New energy output adaptive prediction method based on decomposition-reconstruction-integration

Abstract

The invention provides a new energy output adaptive prediction method based on decomposition-reconstruction-integration, and the method comprises the steps: firstly carrying out the preprocessing of data, and carrying out the decomposition and denoising of a wind power generation power signal through employing an ICEEMDAN technology; decomposing an original complex signal into a series of intrinsic mode functions IMFs; calculating the sample entropy of each IMF, and combining the IMFs with similar complexity to form a new sub-sequence; and finally, according to different subsequences and model characteristics, respectively predicting each subsequence by adopting a CNN model and a BiLSTM model. According to the method, on one hand, the adaptivity of the new energy prediction model to time sequence decomposition with different characteristics is improved, and on the other hand, the prediction precision can be further improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system data analysis and new energy prediction, and specifically relates to a new energy output adaptive prediction method based on decomposition-reconstruction-integration. Background Art

[0002] Renewable energy generation represented by wind power and photovoltaic power generation has received widespread attention, and its installed capacity has increased year by year. However, affected by various factors such as environmental factors and geographical location, its output is highly unstable and volatile, and the uncertainty and irregularity of output at future moments have a great impact on the operation of the power grid. Therefore, accurate and reliable new energy forecasts have received attention in recent years. Commonly used methods include physical methods, statistical methods, and machine learning methods. The physical method is mainly based on numerical weather forecasts and the geographical location characteristics of wind farms, and does not rely on massive historical climate data, but its prediction accuracy is deeply affected by weather forecast accuracy and actual conditions; in contrast, the statistical method is simpler in operation and has lower data requirements, but it performs poorly in dealing with abnormal situations; although machine learning technology requires massive historical observation data support, it can achieve relatively accurate prediction results, the model adjustment is flexible, and it has the potential to enhance the accuracy of predictions by integrating multiple prediction technologies. However, existing machine learning technology still faces two more prominent problems:

[0003] 1. Due to the instability and fluctuation of output, it is difficult to directly build a model for the original sequence. There are challenges in how to more accurately and deeply analyze the characteristics of the original sequence and adaptively process the original sequence according to its characteristics to improve the prediction progress.

[0004] 2. Different sequences have different characteristics. Using the same neural network to predict all subsequences cannot achieve the optimal prediction effect and is likely to reduce the adaptability of the model. Summary of the invention

[0005] In view of the complex problems existing in the field of new energy prediction mentioned above, the present invention proposes a new energy output adaptive prediction method based on decomposition-reconstruction-integration, which provides a feasible solution for improving the accuracy of new energy output prediction results.

[0006] The technical solution adopted by the present invention to solve the technical problem is:

[0007] The present invention provides a new energy output adaptive prediction method based on decomposition-reconstruction-integration, comprising the following steps:

[0008] Step 1. Data preprocessing;

[0009] Step 2. Apply ICEEMDAN technology to decompose and denoise the wind power signal; decompose the original complex signal into a series of intrinsic mode functions (IMFs);

[0010] Step 3. Based on the characteristics of sample entropy, the decomposed signal components are reintegrated. By calculating the sample entropy of each IMF, IMFs with similar complexity are combined to form a new subsequence (reconstructed sequence) to ensure its complexity consistency, and then the recombined subsequence is modeled and predicted.

[0011] Step 4. According to different subsequences and model characteristics, CNN and BiLSTM models are used to predict each subsequence respectively, and these prediction results are integrated to form the final prediction output, and the effectiveness of the model is verified by error evaluation indicators.

[0012] Furthermore, the data preprocessing in step 1 is to perform normalization preprocessing on the collected data.

[0013] Furthermore, step 2 includes the following sub-steps:

[0014] Step 2.1 Use the EMD algorithm to perform local mean decomposition on the signal x(t) to obtain the sequence x i (t) = x(t) + ε0E1(ξ i (t)), and then obtain the first residual: r1= <M(x i (t))>

[0015] The operator M(·) is used to calculate the local average value of the processed signal, E k (·) is the kth intrinsic mode function obtained by the empirical mode decomposition method, ξ i is a Gaussian white noise with zero mean and unit variance. The kth modal component generated by the improved composite empirical mode decomposition noise-assisted method ICEEM DAN is denoted as

[0016] Step 2.2: Based on the calculation in the first stage (k=1), the first modal component is obtained:

[0017] Step 2.3 estimates the second residual as the realization r1(t)+ε1E2[ξ i (t)] and define the second modal component as follows:

[0018]

[0019] Step 2.4 For the subsequent stage k=3,…,K, calculate the kth residual:

[0020] r k (t)= <M(r K-1 (t))+ω k-1 E K (ξ i (t)))> (2-2)

[0021] The kth modal component is calculated:

[0022]

[0023] Step 2.5 Repeat step 2.4 to get the next k until all modal components are extracted.

[0024] Furthermore, step 3 includes the following sub-steps:

[0025] Step 3.1 Calculate the sample entropy of each IMF

[0026] The sample entropy is represented by SE(n,ɑ), which is described as follows:

[0027]

[0028] Where M represents the total length of the sequence; n represents the dimension considered; ɑ is the similarity tolerance; R n+1 (ɑ) and R n (ɑ) represents the probability of two sequences matching n or n+1 points under a given threshold ɑ. When M is assigned a finite value, the estimated sample entropy is calculated according to formula (3-1), expressed as SE(n,ɑ,M):

[0029]

[0030] Step 3.2 combines IMFs of similar complexity to form new subsequences (reconstructed sequences), which establishes conditions for modeling and predicting the reconstructed subsequences in the next step.

[0031] The advantages and positive effects of the present invention are:

[0032] The present invention provides a method for adaptive prediction of new energy output based on decomposition-reconstruction-integration, which can achieve the following beneficial effects: by constructing a "decomposition-reconstruction" strategy, utilizing an improved adaptive noise complete set empirical mode decomposition algorithm and a sequence reconstruction strategy based on sample entropy, the frequency characteristics of the original time series can be accurately mined, and the original time series can be decomposed into subsequences with different frequency characteristics. The subsequences with similar complexity can be reconstructed again, and different neural network models can be used to carry out predictions. On the one hand, the adaptability of the new energy prediction model to the decomposition of time series with different characteristics can be improved, and on the other hand, the prediction accuracy can be further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0034] The present invention is further described in detail below through specific examples. The following examples are only illustrative and not restrictive, and the protection scope of the present invention cannot be limited thereto.

[0035] Figure 1 The overall technical framework diagram of the present invention is shown in the figure. As shown in the figure, the method includes the following steps:

[0036] Step 1. Perform normalization preprocessing on the collected new energy data.

[0037] Step 2. Apply ICEEMDAN technology to decompose and denoise the wind power signal; decompose the original complex signal into a series of intrinsic mode functions (IMFs).

[0038] In the field of time series analysis, empirical mode decomposition (EMD) and its improved algorithms, such as complete ensemble empirical mode decomposition (CEEMDAN) and improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN), have become important tools for processing nonlinear and non-stationary time series. These algorithms provide a new perspective for signal analysis and processing by decomposing complex signals into a series of intrinsic mode functions (IMFs). Among them, the ICEEMDAN algorithm, as a further improvement of EMD and CEEMDAN, has shown its effectiveness in many fields. Experimental results show that ICEEMDAN not only reduces the residual noise in EEMD, but also solves the pseudo-mode problem in CEEMDAN, effectively enhancing the denoising effect. The main steps are:

[0039] Step 2.1 Use the EMD algorithm to perform local mean decomposition on the signal x(t) to obtain the sequence x i (t) = x(t) + ε0E1(ξ i (t)), and then obtain the first residual: r1= <M(x i (t))>

[0040] The operator M(·) is used to calculate the local average value of the processed signal, E k (·) is the kth intrinsic mode function obtained by the empirical mode decomposition method, ξ i is a Gaussian white noise with zero mean and unit variance. The kth modal component generated by the improved composite empirical mode decomposition noise-assisted method ICEEM DAN is denoted as

[0041] Step 2.2 According to the calculation of the first stage (k=1), the first modal component can be obtained:

[0042] Step 2.3 estimates the second residual as the realization r1(t)+ε1E2[ξ i (t)] and define the second modal component as follows:

[0043]

[0044] Step 2.4 For the subsequent stage k=3,…,K, calculate the kth residual:

[0045] r k (t)= <M(r K-1 (t))+ω k-1 E K (ξ i (t)))> (2-2)

[0046] The kth modal component is calculated:

[0047]

[0048] Step 2.5 Repeat step 2.4 to get the next k until all modal components are extracted

[0049] Step 3. Based on the characteristics of sample entropy, the decomposed signal components are reintegrated. By calculating the sample entropy of each IMF, IMFs with similar complexity are combined to form a new subsequence (reconstructed sequence) to ensure its complexity consistency, and then the recombined subsequence is modeled and predicted.

[0050] Considering that there are a large number of IMFs obtained by decomposition, if each IMF is modeled and predicted separately, the complexity of the calculation will increase significantly, and more models need to be designed and trained to meet this challenge. And if there is a certain error in the prediction of each IMF, these small errors may accumulate during the superposition process, resulting in an increase in the error of the final prediction result. Therefore, in order to solve these problems, the present invention proposes a sequence reconstruction strategy based on sample entropy technology.

[0051] Sample entropy is a statistic based on information theory that measures the complexity or irregularity of time series data. The basic idea of ​​sample entropy is that, given a time series, its regularity is evaluated by calculating the number of short-term similar sequence pairs in the series. If the time series exhibits a high degree of regularity or periodicity, the value of sample entropy will be relatively low because there are more similar sequence pairs. On the contrary, if the time series is random or complex, the value of sample entropy will be relatively high because there are fewer similar sequence pairs. The main steps are:

[0052] Step 3.1 Calculate the sample entropy of each IMF

[0053] The sample entropy is represented by SE(n,ɑ), which is described as follows:

[0054]

[0055] Where M represents the total length of the sequence; n represents the dimension considered; ɑ is the similarity tolerance; R n+1 (ɑ) and R n (ɑ) represents the probability of two sequences matching n or n+1 points under a given threshold ɑ. When M is assigned a finite value, the estimated sample entropy is calculated according to formula (3-1), expressed as SE(n,ɑ,M):

[0056]

[0057] Step 3.2 combines IMFs of similar complexity to form new subsequences (reconstructed sequences), which establishes conditions for modeling and predicting the reconstructed subsequences in the next step.

[0058] Step 4. According to different subsequences and model characteristics, CNN and BiLSTM models are used to predict each subsequence respectively, and these prediction results are integrated to form the final prediction output, and the effectiveness of the model is verified by error evaluation indicators.

[0059] The above description is only a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several modifications and improvements without departing from the inventive concept, which all belong to the protection scope of the present invention.

Claims

1. A new energy output adaptive prediction method based on decomposition-reconstruction-integration, characterized in that: The following steps are involved: Step 1. Data preprocessing; Step 2. Apply ICEEMDAN technology to decompose and denoise the wind power signal; decompose the original complex signal into a series of intrinsic mode functions IMFs; Step 3. Calculate the sample entropy of each IMF and combine IMFs with similar complexity to form a new subsequence; Step 4. According to different subsequences and model characteristics, CNN and BiLSTM models are used to predict each subsequence respectively, and these prediction results are integrated to form the final prediction output, and the effectiveness of the model is verified by error evaluation indicators.

2. The adaptive prediction method for renewable energy output based on decomposition-reconstruction-integration according to claim 1 is characterized in that: The data preprocessing described in step 1 is to perform normalization preprocessing on the collected data.

3. The adaptive prediction method for renewable energy output based on decomposition-reconstruction-integration according to claim 1 is characterized in that: Step 2 includes the following sub-steps: Step 2.1 Use the EMD algorithm to perform local mean decomposition on the signal x(t) to obtain the sequence x i (t) = x(t) + ε0E1(ξ i (t)), and then obtain the first residual: r1= <M(x i (t))> The operator M(·) is used to calculate the local average value of the processed signal, E k (·) is the kth intrinsic mode function obtained by the empirical mode decomposition method, ξ i is a Gaussian white noise with zero mean and unit variance. The kth modal component generated by the improved composite empirical mode decomposition noise-assisted method ICEEM DAN is denoted as Step 2.2: Based on the calculation in the first stage (k=1), the first modal component is obtained: Step 2.3 estimates the second residual as the realization r1(t)+ε1E2[ξ i (t)] and define the second modal component as follows: Step 2.4 For the subsequent stages k=3,…,K, calculate the kth residual: r k (t)= <M(r K-1 (t))+ω k-1 E K (x) i (t)))> (2-2) The kth modal component is calculated: Step 2.5 Repeat step 2.4 to get the next k until all modal components are extracted.

4. The adaptive prediction method for renewable energy output based on decomposition-reconstruction-integration according to claim 1 is characterized in that: Step 3 includes the following sub-steps: Step 3.1 Calculate the sample entropy of each IMF The sample entropy is represented by SE(n,ɑ), which is described as follows: Where M represents the total length of the sequence; n represents the dimension considered; ɑ is the similarity tolerance; R n+1 (ɑ) and R n (ɑ) represents the probability of two sequences matching n or n+1 points under a given threshold ɑ. When M is assigned a finite value, the estimated sample entropy is calculated according to formula (3-1), expressed as SE(n,ɑ,M): Step 3.2 combines IMFs of similar complexity to form new subsequences (reconstructed sequences), which establishes conditions for modeling and predicting the reconstructed subsequences in the next step.

Images