequence is non-stationary, if p < 1, the sequence is stationary, y1, y2, …, y p are parameters of lagged difference terms, reflecting the influence of previous difference data on the current value, and p is the lag order; if the absolute value of the ADF unit root test statistic is greater than the critical value, or the p value (the second element in the return value) is less than the significance level (for example, 0.05), the null hypothesis of the existence of the unit root is rejected, i.e., the variable is accepted as stationary (no unit root), and the output result of the ADF unit root test needs to be compared with the critical value of a specific order to determine whether to accept or reject the null hypothesis.

[0072] The 1%, 5%, and 10% critical values in the ADF unit root test correspond to the significance level (or confidence level) selected in the study. These values are commonly selected in statistical decision-making, and they represent the probability of the first type of error (i.e., mistakenly rejecting the null hypothesis) that the model is willing to tolerate. For example, a 1% critical value corresponds to a 99% confidence level, which means that there is a 99% chance that the decision is correct, and only a 1% chance of making a first type of error, i.e., rejecting the null hypothesis. The ADF test commonly uses three empirical models: a model containing a constant term, a model containing both a constant term and a trend term, and a model containing neither a constant term nor a trend term. When performing the ADF test, all three models should meet the statistical requirements.

[0073] Vector autoregressive model is an effective model for analyzing and predicting multiple related indicators. However, when analyzing photovoltaic output data and meteorological factors, the vector autoregressive model only considers the short-term volatility between photovoltaic data and meteorological factors, and does not consider the long-term equilibrium relationship and cointegration relationship between photovoltaic data and meteorological factors and meteorological factors.

[0074] Therefore, an error correction term is introduced on the basis of the vector autoregressive model to continue to test the long-term equilibrium relationship and cointegration relationship in the photovoltaic data. There are several methods for cointegration, including Engle and Granger cointegration test and Johansen cointegration test. However, the Engle and Granger cointegration test does not specify the number of vectors, and the Johansen test can provide information about the number of cointegration vectors. Compared with the Engle and Granger test, the Johansen cointegration test is suitable for multiple variables, and can detect whether there is one or more cointegration relationships between multiple variables. In comparison, the Johansen cointegration test method has more advantages. In the Johansen cointegration test, the error term is needed to estimate the likelihood ratio to determine the cointegration vectors in the cointegration system. The Johansen cointegration test formula is as follows:

[0075]

[0076] Where ΔY t is the difference of time series variable at time t, representing the short-term dynamic relationship, Γ i is the coefficient matrix of lag i, Y t represents the time series under study, p is the lag length, Π is the influence matrix, which contains information about the long-term relationship, and ε t is an n x 1 error vector. In the Johansen cointegration test, the error term is needed to estimate the likelihood ratio (LR) to determine the cointegration vectors in the system Y t .

[0077] Because the cointegration test is carried out under the vector autoregressive model setting, the study uses Akike information criterion and Shward information criterion to estimate the appropriate lag length. After detecting the cointegration relationship, we can continue to identify the short-term deviation between variables in the photovoltaic data from the long-term equation of the vector error correction model. The present invention establishes a vector error correction model to analyze multiple related photovoltaic data variables, and observes whether the selected response variable is affected by other variables.

[0078] Further, the construction method of the quantum long short-term memory network model (QLSTM) comprises:

[0079] The quantum long short-term memory network model uses a variational quantum circuit to replace a long short-term memory network unit to form a quantum long short-term memory network unit;

[0080] The quantum long short-term memory network model architecture is defined as follows:

[0081] f t =σ(VQC1(v t )),

[0082] i t =σ(VQC2(v t )),

[0083]

[0084] o t =σ(VQC4(v t )),

[0085] h t =o t ×tanh(c t ).

[0086] Furthermore, quantum long short-term memory networks also include:

[0087] Input layer: Responsible for receiving and processing raw data;

[0088] Variable quantum circuit unit: Each long short-term memory network unit contains 4 variable quantum circuits, which correspond to the 4 gates of the long short-term memory network and are used to perform quantum feature extraction and data compression.

[0089] Output layer: Converts the processing results of the quantum long short-term memory network into the final prediction output for the next step of time series prediction.

[0090] like Figure 3 The schematic diagram of the variable quantum circuit shown is a quantum circuit with adjustable parameters and iterative optimization. The main functions of the variable quantum circuit are feature extraction and data compression. The variable quantum circuit consists of three parts: a data encoding layer, a variable layer, and a quantum measurement layer. The data encoding layer first encodes the data information into quantum states of qubits. Then, the variable layer, containing trainable weights, performs entanglement operations. The evolved quantum states are then measured to obtain quantum measurement values ​​for subsequent operations. The specific implementation process is as follows:

[0091] 1) Data encoding layer: The encoding layer contains H, R x R y R z Revolving door. Quantum embedding represents classical data as quantum states in Hilbert space through quantum feature mapping. First, a classical data point x is taken and transformed into a set of parameters in a variable quantum circuit, generating the quantum state |ψ|. xThe data is encoded into the amplitude of the quantum state by the amplitude embedding method. A normalized classical n-dimensional data point x is represented by the quantum state |ψ x in amplitude formula as:

[0092]

[0093] where N = 2 n , x i is the i-th element of the n-dimensional data point, and |i> is the i-th computational basis.

[0094] 2) Variational layer: After the classical state data is encoded into the quantum state, it is passed to the variational layer. After the data enters the variational layer, a number of entanglement operations are performed, which are composed of several controlled inverse gates and single-qubit rotation gates. Controlled inverse gates are used for each pair of quantum bits with fixed adjacency (in a cyclic manner) to produce multi-qubit entanglement. The single-qubit rotation gate (R x , R y , R z ) is shown in Figure 3 :

[0095] R(α i ,β i ,γ i ) = R x (α i )R y (β i )R z (γ i );

[0096] where α i , β i and γ i are the iteration optimization parameters of the variational quantum circuit VQC.

[0097] 3) Quantum measurement layer: The variational layer is followed by the quantum measurement layer. On the basis of the calculation, the expected value of each quantum bit is obtained by the Pauli matrix. Using the quantum simulation software PennyLane, the expected value of each quantum bit can be calculated on a classical computer. After a series of quantum entanglement operations, the output of the variational quantum circuit is a fixed-length classical vector. In the quantum long short-term memory network, the measurement values from each variational quantum circuit will be processed in a quantum long short-term memory network unit.

[0098] The application uses the PennyLane quantum machine learning framework to realize the application of a variational quantum circuit in a long short-term memory network. The core of PennyLane is to combine quantum and traditional computing to help build and refine models. One of the outstanding features of PennyLane is that it can seamlessly combine quantum elements based on variational circuits with conventional artificial neural network parts, and can create complex structures such as quantum long short-term memory networks. The quantum long short-term memory network model mixes a traditional long short-term memory network model and quantum nodes that take into account time series data when designed, using specific rotation superposition and entanglement operations.

[0099] The quantum long short-term memory network replaces the long short-term memory network unit with a variational quantum circuit to form a quantum long short-term memory network unit. The variational quantum circuit uses a small number of qubits and gates to represent a complex function. There are four variational quantum circuits in a quantum long short-term memory network unit, namely VQC1, VQC2, VQC3 and VQC4, and their outputs are vectors obtained from the quantum measurement layer. The measurement value is the Pauli expectation value designed for each qubit, and then passes through a nonlinear activation function. The four variational quantum circuits are the same and contain an encoding layer that compresses the dimension of the information to a circuit size suitable for the variational quantum circuit, and then obtains a quantum measurement value through a variational layer. The quantum measurement value is continuously sent to the measurement layer, which converts the quantum measurement value into an output result that conforms to the original data structure.

[0100] As shown in the quantum long short-term memory network structure diagram, Figure 4 σ and tanh represent the s-type function and the hyperbolic tangent activation function respectively. x t is the input at time t, C t represents the cell state; the four variational quantum circuits embedded in each quantum long short-term memory network unit process different logical operations respectively.

[0101] Further, the improved recurrent fuzzy neural network model includes: replacing the recurrent layer in the recurrent fuzzy neural network with a quantum long short-term memory network to obtain an improved recurrent fuzzy neural network model, wherein the improved recurrent fuzzy neural network model includes an input layer, a fuzzy rule layer, a normalization layer, a recurrent layer (QLSTM) and an output layer.

[0102] A recurrent neural network (RNN) is a type of neural network specifically designed to handle sequential data. RNNs are capable of learning and remembering past observations through their structure, which forms a loop of information. Compared to feedforward neural networks with the same approximation accuracy, RNNs perform better in terms of computational power and compactness. Despite the good predictive ability of RNNs, training them can become complex due to the vanishing or exploding gradient problem, which can lead to slow convergence and high computational demands.

[0103] A fuzzy neural network (FNN) is a hybrid approach that combines the learning capabilities of artificial neural networks with the explainability and semantic transparency of fuzzy systems. FNNs have significant advantages in local representation and simulating human reasoning, and are effective in handling uncertainty, allowing them to effectively capture potential relationships in time series data. To combine the potential of RNNs to learn time dependencies with the ability of FNNs to handle fuzzy information, a recurrent fuzzy neural network (RFNN) is introduced, and the recurrent layer in the RFNN is improved by replacing it with a quantum long short-term memory network.

[0104] The various structures in the improved RFNN play different roles in handling fuzzy information and time series data; the functions of each level are described in detail below:

[0105] As shown in the RFNN structure diagram, it includes: Figure 5

[0106] (1) Input layer

[0107] The input layer is the starting point of the neural network, which receives the original input data (usually time series data).

[0108] The role of the membership function is to fuzz the input data. The process of fuzzification is to convert specific input values into membership degrees of fuzzy sets. Each input feature is processed by one or more membership functions to generate membership values associated with fuzzy sets. These membership functions are usually some predefined functions, such as:

[0109] Gaussian function: commonly used to represent fuzzy sets smoothly;

[0110] Triangular function: simple and computationally efficient, suitable for linear fuzzification.

[0111] The output of the membership function is the fuzzy membership of the input data, and each function has K fuzzy rules, which are used for further fuzzy reasoning and recursive processing. In the improved RFNN, the input layer passes these original data to the next layer after calculating the membership function values through the membership function.

[0112] ​The layer takes input variables, whose neurons correspond to membership functions. It is used to calculate the membership values of input variables. Let x = [x1, x2,..., xd]Tbe the input vector, and A = [A1, A2,..., AK]Tbe the output vector of this layer. d ] T where x represents the input, and y represents the predicted output. A i，j is the membership value of x j in the membership function of the i-th rule in the output network, the j-th input variable x j is the membership value of x i in A , thus the output of each neuron A i，j in this layer is the membership value of x j in A i，j , i.e. The output network of this layer has Kxdneurons respectively.

[0113] (2) Fuzzy Rule Layer

[0114] The rule layer applies fuzzy rules to process the fuzzified input data. Fuzzy rules are usually described in the form of "if-then", in this layer, each fuzzy rule generates a fuzzy output by calculating the membership degree matching the input data. These rules can be defined based on expert knowledge or automatically generated by data-driven methods.

[0115] The neurons of this layer represent fuzzy rules, and their outputs represent the firing strength of the rules. Let r i denote the output of the i-th neuron of this layer, which is used for the output network. They can be calculated by applying T-norm operators on the outputs of the previous layer.

[0116] Using algebraic product as the T-norm operator, the firing strength of each rule can be calculated by:

[0117]

[0118] (3) Normalization Layer

[0119] The output of the neurons of this layer represents the normalized firing strength of each rule (r i for the output network), which can be calculated by:

[0120]

[0121] (4) Improved Long Short-Term Memory Network Layer (Recurrent Layer)

[0122] QLSTM is used as the recurrent layer of RFNN. The layer is composed of input gate I d (t), forget gate F d (t), candidate quantity output gate O d (t) and cell state Cd (t) constitutes an internal feedback loop. At time step t, the loop rule node H d (t) is determined by O d (t) and C d (t) in combination, which can be calculated by the following formula:

[0123]

[0124] H d (t) = O d (t) * tanh(C d (t));

[0125] where the definitions of the activation functions are as follows:

[0126] (5) Output layer

[0127] The output layer receives the numerical results generated by the improved long short-term memory network layer and calculates the final output y by the following formula.

[0128]

[0129] where w d is the output weight value connected to the dth fuzzy rule.

[0130] Example 2

[0131] (1) Vector error correction model experimental results

[0132] 1) Unit root test:

[0133] According to the central limit theorem, when the sample size is large enough, the mean distribution of independent and identically distributed random variables will tend to be normally distributed, so the distribution of the test statistic is derived using the central limit theorem in the unit root test. Two statistical values are derived using the central limit theorem through ADF unit root test, and a new critical value is obtained for significance test. Then compare the statistical value with the critical value, when the T statistical value is greater than the critical value, accept the hypothesis, the data sequence is a non-stationary process, otherwise the data sequence is a stationary process. ADF unit root test commonly uses three empirical models: model 1, a model containing a constant term (containing an explained variable lag, an error term lag and a constant term); Model 2, a model containing a constant term and a trend term (adding a linear trend term about time to model 1); Model 3, a model not containing a constant term and a trend term (deleting the constant term from model 1). When performing ADF unit root test, the three models should meet the statistical requirements.

[0134] Taking the photovoltaic data of 69-Site_17-Sanyo power station in Australia as an example, the ADF unit root test is performed on the photovoltaic data of the power station. The ADF statistic, i.e. T statistic, is obtained by dividing the estimated value of the parameter to be tested in the ADF unit root test model by the standard error (SE) of the estimated value. Then, the P value is obtained according to a specific critical value table. The P value is an index for measuring the consistency between the observed data and the null hypothesis in hypothesis testing, which represents the probability of observing the data or more extreme results under the premise that the null hypothesis is true. The test results are shown in Table 1. From the ADF unit root test results, it can be seen that, except for the horizontal scattered radiation statistic value, the inclined scattered radiation statistic value, the wind direction statistic value and the rainfall statistic value, the T statistic values of the other characteristic variables corresponding to the three models are greater than the critical values at the 1%, 5% and 10% confidence intervals. Therefore, the original data of the other characteristic variables except for the horizontal scattered radiation statistic value are non-stationary sequences.

[0135] ADF unit root test results of original data (5% confidence interval)

[0136]

[0137] P value * <0.01

[0138] From Table 1, it can be seen that the horizontal scattered radiation, the inclined scattered radiation, the wind direction and the rainfall are stationary sequences. After being removed, only the new data containing the non-stationary sequences in the original data are obtained. In order to meet the requirements of the error correction model, the first-order difference is used to process the new data to achieve stationarity. After processing, the new ADF unit root test is performed to verify the stationarity. After the ADF unit root test, the test statistic value results of the three models show that the power, the current average phase, the photovoltaic system efficiency and the temperature variables are less than the critical values at the 1%, 5% and 10% levels, indicating that the new data sequence is a stationary time sequence, as shown in Table 2.

[0139] ADF test results of new data under first-order difference (5% confidence interval)

[0140]

[0141] P value * <0.01

[0142] 2) Lag length determination:

[0143] In the determination of lag length, the Akaike information criterion (AIC information criterion), Schwarz criterion (SC information criterion) and LR statistics need to be considered comprehensively. Eviews software is used to determine the optimal delay length, and the Eviews test results show that when the lag length is 4, the AIC information criterion, SC information criterion and LR statistics are significant, so the optimal lag length in this study is 4.

[0144] 3) Cointegration test results:

[0145] Through the analysis of the results in Table 2, the power, current average phase, horizontal global radiation, photovoltaic system efficiency, inclined global radiation, relative humidity and temperature are selected as the new data set, and the Johansen cointegration trace test and maximum eigenvalue test are performed on the new data set. The obtained Table 4 and Table 5 are shown below

[0146] Table 4 unrestricted cointegration rank test (trace test)

[0147]

[0148] When the hypothesis of 0 cointegration relationship exists, the trace statistic is 833.7068, which is greater than the critical value of 5% 95.7536, and the P value is 0.0001, which is less than 0.05. Therefore, the hypothesis is rejected, that is, there is at least one cointegration relationship, and it continues to the hypothesis of 5 cointegration relationships, at this time the trace statistic is 25.7091, which is greater than the critical value of 5% 3.8414, and the P value is less than 0.05, the original hypothesis is rejected, that is, there are at least 6 cointegration relationships.

[0149] Table 5 unrestricted cointegration rank test (maximum eigenvalue test)

[0150]

[0151] When the hypothesis of 0 cointegration relationship exists, the maximum eigenvalue statistic is 860.6586, which is greater than the critical value of 5% 117.7082, and the P value is less than 0.05, so the hypothesis is rejected, that is, there is at least one cointegration relationship, and it continues to the hypothesis of 5 cointegration relationships, at this time the maximum eigenvalue statistic is 27.8952, which is greater than the critical value of 5% 12.5179, and the P value is 0.0001, which is less than 0.05, the original hypothesis is rejected, that is, there are at least 6 cointegration relationships.

[0152] 4) Model stability test:

[0153] As Figure 6The AR root plot shown is used to demonstrate the stability of the vector error correction model. The blue dots represent the eigenvalues, and when the dots are on the horizontal line, the eigenvalues are real values, and the eigenvalues that are not on the horizontal line are imaginary values. When all the eigenvalues are within the unit circle, the model is stable. As can be seen from the figure, all the eigenvalues are within the unit circle, so the established model is stable.

[0154] From the above analysis, it is known that the current average phase, horizontal global radiation, photovoltaic system efficiency, inclined global radiation, relative humidity and temperature which are strongly related to power can be selected from the 10 features of the photovoltaic data set for the next prediction experiment.

[0155] (2) Analysis of photovoltaic power prediction results

[0156] The data set used in the present application is provided by the Australian photovoltaic power station (DKASC), and the data used is the photovoltaic power data of the 69-Site_17-Sanyo power station in Australia, with a time resolution of 5 min, and a total of 11 feature variables including power. The data set is divided into a training set, a validation set and a test set, wherein the historical photovoltaic data set from 00:00 on January 1, 2016 to 23:55 on December 31, 2019 is the training set, the historical data of the power station from 00:00 on January 1, 2020 to 23:55 on June 30 is the validation set, and the photovoltaic power station data from 00:00 on July 1, 2020 to 23:55 on December 31 is the test set, in order to evaluate the effectiveness and generalization ability of the present method.

[0157] The prediction performance evaluation index is shown as follows.

[0158] 1) Mean absolute error

[0159]

[0160] MAE represents the average of the absolute error between the predicted value and the actual value. It is used to measure the size of the prediction error of the model.

[0161] 2) Root mean square error

[0162]

[0163] The square of the error between the predicted value and the actual value is averaged, and then the square root is taken. RMSE is more sensitive to large errors, so it can reflect the influence of large errors.

[0164] 3) Determination coefficient

[0165]

[0166] R 2The degree of explanation of the regression model to the data change. It reflects the goodness of fit of the model. R 2 The value is between 0 and 1, 1 represents perfect fitting, and 0 represents that the model has no prediction ability.

[0167] In the above formula: n is the number of samples; y i is the actual photovoltaic power value at time i; is the photovoltaic power prediction value at time i; is the average value of the actual photovoltaic power value.

[0168] As Figure 7 is the photovoltaic power prediction result graph from 00:00 on August 22, 2020 to 23:55 on August 28, 2020;

[0169] The evaluation indexes of the prediction results of the two models in the period from 00:00 on August 22, 2020 to 23:55 on August 28, 2020 are calculated, and MAE, RMSE and R 2 are used for evaluation, and the results are shown in the following table.

[0170] Table 6 Comparison of photovoltaic power prediction results from August 22, 2020 to August 28, 2020

[0171]

[0172] As Figure 7 and shown in Table 6, when the improved RFNN-QLSTM combined model is used for power prediction, the power prediction curve obtained can basically fit the true power curve without obvious error, while in the four days of August 22, 23, 25 and 28, 2020, the power prediction curve obtained by using the RFNN model for prediction has obvious error compared with the true power curve.

[0173] As Figure 8 shown is the photovoltaic power prediction result graph from August 25, 2020 to August 27, 2020.

[0174] It can be seen that when the improved RFNN-QLSTM combined model is used for power prediction, the power prediction curve obtained can basically fit the true power curve without obvious fluctuation, and the prediction result is significantly better than that of using the RFNN model for photovoltaic power prediction.

[0175] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same and similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the related parts can be referred to the method part.

[0176] The foregoing description of the disclosed embodiments enables one skilled in the art to make or use the application. Numerous modifications of those embodiments can be apparent to those skilled in the art, and the generic principles defined herein can be applied to other embodiments without the use of the innovation falling outside the spirit and scope of the application. Therefore, the application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A photovoltaic power prediction method based on VECM and an improved RFNN-QLSTM, characterized in that, include: Preprocessing of outliers and missing values ​​in the raw photovoltaic data; A vector error correction model (VECM) is constructed to capture the equilibrium regression relationship between photovoltaic data and meteorological factors. An error correction term is introduced into the vector autoregression model to identify the cointegration relationship and short-term bias between variables in the photovoltaic data. The meteorological factors are screened through the vector error correction model, and a new photovoltaic dataset is constructed based on the screening results. An improved recurrent fuzzy neural network-quantum long short-term memory network (RFNN-QLSTM) photovoltaic power short-term prediction model based on the vector error correction model (VECM) is constructed. The new photovoltaic dataset is used to train a short-term photovoltaic power prediction model to predict and output the photovoltaic output prediction results; By introducing an error correction term into the vector autoregression model, we obtain the vector error correction model, as shown in the following formula: Where ΔY t Γ is the difference of time series variables at time t, representing the short-term dynamic relationship. i It is the coefficient matrix of the lag period i, ΔY t Let p represent the time series under investigation, p be the lag length, Π be the influence matrix containing information about long-term relationships, and ε be the latency matrix. t It is an n×1 error vector used to estimate the likelihood ratio LR in order to determine ΔY in the system. t Cointegrated vector; The constructed vector error correction model was subjected to Johansen cointegration test and model stability test in sequence, and the cointegration results were analyzed to identify the variable relationships in the photovoltaic data.

2. The photovoltaic power prediction method based on VECM and improved RFNN-QLSTM according to claim 1, characterized in that, Methods for constructing quantum long short-term memory network models include: The quantum long short-term memory network model replaces the long short-term memory network units with variable quantum circuits to form quantum long short-term memory network units. The corresponding model architecture is defined as follows: f t =σ(VQC1(v t )), i t =σ(VQC2(v t )), the t =σ(VQC4(v t )), h t =o t ×tanh(c t ); Among them, f t Forgot Gate, i t For input gate, For candidate quantity, o t For output gate, c t h is the unit state gate. t It is the hidden state at the current time step, with the current input and the hidden state at the previous time step serving as the input v. t σ and tanh represent the sigmoid and hyperbolic tangent activation functions, respectively.

3. The photovoltaic power prediction method based on VECM and improved RFNN-QLSTM according to claim 2, characterized in that, Quantum long short-term memory network models also include: Input layer: Responsible for receiving and processing raw time series data; Output layer: Converts the processing results of the quantum long short-term memory network model into the final output for the next step of time series prediction.

4. The photovoltaic power prediction method based on VECM and improved RFNN-QLSTM according to claim 1, characterized in that, The improved recurrent fuzzy neural network model includes: the improved recurrent fuzzy neural network obtained by replacing the recurrent layer in the recurrent fuzzy neural network with a quantum long short-term memory network.

5. A system utilizing the improved RFNN-QLSTM photovoltaic power prediction method based on VECM as described in any one of claims 1-4, characterized in that, Includes the following modules: Data preprocessing module: used to preprocess the raw photovoltaic data for outliers and missing values; Vector error correction module: used to capture the equilibrium regression relationship between photovoltaic data and meteorological factors, as well as the relationship between meteorological factors. Based on the vector autoregression model, an error correction term is introduced to identify the cointegration relationship and short-term bias between variables in photovoltaic data. The vector error correction model is used to filter out relevant features in photovoltaic data and construct a new photovoltaic dataset. Photovoltaic power short-term prediction module: An improved recurrent fuzzy neural network model is constructed by replacing the recurrent layer in the recurrent fuzzy neural network with a quantum long short-term memory network. A new photovoltaic dataset constructed by filtering relevant features through a vector error correction model is used as input to the photovoltaic power short-term prediction model to output photovoltaic output prediction results.

6. A computer-readable storage medium, characterized in that, The readable storage medium stores computer instructions. When the computer reads the computer instructions in the storage medium, the computer executes the photovoltaic power prediction method based on VECM and improved RFNN-QLSTM as described in any one of claims 1-4.

Images