A method for matching wind, solar and energy storage combined installed capacity based on power forecasting

Abstract

The present invention discloses a method for matching wind, solar and energy storage combined installed capacity based on power prediction, which relates to the technical field of new energy power systems, and includes the following steps: by calculating the correlation coefficient between features, using scaling factors to gradually process wind and solar multi-dimensional feature data; segmenting and encoding different feature data sets to obtain embedded representation data; constructing a power prediction model; performing wind and solar power prediction; constructing a system power fluctuation model and an energy storage system model; constructing a wind, solar and energy storage combined installed capacity matching model, and solving to obtain a wind, solar and energy storage combined installed capacity matching scheme. This application realizes the accurate prediction of long-sequence wind and solar power by combining the correlation coefficient between features, scaling factors, segmentation, encoding, and a learnable intermediate layer; constructing and solving a wind, solar and energy storage combined installed capacity matching model based on the wind and solar power prediction results, thereby improving the stability of the wind, solar and energy storage combined output and the economy of the installed capacity matching.

Description

Technical Field

[0001] The present application relates to the technical field of new energy power systems, and in particular to a method for allocating wind, solar and storage combined installed capacity based on power prediction. Background Art

[0002] Wind power generation and photovoltaic power generation are important forms of energy in the new energy power system. Wind and solar power generation are highly random and easily affected by weather conditions, which brings challenges to the stability and efficiency of wind and solar power generation systems. The configuration of energy storage systems can smooth out the fluctuations of wind and solar power generation, but the capacity of the energy storage system needs to match the installed capacity of wind power and photovoltaics. Therefore, it is necessary to study the ratio of wind, solar and storage combined installed capacity.

[0003] In the prior art, Chinese patent CN103326388A discloses a microgrid energy storage system and capacity configuration method based on power prediction, which uses a quick-start power generation device and an energy storage device as a backup power source for the microgrid system to balance the power fluctuations of the wind power generation system and the photovoltaic power generation system in the microgrid system; when the microgrid energy storage system is used alone to supply power, the longest power generation time of multiple quick-start power generation devices is measured, and based on the power prediction technology of the wind power generation system and the photovoltaic power generation system, the power fluctuations of the wind power generation system and the photovoltaic power generation system during the longest power generation time are calculated, and the minimum capacity configuration of the energy storage device in the microgrid system is obtained.

[0004] However, the above-mentioned existing technologies are limited by the learning mechanism and have poor prediction effects on long-sequence wind and solar power, resulting in poor stability of the combined output of wind, solar and storage and poor economy of installed capacity ratio. Summary of the invention

[0005] The present application provides a method for matching the installed capacity of a combination of wind, solar and storage based on power prediction, in order to solve the problems that the existing wind-solar power system technology has poor prediction effect on long-sequence wind and solar power, and the stability of the combined output of wind, solar and storage and the economy of the installed capacity matching are poor.

[0006] On the one hand, the present application provides a method for allocating wind, solar and storage combined installed capacity based on power prediction, comprising the following steps:

[0007] Step S1, reading the multi-dimensional feature data of the wind and solar power, calculating the correlation coefficient between the features, using the scaling factor to gradually process the multi-dimensional feature data of the wind and solar power, and dividing them to obtain different feature data sets.

[0008] Step S2, segmenting and encoding the different feature data sets to obtain embedded representation data.

[0009] Step S3, using a learnable intermediate layer and an Encoder-Decoder loop structure to build a power prediction model, wherein the learnable intermediate layer is used to realize the interaction between embedded representation data of different time series, and the Encoder-Decoder loop structure is used to perform information distillation and result output.

[0010] Step S4, setting a training strategy, training the power prediction model, using an optimization algorithm to adjust parameters during the training process, using the trained power prediction model to perform wind and solar power prediction, and obtaining a wind and solar power prediction result.

[0011] Step S5, constructing a system power fluctuation model based on the wind and solar power prediction results and system stability requirements, and constructing an energy storage system model.

[0012] Step S6, considering the economic cost of the wind-solar-storage combined system, constructing a wind-solar-storage combined installed capacity ratio model that considers the dual objectives of power fluctuation and economy based on the system power fluctuation model and the energy storage system model, and solving the wind-solar-storage combined installed capacity ratio scheme.

[0013] In a possible implementation, in step S1, the wind-solar multi-dimensional feature data includes: historical wind power and photovoltaic output power and environmental indicator data.

[0014] Step S1 includes:

[0015] The wind and solar multi-dimensional feature data is read, the correlation coefficient between the features is determined, and the wind and solar multi-dimensional feature data is preprocessed.

[0016] A scaling factor is introduced to perform Norm processing on the scenery multi-dimensional feature data, and the processed scenery multi-dimensional feature data are divided to obtain different feature data sets.

[0017] In a possible implementation, step S2 includes:

[0018] A data set segmentation function is constructed, and the data set segmentation function is used to segment the different feature data sets into segments to obtain data segments.

[0019] The data segment is encoded to obtain the embedded representation data.

[0020] In a possible implementation, in step S3, before using the learnable intermediate layer and the Encode-Decoder loop structure to build the power prediction model, Add and Norm operations are first performed on the embedded representation data of different time series.

[0021] In a possible implementation, in step S4, an evaluation index is also used to perform performance evaluation on the trained power prediction model.

[0022] In a possible implementation, step S5 includes:

[0023] The output variation characteristics of wind power and photovoltaic power are analyzed, a system power fluctuation model is constructed based on the wind and photovoltaic power prediction results and system stability requirements, and the optimal wind and photovoltaic capacity ratio that minimizes the system power fluctuation is calculated.

[0024] An energy storage system model is constructed, and a mapping framework of power fluctuation quantification index and energy storage capacity ratio function is determined according to the energy storage system model.

[0025] In a possible implementation, in step S6, the multi-objective functions and constraints of the wind, solar, and storage combined installed capacity ratio model are determined, and the wind, solar, and storage combined installed capacity ratio model is solved using a multi-objective mantis search algorithm to obtain the wind, solar, and storage combined installed capacity ratio scheme.

[0026] A wind-solar-storage combined installed capacity matching method based on power prediction in this application has the following advantages:

[0027] By combining the correlation coefficient between features, scaling factors, fragment segmentation, encoding, and learnable intermediate layers, the dependency between global data can be captured from the global attention search, and accurate prediction of long-sequence wind and solar power can be achieved; based on the wind and solar power prediction results, a wind, solar, and storage combined installed capacity ratio model that considers power fluctuations and economic goals is constructed and solved, which improves the stability of the wind, solar, and storage combined output and the economy of the installed capacity ratio. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0029] Figure 1 A schematic flow chart of a method for allocating wind, solar and storage combined installed capacity based on power prediction provided in an embodiment of the present application;

[0030] Figure 2 A logical schematic diagram of a wind-solar-storage combined installed capacity matching method based on power prediction provided in an embodiment of the present application;

[0031] Figure 3 A comparison chart of the long-sequence predicted value and the actual value of wind power provided in the embodiment of the present application;

[0032] Figure 4A comparison chart of the photovoltaic power long sequence prediction value and the actual value provided in the embodiment of the present application;

[0033] Figure 5 A graph showing the relationship between wind power capacity, photovoltaic capacity and power shortage provided in the embodiment of the present application;

[0034] Figure 6 This is a graph of the net output power of energy storage provided in an embodiment of the present application. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0036] like Figure 1 and Figure 2 As shown, the embodiment of the present application provides a method for allocating wind, solar and storage combined installed capacity based on power prediction, comprising the following steps:

[0037] Step S1, reading the multi-dimensional feature data of the wind and solar power, calculating the correlation coefficient between the features, using the scaling factor to gradually process the multi-dimensional feature data of the wind and solar power, and dividing them to obtain different feature data sets.

[0038] Step S2, segmenting and encoding the different feature data sets to obtain embedded representation data.

[0039] Step S3, using a learnable intermediate layer and an Encoder-Decoder loop structure to build a power prediction model, wherein the learnable intermediate layer is used to realize the interaction between embedded representation data of different time series, and the Encoder-Decoder loop structure is used to perform information distillation and result output.

[0040] Step S4, setting a training strategy, training the power prediction model, using an optimization algorithm to adjust parameters during the training process, using the trained power prediction model to perform wind and solar power prediction, and obtaining a wind and solar power prediction result.

[0041] Step S5, constructing a system power fluctuation model based on the wind and solar power prediction results and system stability requirements, and constructing an energy storage system model.

[0042] Step S6, considering the economic cost of the wind-solar-storage combined system, constructing a wind-solar-storage combined installed capacity ratio model that considers the dual objectives of power fluctuation and economy based on the system power fluctuation model and the energy storage system model, and solving the wind-solar-storage combined installed capacity ratio scheme.

[0043] Exemplarily, in step S1, the wind-solar multi-dimensional feature data includes: historical wind power and photovoltaic output power and environmental indicator data.

[0044] Step S1 includes:

[0045] The wind and solar multi-dimensional feature data is read, the correlation coefficient between the features is determined, and the wind and solar multi-dimensional feature data is preprocessed.

[0046] A scaling factor is introduced to perform Norm processing on the scenery multi-dimensional feature data, and the processed scenery multi-dimensional feature data are divided to obtain different feature data sets.

[0047] Specifically, in this embodiment, the environmental indicator data includes total radiation, ambient temperature, relative humidity and atmospheric pressure. In other possible embodiments, other environmental indicator data may also be included.

[0048] In this embodiment, the preprocessing includes screening, standardization and correlation coefficient calculation. As shown below:

[0049]

[0050] in, , Respectively represent the multi-dimensional feature data of scenery in different feature dimensions, express , The covariance between , Respectively , The mean of , Respectively , The standard deviation of , Respectively represent indivual , The observed value of Represents the number of observations.

[0051] In this embodiment, the scaling factor is introduced ,pass The multi-dimensional characteristic data of scenery is processed by Norm, where: Represents the processed scenery multi-dimensional feature data, Represents the standard deviation of the multi-dimensional feature data of the wind and solar power. The processed multi-dimensional feature data of the wind and solar power are divided into different feature data sets in a ratio of 8:2. Assume that each sample in the input data set contains D total feature dimensions, and thus set the length of the input data set to , expressed as: ; Set the length of the output data set to , expressed as: .

[0052] Exemplarily, step S2 includes:

[0053] A data set segmentation function is constructed, and the data set segmentation function is used to segment the different feature data sets into segments to obtain data segments.

[0054] The data segment is encoded to obtain the embedded representation data.

[0055] Specifically, in this embodiment, the data set segmentation function is constructed as shown in the following formula:

[0056]

[0057] in, Indicates feature dimensions, D represents the total number of feature dimensions, Indicates that the input data set ranges from 1 to The data sequence, Indicates In the paragraph Data fragments on feature dimensions, represents the length of the input data set, Indicates the length of the data segment, Indicates that the input data is concentrated at a specific time step and The original data on the feature dimension.

[0058] By solving the data set segmentation function, different feature data sets are divided into data fragments, Indicates the total number of fragments. Encode the data fragment to get the embedded representation data: .

[0059] in, Indicates In the paragraph The embedding vector on the feature dimension, Indicates In the paragraph The embedding representation on the feature dimension is Indicates In the paragraph The position information embedding vector on the feature dimension is Indicates the total number of fragments after embedding , total feature dimensions and the embedding vector dimension The interaction matrix dimension on .

[0060] Exemplarily, in step S3, before using the learnable intermediate layer and the Encode-Decoder loop structure to build the power prediction model, Add and Norm operations are performed on the embedded representation data of different time series.

[0061] Specifically, in this embodiment, in step S3, Add and Norm operations are performed on the embedded representation data of different time sequences, and the following mechanism is set:

[0062]

[0063] in, Indicates the embedded data after multi-head self-attention mechanism (MSA) and layer normalization (LayerNorm) in the first The output representation on the feature dimension is Indicates that all feasibility The input feature vector has feature dimensions, represents the output features after multi-layer perceptron (MLP) and layer normalization, Represents the embedded data after the first step of multi-head self-attention mechanism and layer normalization.

[0064] In this embodiment, in step S3, a learnable intermediate layer is introduced Realize the interaction between embedding representation data of different time series, which will have The interaction is as follows:

[0065]

[0066] in, Indicates the number of channels, Indicates The output of the multi-head self-attention mechanism at time step, Indicates The input query vector at time step, Indicates The embedding representation of the data across all feature dimensions at the time step, Represents the output data after the second multi-head self-attention mechanism processing, represents the feature representation after layer normalization, Represents the original time dimension feature, represents the output features after being processed by the multi-head self-attention mechanism, Represents the final output data after multi-layer perceptron and layer normalization.

[0067] In this embodiment, in step S3, the Encoder-Decoder loop structure is used to perform information distillation and result output.

[0068] At the Encode layer, information integration and processing are performed through information distillation. The process is as follows:

[0069]

[0070] in, is the number of encoder layers, For the encoder The intermediate representation of the layer, is the initial embedding of the input data, Indicated in Tier and The merging operation of the time-step embeddings, It means the encoder after the time series attention mechanism The output of the layer.

[0071] At the Decoder layer, Represents the input-output relationship, and performs the following TSA layer, Attention layer, and Add and Norm operations as the final output:

[0072]

[0073] in, Indicates the decoder The input embedding representation of the layer is given in the formula In the example, the skip connection part (i.e., the addition operation Add) uses the intermediate result to directly overwrite the initial input, which not only retains the original input information but also protects the new features; represents the initial embedding vector of the decoder, TSA represents the time series attention operation, Indicates the decoder The output of the previous layer of the layer, Indicates the decoder The input embedding vector of the layer, Indicates the decoder The output representation of the layer after being processed by the multi-head self-attention mechanism, Indicates the decoder The intermediate output representation of the layer is used as the key and value input of the decoder. Indicates the decoder The intermediate output representation of the layer after being processed by the multi-head self-attention mechanism, Indicates the decoder The embedded representation of the layer after being processed by the multi-head self-attention mechanism, Indicates the decoder The final output representation of the layer after multi-layer perceptron and layer normalization.

[0074] Exemplarily, in step S4, an evaluation index is also used to perform performance evaluation on the trained power prediction model.

[0075] Specifically, in this embodiment, in step S4, the initialization strategy of the training parameters is set, and the power prediction model is trained. During the training process, the existing optimization algorithm is used to adjust the parameters to minimize the training error and avoid overfitting, thereby improving the generalization performance of the model. The trained power prediction model is used to predict wind and solar power, and the model output is processed by Inv-norm to obtain the wind and solar power prediction result. In this embodiment, the mean absolute error indicator is used to evaluate the output result of the trained power prediction model.

[0076] Exemplarily, step S5 includes:

[0077] The output variation characteristics of wind power and photovoltaic power are analyzed, a system power fluctuation model is constructed based on the wind and photovoltaic power prediction results and system stability requirements, and the optimal wind and photovoltaic capacity ratio that minimizes the system power fluctuation is calculated.

[0078] An energy storage system model is constructed, and a mapping framework of power fluctuation quantification index and energy storage capacity ratio function is determined according to the energy storage system model.

[0079] Specifically, in this embodiment, the output variation characteristics of wind power and photovoltaic power are analyzed, and a system power fluctuation model is constructed based on the wind and photovoltaic power prediction results and system stability requirements. By setting a reasonable economic output target, the existing optimization algorithm is used to solve the optimal wind and photovoltaic capacity ratio that minimizes the system power fluctuation.

[0080] The system power fluctuation model is as follows:

[0081]

[0082] in, , , , and They are system power, wind power forecast power, photovoltaic forecast power, energy storage output power and load power demand. It is a wind speed sensitive power conversion factor, which is directly related to the real-time wind speed condition and is dynamically determined based on a preset wind speed-power mapping relationship; It is an irradiance-sensitive power conversion coefficient, which can be calculated in real time based on the current solar irradiance level and the corresponding relationship between irradiance and power output.

[0083] In this embodiment, by calculating the power shortage To indicate that at a specific time The gap between the system power and the load demand, the power shortage calculation method is as follows:

[0084]

[0085] In this embodiment, the charging and discharging strategy of the energy storage system is adjusted by power shortage to enhance the stability of the system. The energy storage system model includes analyzing the energy release characteristics of the energy storage system during charging and discharging, where:

[0086] The charging status characteristics are as follows:

[0087]

[0088] The discharge state characteristics are as follows:

[0089]

[0090] in, , Respectively in and The state quantity of the energy storage system at each moment, represents the total capacity of the energy storage system, is the charge and discharge efficiency, is the power loss coefficient, Indicates the time interval, is the Coulombic efficiency of the energy storage battery, is the output power of the energy storage system in the current state, , are the current and voltage in the charging state, , are the current and voltage in the discharge state respectively.

[0091] According to the energy storage system model, the effectiveness of the energy storage system in the power implementation layout and daily operation and maintenance process is comprehensively considered to determine the mapping framework between the power fluctuation quantitative index and the energy storage capacity ratio function.

[0092] Exemplarily, in step S6, the multi-objective functions and constraints of the wind, solar, and storage combined installed capacity ratio model are determined, and the wind, solar, and storage combined installed capacity ratio model is solved using a multi-objective mantis search algorithm to obtain the wind, solar, and storage combined installed capacity ratio scheme.

[0093] Specifically, in this embodiment, in step S6, the multi-objective function and constraint conditions are determined according to the existing wind-solar-storage combined installed capacity ratio model, and the termination conditions of the wind-solar-storage combined installed capacity ratio scheme are set. , randomly generate a set of initial solutions representing the wind, solar and storage combined installed capacity ratio scheme , evaluate the power fluctuation quantitative indicators and energy storage economic costs corresponding to the current solution, and determine the congestion , and select the non-dominated solution, update the position by simulating the mantis attack behavior, and continue to iterate until the optimal wind, solar and storage combined installed capacity ratio scheme is obtained.

[0094] Among them, the mantis attack behavior steps can be divided into exploration, exploitation and sexual cannibalism. The specific position update method is as follows:

[0095]

[0096] in, , , , are the latest solution, current solution, random solution and current optimal solution respectively. represents the control parameters of the current optimal solution, Represents the control parameters of the random solution, which is used to balance the exploration and exploitation of the algorithm.

[0097] In other possible embodiments, other search algorithms may also be used.

[0098] In a possible embodiment, the multi-dimensional feature data of the scenery in a certain place for 8 months from August 1, 2022 to March 31, 2023 is used as input, and according to the following Figure 1 and Figure 2 The flow chart and logic diagram of a wind-solar-storage combined installed capacity matching method based on power prediction are shown in the figure to build a power prediction model, setting the number of iterations to 50, the number of input features to 6, the number of hidden neurons to 64, and the number of heads to 4. Figure 3 The figure shows the comparison between the long series forecast value and the actual value of wind power. Figure 4 The figure shows the comparison between the predicted value and the actual value of the photovoltaic power long series. Figure 3 and Figure 4 It can be seen that the method of this application achieves accurate prediction of long-sequence wind and solar power. Based on the wind and solar power prediction results, the optimal installed capacity ratio of wind power and photovoltaic power generation is calculated and optimized. Figure 5 The figure shows the relationship between wind power capacity, photovoltaic capacity and power shortage. Figure 5 The optimal capacity ratio of wind and solar power that minimizes system power fluctuations can be obtained. On this basis, the role of energy storage is considered, according to Figure 6 The energy storage net output power curve and system objectives and constraints shown in the figure are used to construct a wind-solar-storage combined installed capacity matching model that takes into account the dual objectives of power fluctuation and economy, determine the multi-objective functions and constraints of the wind-solar-storage combined installed capacity matching model, use the multi-objective mantis search algorithm to solve it, verify the obtained wind-solar-storage combined installed capacity matching scheme, and determine the final configuration scheme. In summary, the method of the present application improves the stability of the wind-solar-storage combined output and the economy of the installed capacity matching.

[0099] The embodiments of the present application combine the correlation coefficient between features, scaling factors, segmentation, encoding, and learnable intermediate layers to capture the dependencies between global data from a global attention search, thereby achieving accurate prediction of long-sequence wind and solar power. Based on the wind and solar power prediction results, a wind, solar, and storage combined installed capacity ratio model that considers power fluctuations and economic goals is constructed and solved, thereby improving the stability of the wind, solar, and storage combined output and the economy of the installed capacity ratio.

[0100] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0101] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for allocating wind, solar and energy storage combined installed capacity based on power prediction, characterized in that: The following steps are involved: Step S1, reading the multi-dimensional feature data of the wind and solar power, calculating the correlation coefficient between the features, using the scaling factor to gradually process the multi-dimensional feature data of the wind and solar power, and dividing them to obtain different feature data sets; Step S2, segmenting and encoding the different feature data sets to obtain embedded representation data; Step S3, constructing a power prediction model using a learnable intermediate layer and an Encoder-Decoder loop structure, wherein the learnable intermediate layer is used to realize the interaction between embedded representation data of different time series, and the Encoder-Decoder loop structure is used to perform information distillation and result output; Step S4, setting a training strategy, training the power prediction model, using an optimization algorithm to adjust parameters during the training process, using the trained power prediction model to perform wind and solar power prediction, and obtaining a wind and solar power prediction result; Step S5, constructing a system power fluctuation model based on the wind and solar power prediction results and system stability requirements, and constructing an energy storage system model; Step S6, considering the economic cost of the wind-solar-storage combined system, constructing a wind-solar-storage combined installed capacity ratio model that considers the dual objectives of power fluctuation and economy based on the system power fluctuation model and the energy storage system model, and solving to obtain a wind-solar-storage combined installed capacity ratio scheme; In step S3, before using the learnable intermediate layer and the Encoder-Decoder loop structure to build the power prediction model, Add and Norm operations are performed on the embedded representation data of different time series; Step S5 includes: Analyze the output variation characteristics of wind power and photovoltaic power, build a system power fluctuation model based on the wind and photovoltaic power prediction results and system stability requirements, and calculate the optimal wind and photovoltaic capacity ratio that minimizes system power fluctuations; Constructing an energy storage system model, and determining a mapping framework between a power fluctuation quantification index and an energy storage capacity ratio function according to the energy storage system model; In step S6, the multi-objective functions and constraints of the wind-solar-storage combined installed capacity ratio model are determined, and the wind-solar-storage combined installed capacity ratio model is solved using a multi-objective mantis search algorithm to obtain the wind-solar-storage combined installed capacity ratio scheme.

2. The method for allocating wind, solar and energy storage combined installed capacity based on power prediction according to claim 1 is characterized in that: In step S1, the wind and solar multi-dimensional feature data includes: historical wind and photovoltaic output power and environmental index data; Step S1 includes: Reading the wind and solar multi-dimensional feature data, determining the correlation coefficient between the features, and preprocessing the wind and solar multi-dimensional feature data; A scaling factor is introduced to perform Norm processing on the scenery multi-dimensional feature data, and the processed scenery multi-dimensional feature data are divided to obtain different feature data sets.

3. The method for allocating wind, solar and energy storage combined installed capacity based on power prediction according to claim 1 is characterized in that: Step S2 includes: Constructing a data set segmentation function, and using the data set segmentation function to segment the different feature data sets into segments to obtain data segments; The data segment is encoded to obtain the embedded representation data.

4. The method for allocating wind, solar and energy storage combined installed capacity based on power prediction according to claim 1 is characterized in that: In step S4, an evaluation index is also used to perform performance evaluation on the trained power prediction model.

Images