A method and apparatus for predicting optical power

Abstract

This invention proposes an optical power prediction method and apparatus, which can improve the reliability of optical power prediction. The optical power prediction method includes: collecting static and dynamic parameters that affect optical power; preprocessing the static and dynamic parameters to obtain static and dynamic feature parameters respectively; fusing the static and dynamic feature parameters; training a pre-constructed optical power prediction model to obtain a trained prediction model; and inputting the optical power data to be predicted into the trained prediction model to obtain the optical power prediction result.

Description

Technical Field

[0001] This invention relates to the field of new energy power, and in particular to a method and apparatus for predicting optical power. Background Technology

[0002] Current methods for predicting solar power can categorize the characteristic parameters affecting solar power into dynamic and static parameters based on the characteristics of real-time operational data from photovoltaic (PV) power plants. Dynamic parameters record the real-time operating status of the PV power plant and contain rich information on solar power degradation. Static parameters characterize the physical state of the PV power plant, and the data is relatively stable under normal operating conditions. The two types of characteristic parameters have different effects on solar power prediction, but existing methods ignore the differences between dynamic and static parameters, resulting in low reliability of current solar power prediction methods. Summary of the Invention

[0003] Based on this, the purpose of this invention is to provide an optical power prediction method and apparatus that can solve the problem of low reliability in optical power prediction mentioned above.

[0004] On the one hand, the optical power prediction method proposed in this invention includes:

[0005] Collect static and dynamic parameters that affect optical power;

[0006] The static parameters and the dynamic parameters are preprocessed to obtain static feature parameters and dynamic feature parameters, respectively.

[0007] The static feature parameters and the dynamic feature parameters are fused together;

[0008] The pre-built optical power prediction model is trained to obtain a trained prediction model.

[0009] The optical power data to be predicted is input into the trained prediction model to obtain the optical power prediction result.

[0010] On the other hand, the present invention proposes an optical power prediction device, comprising:

[0011] The acquisition module is used to acquire static and dynamic parameters that affect optical power;

[0012] The preprocessing module is used to preprocess the static parameters and the dynamic parameters respectively to obtain static feature parameters and dynamic feature parameters;

[0013] The fusion module is used to fuse the static feature parameters and the dynamic feature parameters;

[0014] The training module is used to train the pre-built optical power prediction model to obtain the trained prediction model;

[0015] The prediction module is used to input the optical power data to be predicted into the trained prediction model to obtain the optical power prediction result.

[0016] According to the above-mentioned invention, considering the different effects of dynamic parameters and static parameters on optical power prediction, dynamic parameters and static parameters are collected separately; after preprocessing, dynamic feature parameters and static feature parameters are fused, and then the fused parameters are trained to obtain a trained prediction model. By inputting the optical power parameters to be predicted into the model, a highly accurate optical power prediction result can be obtained. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the following detailed description to explain the invention, but should not be construed as limiting the invention. In the drawings,

[0018] Figure 1 This is a schematic diagram of the optical power prediction method proposed in this invention;

[0019] Figure 2 This is another schematic diagram of the optical power prediction method proposed in this invention;

[0020] Figure 3 This is a schematic diagram of the optical power prediction device proposed in this invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] The invention will now be described in detail with reference to the accompanying drawings and examples.

[0023] This invention proposes a method for optical power prediction, the flowchart of which is shown below. Figure 1 As shown, it includes:

[0024] S101. Collect static and dynamic parameters that affect optical power;

[0025] The dynamic parameters affecting solar power include: real-time irradiance, wind speed, wind direction, temperature, humidity, and air pressure at the photovoltaic power station site; the static parameters affecting solar power include: the number of inverters, inverter power, and whether the inverters are turned on.

[0026] S102. Preprocess the static parameters and the dynamic parameters respectively to obtain static feature parameters and dynamic feature parameters;

[0027] When preprocessing the dynamic parameters:

[0028] (1) The dynamic parameters can be normalized using the minimax normalization method; wherein, a mathematical expression for minimax normalization is:

[0029]

[0030] In the formula, x′ is the normalized dynamic parameter, x is the original dynamic parameter, and x' is the normalized dynamic parameter. min and x max These are the maximum and minimum values ​​in the original dynamic parameters.

[0031] (2) Then, the normalized dynamic parameters are divided into an input time window sequence using the sliding window method; where, a mathematical expression for the input time window at time t is:

[0032]

[0033] In the formula, X(t) is the input time window at time t. Let t be the value of the first dynamic parameter at time t, n be the size of the time window, and m be the number of dynamic parameters.

[0034] (3) Perform noise reduction and smoothing processing on each input time window; among which, a mathematical expression for the noise reduction method based on statistics is:

[0035]

[0036] In the formula, x i,j,k μ represents the k-th dynamic parameter value in the j-th column of the i-th window; i,j σ represents the mean of the dynamic parameter values ​​in the j-th column of the i-th window; i,j This represents the standard deviation of the dynamic parameter values ​​in the j-th column of the i-th window.

[0037] (4) Add an optical power label to each time window. The label for the input time window at time t is:

[0038]

[0039] In the formula, IP t i Input the optical power label of the time window for the i-th photovoltaic power station at time t. Let be the actual power value of the i-th photovoltaic power station at time t.

[0040] In one embodiment, when preprocessing the static parameters, constant values ​​are used as representative values ​​of the corresponding static parameters, and the variable values ​​are averaged and then used as representative values ​​of the corresponding static parameters.

[0041] S103. Merge the static feature parameters and the dynamic feature parameters;

[0042] S104. Train the pre-built optical power prediction model to obtain the trained prediction model;

[0043] S105. Input the optical power data to be predicted into the trained prediction model to obtain the optical power prediction result.

[0044] According to the above-mentioned invention, considering the different effects of dynamic parameters and static parameters on optical power prediction, dynamic parameters and static parameters are collected separately; after preprocessing, dynamic feature parameters and static feature parameters are fused, and then the fused parameters are trained to obtain a trained prediction model. By inputting the optical power parameters to be predicted into the model, a highly accurate optical power prediction result can be obtained.

[0045] In one embodiment, another flowchart of the optical power prediction method proposed in this invention is shown below. Figure 2 :

[0046] 1. First, acquire data, that is, collect static and dynamic parameters that affect optical power;

[0047] 2. Perform data preprocessing to obtain the training dataset, including: static feature parameters and dynamic feature parameters; see S102 for details;

[0048] 3. Feature fusion based on training dataset; wherein, when fusing the static feature parameters and the dynamic feature parameters, a Multiscale-ConvBiLSTM-Attention network can be constructed based on a convolutional bidirectional long short-term memory network and an attention mechanism.

[0049] The calculation process for the convolution operation is as follows:

[0050] Y(t)=σ(W*X(t)+b)

[0051] In the above formula: Y(t) is the output of the convolutional layer, σ is the sigmoid activation function, W represents the weight matrix, X(t) represents the input time window at time t, and b represents the bias.

[0052] The dynamic feature parameters are processed to extract spatiotemporal features; firstly, ConvBiLSTM is used to mine the spatiotemporal features contained in the dynamic parameters; the calculation process of BiLSTM is as follows:

[0053]

[0054]

[0055] In the above formula: and These represent the input gate, forget gate, new cell state candidates, output gate, hidden output, and cell state at time t in the positive time direction, respectively. and These represent the input gate, forget gate, new cell state candidates, output gate, hidden output, and cell state at time t in the negative time direction, respectively; h final (t) represents the hidden output of BiLSTM at time t; σ represents the sigmoid function, tanh represents the tanh function; ⊙ represents the Hadamard product; and Represents the weight matrix; and This represents the bias term.

[0056] Weights are assigned to spatiotemporal features; these weights can be assigned using Attention. The calculation process for Attention is as follows:

[0057]

[0058] In the above formula: Q, K, and V are linear mappings of the input information; T is the matrix transpose operation; d K Let K be the dimension of the key matrix.

[0059] A multi-scale learning strategy is employed to obtain multi-scale weighted spatiotemporal features of dynamic feature parameters.

[0060] Clustering is performed on the static feature parameters to determine cluster centers. The category probability of each photovoltaic power station is calculated based on its Euclidean distance from the cluster center. The category features of the static feature parameters are then obtained based on these category probabilities. The mathematical expression for the Euclidean distance is:

[0061]

[0062] In the above formula: This represents the Euclidean distance between the i-th photovoltaic power station and the j-th cluster center. This represents the first static characteristic parameter value of the i-th photovoltaic power station. The value of the first static feature parameter of the j-th cluster center, where y is the number of static feature parameters.

[0063] The mathematical expression for category probability is:

[0064]

[0065] In the above formula: This represents the probability that the i-th photovoltaic power station belongs to the j-th class. K represents the distance between the i-th photovoltaic power station and the j-th cluster center, and K is the number of cluster centers.

[0066] The mathematical expression for the category feature is:

[0067]

[0068] In the above formula: fea i This represents the category characteristics of the i-th photovoltaic power station.

[0069] The multi-scale weighted spatiotemporal features of dynamic feature parameters are fused together, and then fused together with the category features of static feature parameters.

[0070] 4. Train the pre-built optical power prediction model to obtain a trained prediction model; the trained prediction model can be a fully connected neural network.

[0071] 5. Input the optical power data to be predicted into the trained prediction model to obtain the optical power prediction result.

[0072] This invention uses a dynamic and static feature fusion method for optical power prediction, which can predict the short-term and ultra-short-term output power of power plants.

[0073] In another aspect, the present invention provides an optical power prediction device, comprising:

[0074] The acquisition module is used to collect static and dynamic parameters that affect the photovoltaic power. The static parameters include the number of inverters, inverter power, and whether the inverters are turned on. The dynamic parameters are the real-time irradiance, wind speed, wind direction, temperature, humidity, and air pressure at the photovoltaic power station.

[0075] The preprocessing module is used to preprocess the static parameters and dynamic parameters respectively to obtain static feature parameters and dynamic feature parameters. Specifically, when preprocessing the dynamic parameters, the preprocessing module can use the max-min normalization method to normalize the dynamic parameters, and then use a sliding window method to divide the normalized dynamic parameters into an input time window sequence. Noise reduction and smoothing processing is performed on each input time window; and an optical power label is added to each time window. When preprocessing the static parameters, the preprocessing module uses constant values ​​as representative values ​​for the corresponding static parameters, and averages the changing values ​​as representative values ​​for the corresponding static parameters.

[0076] A fusion module is used to fuse the static feature parameters and the dynamic feature parameters; wherein, when fusing the static feature parameters and the dynamic feature parameters, the fusion module processes the dynamic feature parameters to extract spatiotemporal features and assigns weights to the spatiotemporal features; a multi-scale learning strategy is used to obtain multi-scale weighted spatiotemporal features of the dynamic feature parameters;

[0077] Clustering is performed on the static feature parameters to determine cluster centers; the category probability of each photovoltaic power station is calculated based on the Euclidean distance of each photovoltaic power station from the cluster center, and the category features of the static feature parameters are obtained based on the category probability of each photovoltaic power station; the multi-scale weighted spatiotemporal features of the dynamic feature parameters are fused, and then fused with the category features of the static feature parameters.

[0078] The training module is used to train the pre-built optical power prediction model to obtain the trained prediction model;

[0079] The prediction module is used to input the optical power data to be predicted into the trained prediction model to obtain the optical power prediction result.

[0080] According to the above-mentioned invention, considering the different effects of dynamic parameters and static parameters on optical power prediction, dynamic parameters and static parameters are collected separately; after preprocessing, dynamic feature parameters and static feature parameters are fused, and then the fused parameters are trained to obtain a trained prediction model. By inputting the optical power parameters to be predicted into the model, a highly accurate optical power prediction result can be obtained.

[0081] Any combination of various embodiments of the present invention, provided it does not violate the inventive concept of the present invention, shall be regarded as the content disclosed by the present invention; within the scope of the technical concept of the present invention, any simple modifications to the technical solution and any combination of different embodiments that do not violate the inventive concept of the present invention shall be within the protection scope of the present invention.

Claims

1. A method for predicting optical power, characterized in that, include: Collect static and dynamic parameters that affect optical power; The static parameters include: number of inverters, inverter power, and whether the inverters are turned on; the dynamic parameters include real-time irradiance, wind speed, wind direction, temperature, humidity, and air pressure at the photovoltaic power station site. The static parameters and the dynamic parameters are preprocessed to obtain static feature parameters and dynamic feature parameters, respectively. The static and dynamic feature parameters are fused. During fusion, the dynamic feature parameters are processed to extract spatiotemporal features, and weights are assigned to these features. A multi-scale learning strategy is employed to obtain multi-scale weighted spatiotemporal features of the dynamic feature parameters. The static feature parameters are clustered to determine cluster centers. The class probability of each photovoltaic power station is calculated based on its Euclidean distance from the cluster center, and the class features of the static feature parameters are obtained based on these probabilities. The multi-scale weighted spatiotemporal features of the dynamic feature parameters are then fused with the class features of the static feature parameters. The pre-built optical power prediction model is trained to obtain a trained prediction model. The optical power data to be predicted is input into the trained prediction model to obtain the optical power prediction result.

2. The optical power prediction method according to claim 1, characterized in that, When preprocessing the dynamic parameters, the dynamic parameters are normalized; the normalized dynamic parameters are divided into a sequence of input time windows using a sliding window method; noise reduction and smoothing are performed on each input time window; and an optical power label is added to each time window.

3. The optical power prediction method according to claim 1, characterized in that, When preprocessing the static parameters, constant values ​​are used as representative values ​​of the corresponding static parameters, and the variable values ​​are averaged and then used as representative values ​​of the corresponding static parameters.

4. An optical power prediction device, characterized in that, include: The acquisition module is used to acquire static and dynamic parameters that affect optical power; The static parameters include: number of inverters, inverter power, and whether the inverters are turned on; the dynamic parameters include real-time irradiance, wind speed, wind direction, temperature, humidity, and air pressure at the photovoltaic power station site. The preprocessing module is used to preprocess the static parameters and the dynamic parameters respectively to obtain static feature parameters and dynamic feature parameters; A fusion module is used to fuse the static feature parameters and the dynamic feature parameters. When fusing the static and dynamic feature parameters, the fusion module processes the dynamic feature parameters and extracts spatiotemporal features, assigning weights to the spatiotemporal features. A multi-scale learning strategy is used to obtain multi-scale weighted spatiotemporal features of the dynamic feature parameters. The static feature parameters are clustered to determine cluster centers. The class probability of each photovoltaic power station is calculated based on its Euclidean distance from the cluster center, and the class features of the static feature parameters are obtained based on the class probability of each photovoltaic power station. The multi-scale weighted spatiotemporal features of the dynamic feature parameters are then fused, and finally fused with the class features of the static feature parameters. The training module is used to train the pre-built optical power prediction model to obtain the trained prediction model; The prediction module is used to input the optical power data to be predicted into the trained prediction model to obtain the optical power prediction result.

5. The optical power prediction device according to claim 4, characterized in that, When the preprocessing module preprocesses the dynamic parameters, it normalizes the dynamic parameters; divides the normalized dynamic parameters into a sequence of input time windows using a sliding window method; performs noise reduction and smoothing processing on each input time window; and adds an optical power label to each time window.

6. The optical power prediction device according to claim 4, characterized in that, When the preprocessing module preprocesses the static parameters, it uses constant values ​​as representative values ​​of the corresponding static parameters and averages the changing values ​​as representative values ​​of the corresponding static parameters.

Images