Self-adaptive voltage regulation method and system for micro-grid with multi-source power generation characteristic

Abstract

The invention discloses a self-adaptive voltage regulation method and system for a multi-source power generation characteristic microgrid, relates to the technical field of smart grids, and aims to predict voltage fluctuation signals by collecting physical characteristic data of each renewable energy source in real time, generating characteristic embedded representation and utilizing a multi-head attention mechanism. And according to the generated voltage fluctuation prediction signal, the system dynamically adjusts the voltage output of each sub-grid through an adaptive control algorithm. The method specifically comprises the steps of calculating a target voltage adjustment amount in real time, adjusting control parameters of the power conversion device, and updating parameters of a control algorithm through feedback signals. Meanwhile, a gradient descent algorithm is adopted to minimize a voltage regulation error, and the regulation precision and response speed of the system are improved. The method has the advantages of being high in self-adaptive capacity, capable of processing fluctuation characteristics of different time scales at the same time, capable of remarkably improving the stability and accuracy of voltage regulation of the micro-grid, and suitable for a complex multi-source power generation environment.

Description

Technical Field

[0001] The present application relates to the field of smart grid technology, and in particular to an adaptive voltage regulation method and system for a microgrid with multi-source power generation characteristics. Background Art

[0002] With the increasing global energy shortage and environmental problems, clean energy has gradually become an important direction for the development of modern power systems. Microgrid technology has been widely used due to its flexibility and high efficiency, especially in multi-source power generation systems that integrate multiple renewable energy sources. Microgrids can effectively integrate renewable energy such as solar energy and wind energy, while providing coordinated control of distributed power sources to ensure a stable supply of electricity.

[0003] However, there are many technical difficulties in voltage regulation and power distribution of multi-source microgrids. Due to the large differences in the power generation characteristics of different renewable energy sources and the strong volatility of these energy sources, traditional voltage regulation methods are difficult to adapt to the dynamic changes of multi-source power generation systems, often resulting in voltage instability or unreasonable power distribution. In addition, the existing technology usually adopts a fixed control strategy, which is difficult to respond to complex grid fluctuations in real time. This is particularly prone to problems such as excessive voltage fluctuations and unbalanced power distribution between sub-grids when multiple renewable energy sources coexist.

[0004] For example, the Chinese patent with the authorization announcement number CN110661248B discloses a multi-source DC microgrid adaptive robust power coordination allocation method. This method establishes a continuous and smooth nonlinear equivalent resistance regulation law, calculates the stable regulation control amount of the bus voltage and the power coordination control amount of the access converter, and realizes effective control of the access converter to deal with the problems of disordered power distribution and uneven line flow between multiple distributed power sources.

[0005] The above methods have the problems raised by this background technology, that is, in a multi-source power generation microgrid system, due to the lack of accurate description of the dynamic characteristics of the power generation source, especially in a renewable energy system with strong volatility, the existing technology is difficult to adapt to voltage fluctuations in real time, resulting in insufficient accuracy of power distribution and voltage regulation. In addition, traditional methods lack flexible response capabilities to complex dynamic changes and cannot effectively integrate grid fluctuation information at different time scales. The grid stability and response efficiency need to be improved. Summary of the invention

[0006] The purpose of the present invention is to provide an adaptive voltage regulation method and system for a microgrid with multi-source power generation characteristics, which can collect physical characteristic data of each renewable energy source in real time, generate feature embedding representation, predict voltage fluctuation signals through a multi-head attention mechanism, and dynamically adjust the voltage output of each subgrid based on the predicted signal, thereby achieving stable regulation of the microgrid voltage.

[0007] In a first aspect, the present application provides an adaptive voltage regulation method for a microgrid with multi-source power generation characteristics, wherein the microgrid includes a plurality of subgrids, each subgrid being connected to at least one renewable energy power generation source, and the method includes:

[0008] Collecting physical property data of each renewable energy source in real time;

[0009] Based on the physical property data, generating a feature embedding representation of each renewable energy source; wherein the feature embedding representation is used to determine the dynamic characteristics and volatility of each renewable energy source;

[0010] The feature embedding representation is input into a multi-head attention mechanism to generate a comprehensive voltage fluctuation prediction signal;

[0011] Based on the comprehensive voltage fluctuation prediction signal, the voltage output of each sub-grid is dynamically adjusted.

[0012] As an optional implementation, the method for generating a feature embedding representation of each renewable energy source includes:

[0013] Collecting the physical property data in real time to generate an original data set; wherein the data in the original data set is in the form of time series data;

[0014] Extracting spatial features from the preprocessed raw data set using a convolutional neural network;

[0015] Mapping the spatial features into a one-dimensional vector through a fully connected layer to form an intermediate feature representation;

[0016] The intermediate feature representation is nonlinearly transformed through an activation function to generate a feature embedding representation for each renewable energy source.

[0017] As an optional implementation, the activation function includes: a parameterized exponential linear unit;

[0018] The function expression of the parameterized exponential linear unit is:

[0019]

[0020] represents the input value, that is, each element of the intermediate feature representation after the fully connected layer, It is an adjustable parameter used to control the output amplitude of the activation function in the negative region;

[0021] Among them, the adjustable parameters Determined based on statistical characteristics of the physical property data.

[0022] As an optional implementation, the method for generating a comprehensive voltage fluctuation prediction signal includes:

[0023] The feature embedding representation is copied into multiple parallel inputs and input into multiple attention heads respectively;

[0024] Each attention head corresponds to a preset time scale, performs weighted processing on the feature embedding representation, and extracts the time correlation features of the corresponding time scale;

[0025] The self-attention mechanism is used to calculate the correlation between the elements in the feature embedding representation and generate the attention weight matrix;

[0026] The outputs of each attention head are fused to generate a comprehensive voltage fluctuation prediction signal.

[0027] As an optional implementation, fusing the outputs of the attention heads includes:

[0028] Based on the feature differences output by each attention head, the corresponding fusion weights are calculated;

[0029] The calculation of the fusion weight includes:

[0030] Use the weight generation network to analyze the output of each attention head and generate dynamic fusion weights;

[0031] Wherein, the weight generation network is trained based on the real-time volatility index of renewable energy to adapt to different volatility characteristics;

[0032] Multiply the output of each attention head by the corresponding fusion weight to get the weighted output;

[0033] The weighted outputs are summed to generate a comprehensive voltage fluctuation prediction signal.

[0034] As an optional implementation, dynamically adjusting the voltage output of each subgrid based on the comprehensive voltage fluctuation prediction signal includes:

[0035] Using an adaptive control algorithm, based on the comprehensive voltage fluctuation prediction signal, a target voltage adjustment amount of each sub-grid is calculated in real time;

[0036] According to the target voltage adjustment amount, adjusting the control parameters of the power conversion device of each sub-grid;

[0037] Monitor the actual voltage output of each sub-grid, compare it with the target voltage adjustment amount, and generate a feedback signal;

[0038] Based on the feedback signal, parameters of the adaptive control algorithm are updated.

[0039] As an optional implementation manner, updating the parameters of the adaptive control algorithm based on the feedback signal includes:

[0040] Calculating the gradient of the loss function according to the feedback signal using a gradient descent algorithm;

[0041] Based on the gradient, parameters of an adaptive control algorithm are adjusted to minimize the voltage regulation error.

[0042] In a second aspect, the present application also provides an adaptive voltage regulation system for a microgrid with multi-source power generation characteristics, the system comprising: a collection unit, a first processing unit, a second processing unit, and a regulation unit; wherein:

[0043] The collection unit is used to collect physical characteristic data of each renewable energy source in real time;

[0044] The first processing unit is used to generate a feature embedding representation of each renewable energy source based on the physical property data; wherein the feature embedding representation is used to determine the dynamic characteristics and volatility of each renewable energy source;

[0045] The second processing unit is used to embed the feature into a multi-head attention mechanism to generate a comprehensive voltage fluctuation prediction signal;

[0046] The regulating unit is used to dynamically regulate the voltage output of each sub-grid based on the comprehensive voltage fluctuation prediction signal.

[0047] In a third aspect, the present application also provides a computer device, comprising a processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and the processor is used to execute the machine-readable instructions stored in the memory, and when the machine-readable instructions are executed by the processor, the machine-readable instructions perform the steps in the above-mentioned first aspect, or any possible implementation of the first aspect.

[0048] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed, the steps of the above-mentioned first aspect, or any possible implementation of the first aspect are executed.

[0049] Compared with the prior art, the beneficial effects of the present invention are mainly reflected in its strong adaptability and more comprehensive analysis of voltage fluctuation characteristics at multiple time scales. By collecting the physical property data of renewable energy in real time and using the multi-head attention mechanism to generate a comprehensive voltage fluctuation prediction signal, the voltage output can be dynamically adjusted according to the fluctuation characteristics of different power generation sources. In addition, the dynamic fusion of multi-head attention outputs is realized through the weight generation network, making the voltage fluctuation prediction at different time scales more accurate. The gradient descent algorithm is used for parameter optimization, so that the adaptive control algorithm can be updated in real time, the voltage regulation error is minimized, and the stability and regulation accuracy of the power grid are significantly improved. Compared with the prior art, it can better cope with the fluctuation characteristics of complex multi-source power generation systems, ensure the stability and rapid response capability of the system, and improve the voltage regulation accuracy, enhance the robustness and flexibility of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of an adaptive voltage regulation method for a microgrid with multi-source generation characteristics provided in an embodiment of the present application;

[0051] Figure 2 A flowchart of a method for generating a feature embedding representation of each renewable energy source provided in an embodiment of the present application;

[0052] Figure 3 A schematic diagram of an adaptive voltage regulation system for a microgrid with multi-source power generation characteristics provided in an embodiment of the present application;

[0053] Figure 4 A schematic diagram of a microgrid structure provided in an embodiment of the present application.

[0054] Reference numerals: 10, acquisition unit; 20, first processing unit; 30, second processing unit; 40, adjustment unit. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0056] Research has found that traditional microgrid voltage regulation methods usually rely on fixed rules or empirical formulas to regulate system voltage, and lack the ability to accurately capture the dynamic characteristics of various types of renewable energy. For example, solar and wind energy have obvious random fluctuations, and their power generation characteristics are greatly affected by environmental conditions (such as irradiance and wind speed). This uncertainty makes it difficult for traditional regulation methods to respond flexibly when faced with multi-source power generation systems, and the regulation accuracy and speed are limited.

[0057] For this purpose, see Figure 1 and Figure 4As shown, Figure 1 A flowchart of an adaptive voltage regulation method for a microgrid with multi-source power generation characteristics provided in an embodiment of the present application, Figure 4 A schematic diagram of a microgrid structure provided in an embodiment of the present application.

[0058] The microgrid includes a plurality of subgrids, each of which is connected to at least one renewable energy power generation source; the method includes steps S101 to S104, wherein:

[0059] S101: Collecting physical property data of each renewable energy source in real time.

[0060] S102: Based on the physical property data, generate a feature embedding representation for each renewable energy source; wherein the feature embedding representation is used to determine the dynamic characteristics and volatility of each renewable energy source.

[0061] S103: embedding the feature into a multi-head attention mechanism to generate a comprehensive voltage fluctuation prediction signal.

[0062] S104: Dynamically adjust the voltage output of each sub-grid based on the comprehensive voltage fluctuation prediction signal.

[0063] The present invention aims to solve the following technical problems: How to achieve adaptive voltage regulation in a multi-source power generation environment: By embedding the physical characteristic data, the present invention can accurately capture the fluctuation characteristics of each subgrid in the multi-source power generation system, generate a voltage fluctuation prediction signal, and dynamically adjust the voltage output accordingly, thus solving the problem that traditional methods cannot adapt to the multi-source power generation environment.

[0064] How to improve the accuracy and stability of voltage regulation: The present invention analyzes and weightedly processes the feature embedding representation of each sub-grid through a multi-head attention mechanism to generate a high-precision voltage fluctuation prediction signal, thereby achieving more accurate voltage regulation and improving the stability of the microgrid.

[0065] For example, Figure 4 As shown, the microgrid consists of multiple subgrids (subgrid 1, subgrid 2, subgrid 3), where subgrid 1 is connected to source 1, subgrid 2 is connected to source 2, and subgrid 3 is connected to source 3, that is, each subgrid is connected to at least one renewable energy power generation source, which may include: solar power generation, wind power generation, biomass power generation, etc. Each subgrid can operate independently, and all subgrids are connected to the main grid or other subgrids through power electronic equipment (such as inverters).

[0066] Regarding S101 above:

[0067] In order to achieve adaptive voltage regulation, it is first necessary to collect the physical characteristics of each renewable energy source in real time. In order to ensure that the dynamic characteristics and volatility of each renewable energy source are monitored in real time, the physical characteristic data of each renewable energy source can be collected through sensors. Specific physical characteristic data include but are not limited to: irradiance, ambient temperature, component temperature, power generation and other data of solar power generation system; wind speed, wind direction, blade speed, power generation and other data of wind power generation system; fuel consumption rate, power generation efficiency, power generation and other data of biomass power generation system.

[0068] These physical property data are collected by sensors and transmitted to the central control system in real time for further analysis. The collection frequency can be set based on the volatility of the power generation source. For example, the characteristic data of solar and wind energy can be set to update once per second.

[0069] Regarding S102 above:

[0070] By analyzing the physical property data, a feature embedding representation of each renewable energy source is generated. This feature embedding representation is used to quantify the dynamic characteristics and volatility of each renewable energy source.

[0071] In a specific implementation, the physical property data collected in S101 can be filtered and denoised through preprocessing to eliminate possible environmental interference and collection errors; in addition, in order to eliminate the differences between different physical quantities, all data need to be normalized so that the physical properties of different types of power generation sources can be compared under the same standard.

[0072] Based on the normalized data, specific algorithms are applied to extract the dynamic characteristics of renewable energy. For example, principal component analysis (PCA) or other machine learning algorithms can be used to extract the most significant features in the data, which can represent the changing patterns and fluctuations of each power source.

[0073] The extracted key feature data is converted into an embedded representation, which compresses the physical characteristic data into a high-dimensional vector representation to summarize all the important characteristics of the renewable energy at the current moment. This vector will serve as input data in the subsequent steps.

[0074] See also Figure 2 , Figure 2 Flow chart of a method for generating a feature embedding representation of each renewable energy source. As an optional implementation, the method for generating a feature embedding representation of each renewable energy source includes steps S201 to S204, wherein:

[0075] S201: collecting the physical property data in real time to generate an original data set; wherein the data in the original data set is in the form of time series data;

[0076] S202: extracting spatial features from the preprocessed original data set using a convolutional neural network;

[0077] S203: Mapping the spatial features into a one-dimensional vector through a fully connected layer to form an intermediate feature representation;

[0078] S204: Performing a nonlinear transformation on the intermediate feature representation through an activation function to generate a feature embedding representation of each renewable energy source.

[0079] Regarding the above S201:

[0080] In the specific implementation, a sensor device with high-frequency sampling is used, and the specific sampling frequency can be set according to the actual application. For example, for solar panels, 1-10 times per second can be collected; while for wind turbines, changes in wind speed and direction require a higher sampling frequency.

[0081] All collected data is stored in a distributed database in the form of time series for subsequent processing and analysis. The storage structure of time series data allows efficient historical backtracking and trend analysis of data, and ensures real-time performance.

[0082] It is understandable that time series data has a clear temporal relationship, which is very important in renewable energy systems. For example, the power change of solar power generation is not only related to the current irradiance, but also to the irradiance change in the previous period. Therefore, time series data can reflect the dynamic characteristics of the energy system and provide richer background information than static data. This form of data enables the model to capture short-term fluctuations and long-term trends, which is conducive to subsequent feature extraction and voltage regulation.

[0083] Exemplarily, a low-latency communication protocol (such as Modbus, IEC 61850, etc.) can be used to transmit the collected physical characteristic data to the processing part of the main power grid in real time.

[0084] Regarding the above S202:

[0085] Convolutional Neural Networks (CNN) are used to extract spatial features from time series data. Spatial features refer to local patterns hidden in time series data, such as the effect of wind speed on power generation and the effect of irradiance on current output.

[0086] Exemplarily, the specific structure of CNN includes:

[0087] Input layer: Receives preprocessed time series data in the form of a two-dimensional array, where rows represent different physical properties (such as wind speed, temperature, etc.) and columns represent data at different time points.

[0088] Convolutional layer: The first convolutional layer can be composed of several one-dimensional convolution kernels. The size of each convolution kernel can be set according to the data characteristics, such as a convolution window of size 5 or 7. The convolution kernel slides the time series to extract the local pattern at each moment and within a certain period of time before and after.

[0089] Activation function: ReLU (Rectified Linear Unit) activation function is used after each convolutional layer. Its main purpose is to introduce nonlinearity and enhance the network's ability to recognize complex features.

[0090] Pooling layer: Each convolutional layer can be followed by a Max Pooling layer for dimensionality reduction. The pooling window size can be set to 2 or 3, depending on the frequency and characteristics of the input data.

[0091] In this way, through CNN, important patterns in time series data can be extracted, such as the impact of wind speed changes on power generation output at a specific moment. The convolution operation enables local change features to be captured and used for subsequent analysis, providing a basis for feature embedding representation.

[0092] Regarding the above S203:

[0093] After the convolutional layer extracts the spatial features of the time series data, these high-dimensional spatial features are mapped into one-dimensional vectors through the fully connected layer.

[0094] In the specific implementation, the output of CNN is a multi-dimensional feature map, which is flattened to form a one-dimensional input vector and input to the fully connected layer. The fully connected layer consists of several neurons, each of which receives all the inputs from the previous layer.

[0095] The fully connected layer is embedded at the end of the entire CNN as the structure after the convolutional layer. After the convolutional layer extracts local features, the fully connected layer combines all local features through a weighted sum operation to generate a one-dimensional vector that can reflect the dynamic changes of the entire time series, so as to achieve efficient feature dimensionality reduction and prepare for the subsequent generation of feature embedding representation.

[0096] Regarding the above S204:

[0097] After the linear transformation of the fully connected layer, nonlinear transformation is performed through the activation function to further improve the expressive power of the model.

[0098] For example, a ReLU (rectified linear unit) activation function or a more complex PELU (parameterized exponential linear unit) activation function may be selected. For example, PELU can effectively process negative input features, making the model more flexible in processing negative inputs.

[0099] Through nonlinear activation functions, the model's ability to capture complex relationships can be enhanced. Especially in multi-source power generation systems, the dynamic changes of various energy sources are not simple linear relationships, so nonlinear transformation is crucial.

[0100] The output after activation function processing is the feature embedding representation of each renewable energy source. This embedding representation can fully reflect the dynamic characteristics of each energy system and provide accurate input basis for subsequent voltage regulation.

[0101] As an optional implementation, the activation function includes: a parameterized exponential linear unit (PELU); the function expression of the parameterized exponential linear unit is:

[0102]

[0103] represents the input value, that is, each element of the intermediate feature representation after the fully connected layer, It is an adjustable parameter used to control the output amplitude of the activation function in the negative region;

[0104] Among them, the adjustable parameters Determined based on statistical characteristics of the physical property data.

[0105] In the specific implementation, for positive input, PELU maintains linear output; for negative input, PELU generates smooth nonlinear output through exponential function. The main purpose of this design is to enhance the model's ability to handle negative input, especially in complex time series data. When the input contains more negative values, PELU can better retain negative information and prevent the gradient disappearance problem.

[0106] also, Controls the output amplitude of the activation function in the negative region. This parameter is not fixed, but is dynamically determined based on the statistical characteristics of the physical characteristics data of each renewable energy source. For example, when dealing with the irradiance and power output of a solar power generation system, It can be dynamically adjusted according to the variance or mean of the data to adapt to different volatility characteristics. Similarly, in wind power generation systems, It can be optimized based on historical data of wind speed and power fluctuations to more accurately describe the characteristics of its negative part.

[0107] For example, taking a wind power generation system as an example, the physical characteristic data includes wind speed and power generation In order to make the parameters of the activation function Able to adapt to different fluctuations in wind speed and power, and can dynamically adjust by calculating the statistical characteristics of these data in real time The value of .

[0108] For example, high-precision sensors installed on wind turbines collect wind speed in real time. and power generation , the sampling frequency is set to once per second.

[0109] Use a sliding window of length N (for example, the data of the most recent 60 seconds) to perform statistical analysis on the collected data.

[0110] For example, we can define The calculation formula is:

[0111]

[0112] in, is the adjustment coefficient, and the value range can be set to , used to control so that it fluctuates within a reasonable range. and are the standard deviations of power generation and wind speed, respectively.

[0113] When the variance of the generated power Larger wind speed variance When it is small, it means that the power generation has a large fluctuation when the wind speed does not change much, which may be an abnormality or fault. As the value increases, the response of the activation function in the negative area is enhanced, making the model more sensitive to abnormal features.

[0114] When the wind speed variance Larger, indicating that the wind speed itself fluctuates greatly. In order to prevent the model from overfitting to too much noise, The value is reduced, the output amplitude of the activation function in the negative region is reduced, and the robustness of the model is improved. Value, in PELU for each moment Perform nonlinear transformation.

[0115] Compared with the traditional ReLU activation function, the PELU activation function can handle negative input more flexibly and adjust the parameters , dynamically optimizes the output of the activation function according to the specific physical characteristics of renewable energy, thereby improving the nonlinear expression ability of the model. When dealing with different types of renewable energy sources, the PELU activation function can adapt to different physical data distributions to achieve more accurate feature extraction. This flexibility is of great significance for microgrid adaptive voltage regulation in multi-source power generation environments, and can ensure that the model exhibits better robustness and accuracy when dealing with complex volatility.

[0116] In a specific implementation, in order to determine whether the variance is "large" or "small", a sliding window for calculating the variance (for example, the last thirty seconds or sixty seconds) can be set first, and the real-time variance of wind speed and power generation can be calculated in each sliding window. In order to determine a specific threshold to determine whether a certain variance is "large" or "small", one or more representative benchmark values ​​can be determined based on historical operating data. For example, by statistically analyzing the variance of past wind speed and power generation, their average level and standard deviation over a long period of time can be obtained, and the judgment standard can also be selected by observing different percentiles of their distribution.

[0117] In addition, if several "variance threshold" data sets have been formed when the system is initialized or when historical data is sufficient, after the wind speed variance or power generation variance at a certain moment is calculated in real time, it is compared with the pre-selected benchmark value: if the real-time variance is higher than the corresponding threshold, it can be judged as "large variance"; if it is lower than the threshold, it can be judged as "small variance". When the real-time variance of power generation is much higher than its threshold, but the wind speed variance is lower than its threshold, it can be understood that the power generation fluctuates greatly when the wind speed fluctuation is not significant, indicating that there may be system abnormalities or failures, which require further detection or processing; on the contrary, when the wind speed variance is detected to be much higher than the corresponding threshold, it means that the current wind speed fluctuation is large, and it should be reserved in the control algorithm to avoid overfitting or frequent adjustments in a high-volatility environment, thereby ensuring the stability and robustness of the system operation.

[0118] As an optional implementation, the method for generating a comprehensive voltage fluctuation prediction signal includes: copying the feature embedding representation as multiple parallel inputs, and inputting them into multiple attention heads respectively; each attention head corresponds to a preset time scale, and weighted processing is performed on the feature embedding representation to extract the time correlation features of the corresponding time scale; using a self-attention mechanism to calculate the correlation between the elements in the feature embedding representation to generate an attention weight matrix; and fusing the outputs of each attention head to generate a comprehensive voltage fluctuation prediction signal.

[0119] Regarding S103 above:

[0120] In the specific implementation, the feature embedding representation of each renewable energy source is replicated as multiple parallel inputs. These embedding representations will be input into multiple attention heads respectively.

[0121] The number of attention heads can be set according to actual needs, for example, 4, 8 or 12, and the specific number can be optimized through model training.

[0122] Each attention head corresponds to a different time scale. The time scale can be set based on the volatility characteristics of renewable energy. For example, a short time scale can capture voltage fluctuations at the second level, while a long time scale can capture trends at the minute and hour level.

[0123] For each specific time scale, the attention head performs weighted processing on the feature embedding representation by calculation, thereby extracting the correlation features at that time scale.

[0124] The core of the self-attention mechanism is to calculate the correlation between the elements in the feature embedding representation. In the specific implementation, for the input feature embedding representation y, firstly generate three sets of vectors of query, key, and value through linear transformation:

[0125]

[0126] in, , , are the weight matrices for query, key, and value, respectively, which are determined after model training. Then the attention score S is calculated by the dot product of the query vector and the key vector, and the score is normalized using the Softmax function:

[0127]

[0128] in, are the dimensions of the query and key vectors.

[0129] Finally, the attention output is obtained by weighted summation:

[0130]

[0131] The output results of multiple attention heads are weighted fused to generate a comprehensive voltage fluctuation prediction signal. The purpose of fusion is to comprehensively process information of different time scales to obtain a unified voltage fluctuation prediction signal.

[0132] As an optional implementation, fusing the outputs of the attention heads includes:

[0133] The output of each attention head will have some differences due to different time scales and features. These differences are adjusted through fusion weights to ensure that the contribution of each attention head to the final output is adaptive.

[0134] In a specific implementation, a weight generation network can be used to calculate the fusion weights. The weight generation network is a small neural network whose input is the output features of each attention head.

[0135] The network dynamically generates fusion weights by analyzing the feature differences of the outputs of different attention heads. The network structure can include several fully connected layers, each of which is followed by a ReLU activation function, and the output is the fusion weight corresponding to each attention head. .

[0136] The training process of the weight generation network is based on real-time volatility indicators of renewable energy. For example, in solar power generation systems, the volatility of irradiance and power is a key factor, and the training data can include power generation fluctuations under different weather conditions.

[0137] The network adjusts its parameters by minimizing the voltage prediction error, ensuring that the weights can be adaptively adjusted under different volatility characteristics.

[0138] The calculated fusion weight It will be used to weight the output of each attention head. The specific weighting formula is:

[0139]

[0140] Where n is the number of attention heads, represents the output of the i-th attention head.

[0141] Finally, the weighted outputs are summed to generate a comprehensive voltage fluctuation prediction signal, which will be used to dynamically adjust the voltage output of each subgrid in the microgrid.

[0142] In this way, the accuracy of voltage fluctuation prediction is effectively improved through multi-time scale feature extraction, correlation analysis of self-attention mechanism, and dynamic fusion of weight generation network. The strategy of integrating information of different time scales ensures that the system can balance short-term fluctuations and long-term trends, thereby improving the response speed and stability of voltage regulation. In particular, the weight generation network based on real-time volatility dynamic training further enhances the adaptability of the system under multi-source power generation conditions, making voltage regulation more flexible and accurate.

[0143] As an optional implementation, dynamically adjusting the voltage output of each subgrid based on the comprehensive voltage fluctuation prediction signal includes: using an adaptive control algorithm to calculate the target voltage adjustment amount of each subgrid in real time based on the comprehensive voltage fluctuation prediction signal; adjusting the control parameters of the power conversion device of each subgrid according to the target voltage adjustment amount; monitoring the actual voltage output of each subgrid, comparing it with the target voltage adjustment amount, and generating a feedback signal; and updating the parameters of the adaptive control algorithm based on the feedback signal.

[0144] As an optional implementation, updating the parameters of the adaptive control algorithm based on the feedback signal includes: using a gradient descent algorithm to calculate the gradient of the loss function according to the feedback signal; and adjusting the parameters of the adaptive control algorithm according to the gradient to minimize the voltage regulation error.

[0145] In the specific implementation, a comprehensive voltage fluctuation prediction signal generated by a multi-head attention mechanism is received, which can reflect the possible voltage fluctuation trend of the microgrid in the future.

[0146] Based on this signal, an adaptive control algorithm (eg, adaptive PID control algorithm, adaptive fuzzy control algorithm) is used to calculate the target voltage adjustment of each subgrid in real time. The target voltage adjustment is used to indicate the voltage output that needs to be increased or decreased for each subgrid.

[0147] According to the calculated target voltage adjustment amount, the control parameters of the power conversion devices (such as inverters and transformers) of each sub-grid are adjusted in real time.

[0148] The control parameters may include the output voltage amplitude or phase angle of the inverter, the tap position of the transformer, etc. Adjustment of these control parameters directly affects the voltage output of each sub-grid to achieve the desired voltage level.

[0149] In a specific implementation, the actual voltage output of each sub-grid can be monitored in real time by means of a voltage sensor installed in each sub-grid.

[0150] The actual voltage output is compared with the target voltage adjustment amount, and the voltage deviation is calculated, which is the feedback signal.

[0151] According to the above feedback signal, the parameters of the adaptive control algorithm are dynamically updated.

[0152] Among them, the update of the adaptive control algorithm can be achieved through the gradient descent algorithm. For example, the loss function Defined as the square of the voltage regulation error:

[0153]

[0154] in, represents the parameter set of the adaptive control algorithm, is the voltage deviation for each subgrid.

[0155] By calculating the loss function, the control parameters Gradient , which can determine the direction of parameter adjustment. The gradient calculation formula is:

[0156]

[0157] Based on the calculated gradient, the gradient descent algorithm is used to update the parameters of the adaptive control algorithm to minimize the loss function The update formula is:

[0158]

[0159] in, is the learning rate, which is used to control the step size of each parameter update.

[0160] This step gradually reduces the voltage regulation error by repeatedly iteratively adjusting the parameters of the adaptive control algorithm to ensure that the difference between the actual voltage of each subgrid and the predicted value is minimized.

[0161] In this way, the adaptive control algorithm can calculate the target voltage adjustment of each subgrid according to the real-time voltage fluctuation, and dynamically update the control parameters by monitoring the feedback signal to ensure the accuracy and response speed of voltage regulation. The gradient descent algorithm is used to continuously optimize the parameters of the control algorithm and reduce the voltage regulation error, so that the system can show higher robustness and adaptability in a complex multi-source power generation environment. This mechanism can significantly improve the voltage stability of the microgrid and ensure the efficient and stable operation of the power system.

[0162] Those skilled in the art will appreciate that, in the above method of specific implementation, the order in which the steps are written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of the steps should be determined by their functions and possible internal logic.

[0163] Based on the same inventive concept, an adaptive voltage regulation system for a microgrid with multi-source power generation characteristics is also provided in an embodiment of the present application. Since the principle of solving the problem by the system in the embodiment of the present application is similar to the adaptive voltage regulation method for a microgrid with multi-source power generation characteristics in the embodiment of the present application, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be repeated.

[0164] Reference Figure 3As shown, it is a schematic diagram of an adaptive voltage regulation system for a multi-source power generation characteristic microgrid provided in an embodiment of the present application, wherein the microgrid includes a plurality of subgrids, each of which is connected to at least one renewable energy power generation source, and the system includes: a collection unit 10, a first processing unit 20, a second processing unit 30, and a regulation unit 40; wherein:

[0165] The collection unit 10 is used to collect physical characteristic data of each renewable energy source in real time;

[0166] The first processing unit 20 is used to generate a feature embedding representation of each renewable energy source based on the physical property data; wherein the feature embedding representation is used to determine the dynamic characteristics and volatility of each renewable energy source;

[0167] The second processing unit 30 is used to embed the feature representation into a multi-head attention mechanism to generate a comprehensive voltage fluctuation prediction signal;

[0168] The regulating unit 40 is used to dynamically regulate the voltage output of each sub-grid based on the comprehensive voltage fluctuation prediction signal.

[0169] For the description of the processing flow of each unit in the system and the interaction flow between each unit, reference may be made to the relevant description in the above method embodiment, which will not be described in detail here.

[0170] The present application also provides a computer device, including:

[0171] A processor and a memory; the memory stores machine-readable instructions executable by the processor, and the processor is used to execute the machine-readable instructions stored in the memory. When the machine-readable instructions are executed by the processor, the processor performs the following steps:

[0172] Collecting physical property data of each renewable energy source in real time;

[0173] Based on the physical property data, generating a feature embedding representation of each renewable energy source; wherein the feature embedding representation is used to determine the dynamic characteristics and volatility of each renewable energy source;

[0174] The feature embedding representation is input into a multi-head attention mechanism to generate a comprehensive voltage fluctuation prediction signal;

[0175] Based on the comprehensive voltage fluctuation prediction signal, the voltage output of each sub-grid is dynamically adjusted.

[0176] The above-mentioned memory includes internal memory and external memory; the internal memory here is also called internal memory, which is used to temporarily store the calculation data in the processor and the data exchanged with external memory such as hard disk. The processor exchanges data with the external memory through the internal memory.

[0177] The specific execution process of the above instructions can refer to the steps of the adaptive voltage regulation method for a microgrid with multi-source power generation characteristics described in the embodiment of the present application, which will not be repeated here.

[0178] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the adaptive voltage regulation method for a microgrid with multi-source power generation characteristics described in the above method embodiment are executed. The storage medium can be a volatile or non-volatile computer-readable storage medium.

[0179] An embodiment of the present application also provides a computer program product, which carries a program code. The instructions included in the program code can be used to execute the steps of the adaptive voltage regulation method for a microgrid with multi-source power generation characteristics described in the above method embodiment. For details, please refer to the above method embodiment, which will not be repeated here.

[0180] The computer program product may be implemented in hardware, software or a combination thereof. In one optional embodiment, the computer program product is implemented as a computer storage medium. In another optional embodiment, the computer program product is implemented as a software product, such as a software development kit (SDK).

[0181] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the system and device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the several embodiments provided in the present disclosure, it should be understood that the disclosed system, device and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of the device or unit can be electrical, mechanical or other forms.

[0182] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present disclosure, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0183] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present disclosure, which are used to illustrate the technical solutions of the present disclosure, rather than to limit them. The protection scope of the present disclosure is not limited thereto. Although the present disclosure is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present disclosure, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims

1. An adaptive voltage regulation method for a microgrid with multi-source generation characteristics, characterized in that: The microgrid comprises a plurality of subgrids, each subgrid being connected to at least one renewable energy power generation source, and the method comprises: Collecting physical property data of each renewable energy source in real time; Based on the physical property data, generating a feature embedding representation of each renewable energy source; wherein the feature embedding representation is used to determine the dynamic characteristics and volatility of each renewable energy source; The feature embedding representation is input into a multi-head attention mechanism to generate a comprehensive voltage fluctuation prediction signal; Based on the comprehensive voltage fluctuation prediction signal, the voltage output of each sub-grid is dynamically adjusted.

2. The adaptive voltage regulation method for a multi-source power generation characteristic microgrid according to claim 1, characterized in that: The method for generating a feature embedding representation of each renewable energy source comprises: Collecting the physical property data in real time to generate an original data set; wherein the data in the original data set is in the form of time series data; Extracting spatial features from the preprocessed raw data set using a convolutional neural network; Mapping the spatial features into a one-dimensional vector through a fully connected layer to form an intermediate feature representation; The intermediate feature representation is nonlinearly transformed through an activation function to generate a feature embedding representation for each renewable energy source.

3. The adaptive voltage regulation method for a multi-source power generation characteristic microgrid according to claim 2, characterized in that: The activation function includes: a parameterized exponential linear unit; The function expression of the parameterized exponential linear unit is: ； represents the input value, that is, each element of the intermediate feature representation after the fully connected layer, It is an adjustable parameter used to control the output amplitude of the activation function in the negative region; Among them, the adjustable parameters Determined based on statistical characteristics of the physical property data.

4. The adaptive voltage regulation method for a multi-source power generation characteristic microgrid according to claim 3, characterized in that: The method for generating a comprehensive voltage fluctuation prediction signal comprises: The feature embedding representation is copied into multiple parallel inputs and input into multiple attention heads respectively; Each attention head corresponds to a preset time scale, performs weighted processing on the feature embedding representation, and extracts the time correlation features of the corresponding time scale; The self-attention mechanism is used to calculate the correlation between the elements in the feature embedding representation and generate the attention weight matrix; The outputs of each attention head are fused to generate a comprehensive voltage fluctuation prediction signal.

5. The adaptive voltage regulation method for a multi-source power generation characteristic microgrid according to claim 4, characterized in that: The fusing of the outputs of the attention heads includes: Based on the feature differences output by each attention head, the corresponding fusion weights are calculated; The calculation of the fusion weight includes: Use the weight generation network to analyze the output of each attention head and generate dynamic fusion weights; Wherein, the weight generation network is trained based on the real-time volatility index of renewable energy to adapt to different volatility characteristics; Multiply the output of each attention head by the corresponding fusion weight to get the weighted output; The weighted outputs are summed to generate a comprehensive voltage fluctuation prediction signal.

6. The adaptive voltage regulation method for a multi-source generation characteristic microgrid according to claim 5, characterized in that: Based on the comprehensive voltage fluctuation prediction signal, dynamically adjusting the voltage output of each sub-grid includes: Using an adaptive control algorithm, based on the comprehensive voltage fluctuation prediction signal, a target voltage adjustment amount of each sub-grid is calculated in real time; According to the target voltage adjustment amount, adjusting the control parameters of the power conversion device of each sub-grid; Monitor the actual voltage output of each sub-grid, compare it with the target voltage adjustment amount, and generate a feedback signal; Based on the feedback signal, parameters of the adaptive control algorithm are updated.

7. The adaptive voltage regulation method for a multi-source power generation characteristic microgrid according to claim 6, characterized in that: Based on the feedback signal, updating the parameters of the adaptive control algorithm includes: Calculating the gradient of the loss function according to the feedback signal using a gradient descent algorithm; Based on the gradient, parameters of an adaptive control algorithm are adjusted to minimize the voltage regulation error.

8. An adaptive voltage regulation system for a microgrid with multi-source power generation characteristics, used to implement an adaptive voltage regulation method for a microgrid with multi-source power generation characteristics as claimed in any one of claims 1 to 7, characterized in that: The system comprises: a collection unit, a first processing unit, a second processing unit, and an adjustment unit; wherein: The collection unit is used to collect physical characteristic data of each renewable energy source in real time; The first processing unit is used to generate a feature embedding representation of each renewable energy source based on the physical property data; wherein the feature embedding representation is used to determine the dynamic characteristics and volatility of each renewable energy source; The second processing unit is used to embed the feature into a multi-head attention mechanism to generate a comprehensive voltage fluctuation prediction signal; The regulating unit is used to dynamically regulate the voltage output of each sub-grid based on the comprehensive voltage fluctuation prediction signal.

9. A computer device, characterized in that: include: A processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and the processor is used to execute the machine-readable instructions stored in the memory. When the machine-readable instructions are executed by the processor, the processor executes the steps of the adaptive voltage regulation method for a microgrid with multi-source power generation characteristics as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program. When the computer program is executed by a computer device, the computer device executes the steps of the adaptive voltage regulation method for a microgrid with multi-source power generation characteristics as described in any one of claims 1 to 7.

Images