A method, device and system for coordinated operation of a virtual power plant considering multi-energy complementation

Abstract

The application relates to the technical field of smart grids, in particular to a virtual power plant coordinated operation method, device and system considering multi-energy complementation, which comprises the following steps: extracting charging information of different electric vehicles during charging, composing a data set, dividing a training set and a test set, taking each record as a sample; calculating the instability, power boundary ratio, clustering evaluation value and difference of each sample in the training set, adjusting the neighborhood radius of a classification model, iteratively training the classification model, evaluating the model by using the test set to obtain a deployment model, classifying real-time charging information of the electric vehicles, obtaining real-time schedulable resource capacity of each category, establishing a multi-objective optimization function and solving the function to adjust output of distributed energy in the virtual power plant. The application can improve the classification accuracy and clustering effect of the charging behavior of the electric vehicles, accurately and efficiently coordinates the output of the distributed energy in the virtual power plant, and improves the stability of power grid operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart grid, in particular to a virtual power plant coordinated operation method, device and system considering multi-energy complementation. BACKGROUND

[0002] Due to the randomness and intermittence of photovoltaic power generation and wind power generation, the difficulty of grid regulation and control of supply-side multi-energy sources is increased, and with the wide application of electric vehicles, the charging behavior of electric vehicles has a significant impact on load balance and stability of the power grid. Electric vehicles can be used as load-side resources and as a distributed energy storage unit. Virtual power plants can integrate distributed energy sources and participate in the coordinated operation of the power grid through vehicle-to-grid interaction, achieving multi-energy complementation and load balance.

[0003] Due to the high randomness and flexibility of the charging behavior of electric vehicles, disordered charging of electric vehicles will exacerbate power grid load fluctuations. Precise coordination and control of charging and discharging of electric vehicles is premised on accurate prediction and classification of the cluster behavior of electric vehicles. When self-organizing mapping algorithm is used to classify electric vehicles, a fixed neighborhood radius update strategy is usually adopted, which has limited ability to capture local features of complex charging behavior and is difficult to effectively distinguish vehicle groups with different behavior patterns, resulting in poor classification effect of the model, affecting effective cluster management of electric vehicles, and further leading to inaccurate subsequent optimization and scheduling of virtual power plants and reducing the stability of power grid operation. SUMMARY

[0004] In order to solve the above technical problems, a virtual power plant coordinated operation method, device and system considering multi-energy complementation are provided to solve the existing problems.

[0005] The technical problem of the present application is solved by providing a virtual power plant coordinated operation method, device and system considering multi-energy complementation, including the following steps:

[0006] In a first aspect, the embodiments of the present application provide a virtual power plant coordinated operation method considering multi-energy complementation, which includes the following steps:

[0007] The charging information of different electric vehicles when charging is extracted from the historical charging pile records, a data set is formed, and a training set and a test set are divided, each record being taken as a sample, wherein the charging information includes access time, charging and discharging state at each time, and charging and discharging power; for the training set, the fluctuation of the charging and discharging state and the frequency of the state change of each sample are analyzed, and the instability of each sample is calculated; based on the cumulative deviation of the charging and discharging power of each sample, the power boundary ratio of each sample is calculated, and combined with the instability, the clustering evaluation value of each sample is obtained; the difference between each sample and the charging and discharging power of the remaining samples adjacent to the access time of each sample and the number of adjacent samples are determined to determine the difference of each sample, and combined with the clustering evaluation value, the neighborhood radius of the classification model corresponding to each sample in the training set is adjusted, the classification model is iteratively trained, the deployment model is obtained by evaluating the model with the test set, and the real-time charging information of the electric vehicle is classified to obtain the real-time schedulable resource capacity of each category, which is used as a boundary condition to establish a multi-objective optimization function and solve it to adjust the output of the distributed energy in the virtual power plant.

[0008] Preferably, the instability of each sample is calculated, comprising:

[0009] For each sample in the training set, the number of inconsistent charging and discharging states of all adjacent two time points of each sample is counted as the number of reversals of each sample;

[0010] The discrete degree of the charging and discharging state of all time points of each sample is calculated.

[0011] The instability is the product of the number of reversals and the discrete degree.

[0012] Preferably, the power boundary ratio of each sample is calculated, comprising:

[0013] Based on the charging information of each sample in the training set, a charging uplink boundary and a charging downlink boundary are respectively constructed; the total area surrounded by the charging uplink boundary and the charging downlink boundary is calculated;

[0014] The charging and discharging power of all time points of each sample in the training set is curve fitted to obtain a fitting curve; the integral curve of the fitting curve with respect to time is taken as an actual power curve;

[0015] The area surrounded by the charging uplink boundary and the actual power curve is calculated, and is recorded as a deviation area;

[0016] The power boundary ratio is the ratio of the deviation area to the total area.

[0017] Preferably, the clustering evaluation value is the product of the instability and the power boundary ratio.

[0018] Preferably, the determining the difference degree of each sample comprises:

[0019] A local time period of a preset time length is constructed with the access time of each sample in the training set as the center; all samples in the training set with the access time in the local time period are obtained, denoted as adjacent samples;

[0020] The charging and discharging power of all samples at each time is normalized, and the average of the difference between each sample and all adjacent samples at all times corresponding to the normalized charging and discharging power is calculated as the relative difference;

[0021] The number of all adjacent samples of each sample is counted and positively mapped; the ratio of the relative difference to the positively mapped result is taken as the difference degree of each sample in the training set.

[0022] Preferably, the calculation formula of the adjusted neighborhood radius of the i-th sample in the training set is: The calculation formula of the adjusted neighborhood radius of the i-th sample in the training set is: wherein, is a preset initial neighborhood radius, is a preset adjustment amplitude, is an adjustment coefficient of the i-th sample in the training set, and the adjustment coefficient is the normalized result of the product of the clustering evaluation value and the difference degree.

[0023] Preferably, the iterative training of the classification model and the use of the test set to evaluate the model to obtain a deployment model comprise: based on the adjusted neighborhood radius corresponding to different samples in the training set, taking the training set as the input of the self-organizing mapping model, training through multiple iterations, wherein the adjusted neighborhood radius gradually decays in each iteration, and the model after each iteration is completed is taken as a candidate model; each candidate model is used to classify the test set, the clustering effect of each candidate model on the test set is evaluated by using the Davies-Bouldin index, and the model with the smallest Davies-Bouldin index is selected as the deployment model.

[0024] Preferably, the establishment of the multi-objective optimization function comprises:

[0025] The target function of the minimum load variance is: , wherein, is the net load, , , are the output powers of wind power generation, photovoltaic power generation and thermal power generation at t time, respectively, is the output power of the i-th category at t time, and L is the number of all categories, ​​​P is the conventional load power at time t, T is all time points;

[0026] The objective function of minimizing the operation cost of thermal power generation is established as: wherein, represents the cost coefficient of generating unit electricity during thermal power generation;

[0027] The objective function of maximizing the consumption of photovoltaic and wind power is established as:

[0028] In a second aspect, the embodiments of the present application further provide a virtual power plant coordinated operation device considering multi-energy complementation, wherein a computer program is stored in the device, and the computer program is executed by a processor to realize the steps of the virtual power plant coordinated operation method considering multi-energy complementation.

[0029] In a third aspect, the embodiments of the present application further provide a virtual power plant coordinated operation system considering multi-energy complementation, comprising a memory, a processor and a computer program stored in the memory and running on the processor, and the processor executes the computer program to realize the steps of the virtual power plant coordinated operation method considering multi-energy complementation.

[0030] The present application has at least the following beneficial effects:

[0031] ​The application calculates the instability of each sample by analyzing the change and fluctuation degree of the charging and discharging state of the electric vehicle in the charging process, which has the beneficial effect of considering the frequency of interaction between the electric vehicle and the power grid, reflecting the complex charging behavior of the electric vehicle, and evaluating the instability of the power grid at this time, and preliminarily indicating the difficulty of clustering the charging behavior of the electric vehicle represented by the sample; the power boundary ratio of each sample is calculated, which has the beneficial effect of considering the deviation of the charging and discharging power from the ideal mode in the actual charging process of the electric vehicle, reflecting the complex charging and discharging behavior of the electric vehicle, and further evaluating the difficulty of clustering the charging behavior of the electric vehicle at this time, obtaining the clustering evaluation value of each sample, which has the beneficial effect of comprehensively evaluating the charging behavior of the electric vehicle in multiple dimensions, and reflecting the degree of difficulty in accurate classification; the difference degree of each sample is determined, which has the beneficial effect of considering the difference between the charging and discharging power of the electric vehicles represented by the samples with adjacent access time, reflecting the specificity of the charging behavior of the electric vehicle, and identifying those abnormal or rare mode samples with unique behavior in a specific period; the neighborhood radius of the classification model corresponding to each sample in the training set is adjusted, which has the beneficial effect of dynamically adjusting the neighborhood radius of the electric vehicle in the SOM model by considering the complexity and specificity of the charging behavior of the electric vehicle represented by the sample, so that the neighborhood radius of the sample with high complexity and specificity is increased, and more neurons are involved in the learning process, thereby improving the adaptability of the model to complex behavior; the classification model is iteratively trained, the model is evaluated by the test set, and the deployed model is obtained, which has the beneficial effect of evaluating the classification effect by the test set, selecting the deployed model, accurately capturing the distribution characteristics of complex charging behavior, effectively avoiding overfitting, ensuring good generalization ability of the model, and improving the classification precision and clustering effect; the real-time charging information of the electric vehicle is classified to obtain the real-time schedulable resource capacity of each category, and a multi-objective optimization function is established with the new energy consumption, load variance and minimum cost of thermal power generation as the boundary conditions and is solved to adjust the output of the distributed energy in the virtual power plant, which has the beneficial effect of considering the new energy consumption, load variance and thermal power generation cost by establishing a multi-objective optimization function, realizing comprehensive optimization, and based on the accurate classification of the real-time charging information of the electric vehicle, the virtual power plant can clearly grasp the current access to the power grid of the electric vehicle group with different behavior characteristics and its real-time schedulable charging and discharging capacity, so that the virtual power plant can accurately and efficiently coordinate the output of the distributed energy, and improve the stability of the power grid. BRIEF DESCRIPTION OF DRAWINGS

[0032] The application will be further described in detail below in combination with the drawings.

[0033] Figure 1A step flow chart of a method for coordinated operation of a virtual power plant considering multi-energy complementarity is provided in an embodiment of the present application.

[0034] Figure 2 A behavior boundary diagram of an electric vehicle is provided in an embodiment of the present application. DETAILED DESCRIPTION

[0035] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application provides a method, device and system for coordinated operation of a virtual power plant considering multi-energy complementarity, which are further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0036] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0037] Please refer to Figure 1 which shows a step flow chart of a method for coordinated operation of a virtual power plant considering multi-energy complementarity provided in an embodiment of the present application. The method comprises the following steps:

[0038] Step 1: Extract charging information of different electric vehicles when charging from historical charging pile records, form a data set, and divide training set and test set, and take each record as a sample, wherein the charging information includes access time, charging and discharging state and charging and discharging power at each time.

[0039] With the rapid increase in the number of electric vehicles, if the electric vehicles are allowed to charge in disorder, it will bring a great burden to the safety and stability of energy supply side. However, the charging and discharging characteristics of electric vehicles can be regarded as portable energy storage units to participate in virtual power plant aggregation scheduling, so as to achieve the goals of demand side load peak clipping and valley filling, increasing income or reducing cost, promoting new energy consumption, etc. The virtual power plant technology containing electric vehicles uses advanced communication technology to coordinate and control multiple distributed new energies with different regions and types. Through reasonable construction and coordinated optimization, controllable power sources such as gas turbines and energy storage systems in the region are flexibly dispatched to smooth the randomness and volatility of uncontrollable new energies such as wind power and photovoltaic power.

[0040] Thus, the virtual power plant integrates distributed energy and demand-side resources together to form a virtual power plant, wherein the distributed energy includes photovoltaic power generation, wind power generation and thermal power generation, etc. Through the integration of these resources, the virtual power plant can realize multi-energy complementation and improve the stability and reliability of energy supply. Electric vehicles play a dual role in the virtual power plant, which is both a demand side and a power supply side, and is a very special flexible load or mobile energy storage unit. The regular periodicity of production and living electricity in the demand side is strong, but the charging behavior of electric vehicles is random, and the overall market electricity demand of electric vehicles is large and cannot be accurately predicted. Therefore, the charging and discharging behavior of electric vehicles has a great influence on the coordinated operation of the virtual power plant.

[0041] By extracting the historical records of all charging piles in the coverage area of the virtual power plant, the charging information of different electric vehicles during the charging process in the historical period is obtained to form a data set, wherein the charging information includes the access time, disconnection time, real-time charging and discharging state, real-time charging and discharging power, rated battery capacity and initial electricity of the electric vehicle. When the electric vehicle is charging, the charging and discharging state is 1 and the charging and discharging power is positive. When the electric vehicle is discharging, the charging and discharging state is 0 and the charging and discharging power is negative.

[0042] In this embodiment, the length of the historical period is one day, that is, the charging record of the previous day; the time interval for collecting the charging and discharging state and the charging and discharging power is 1 min. As other implementation manners, the implementer can set it according to the actual situation.

[0043] The data set is divided into a training set and a test set; wherein each charging process of an electric vehicle in the data set, that is, each record is taken as a sample;

[0044] In this embodiment, the K-fold cross-validation method is used, wherein K is selected as 10, 80% of the data set is selected as the training set, and 20% of the data set is selected as the test set. As other implementation manners, the implementer can set it according to the actual situation.

[0045] Thus, the charging information of different electric vehicles during the charging process in the data set is obtained.

[0046] Step 2, for the training set, analyze the fluctuation of the charging and discharging state and the frequency of the state change under each sample, calculate the instability of each sample, and based on the cumulative deviation of the charging and discharging power under each sample, calculate the power boundary ratio of each sample, and combine the instability to obtain the clustering evaluation value of each sample.

[0047] Further, the periodicity of production and life electricity is strong, so it is relatively easy to predict, and the market share of electric vehicles is increasing, which can be used as a flexible energy storage unit. At the same time, due to the input and output characteristics, the charging and discharging behavior of electric vehicles has high flexibility. In order to better manage the charging and discharging behavior of electric vehicles, the charging information of electric vehicles is usually clustered, so that electric vehicles with similar charging and discharging behavior are grouped, and more refined management is realized.

[0048] The self-organizing map (SOM) neural network model is used to solve the classification problem in nonlinear space. It can obtain the maximum correlation group, i.e. the optimal classification, from high-dimensional data through self-competitive learning. Therefore, the SOM model maps the charging information of electric vehicles to a predefined grid structure to classify the charging behavior of electric vehicles. When using the model for classification, the weights are adjusted iteratively to improve the ability of the model to capture local features in the charging information, thereby improving the quality and adaptability of the clustering structure.

[0049] Secondly, the neighborhood radius of the SOM model determines the range of weight update around the winning neuron (BMU) found by an input sample in each iteration. However, in the traditional SOM model, a fixed neighborhood radius is usually set, which affects the weight update speed of the neuron. Moreover, a fixed neighborhood radius may increase the sensitivity to the initial parameter setting of the algorithm. If the initial neighborhood radius is not properly set, it may be necessary to constantly try to get a more appropriate neighborhood radius, which reduces the adaptability of the algorithm.

[0050] When the electric vehicle is connected to the charging pile, the electric vehicle can be used as a distributed power source. At this time, it only needs to ensure that the electric quantity of the electric vehicle is charged to the preset electric quantity of the customer within a fixed time period. In this embodiment, it is defaulted to reach the full electric quantity. Therefore, after being connected to the charging pile, in order to balance the demand of the power grid, when the load of the power grid is heavy, the electric vehicle can be used as a distributed power source to supply power to the power grid. When the load of the power grid decreases, the charging of the electric vehicle begins. Therefore, the charging and discharging state of the electric vehicle will change reversely, i.e. the charging state changes to the discharging state, or the discharging state changes to the charging state. Therefore, if the number of reversals of the real-time charging and discharging state of the electric vehicle is more, it indicates that the stability of the power grid during the charging process of the electric vehicle is poor, and the influence on the coordinated operation of the virtual power grid is greater.

[0051] Based on this analysis, the instability degree is calculated by analyzing the change of the charging and discharging state of the electric vehicle during the charging process in the training set, specifically as follows:

[0052] For each sample in the training set, the number of inconsistent charge and discharge states of all adjacent two time points is counted as the number of reversals of each sample;

[0053] The dispersion degree of the charge and discharge states of all time points of each sample is calculated;

[0054] In this embodiment, the dispersion degree is measured by calculating the variance of the charge and discharge states of all time points of each sample. As an alternative, the implementer can use other methods in the prior art, such as standard deviation, and this embodiment does not make special restrictions.

[0055] The product of the number of reversals and the dispersion degree is calculated as the instability of each sample in the training set;

[0056] It should be noted that the greater the number of reversals, the more reversals occur during this charging process, reflecting the poor stability of the power grid operation state. The greater the dispersion degree, the higher the randomness and unpredictability of the charge and discharge behavior, and the power grid is in a state of severe fluctuations. Therefore, the higher the instability, the more complex the charge and discharge behavior of the electric vehicle, and the more difficult the clustering. In order to reasonably classify these complex behavior electric vehicles, a larger neighborhood radius is needed in the future to increase the number of neurons affected by each sample in the training process, so that the model can learn the characteristics of complex behavior more comprehensively, thereby improving the adaptability to these complex samples.

[0057] Further, during the charging process of the electric vehicle, the charging and discharging state may be switched continuously until the battery is fully charged. Therefore, when the electric vehicle is connected, the behavior of the electric vehicle has two boundaries. Ideally, after the electric vehicle is connected, it will immediately start the charging mode with the maximum charging power until it leaves the charging pile after the battery is fully charged. This charging process can be recorded as the charging uplink boundary. After the electric vehicle is connected, it will quickly enter the efficient discharging mode to continuously discharge to the power grid with the maximum discharging power until it reaches the critical point of the battery level requirement. Before the vehicle leaves the charging pile, it will quickly start the charging mode and use the maximum charging power required for the shortest charging time to restore the full battery level. This charging process can be recorded as the charging downlink boundary. The behavior boundary diagram of the electric vehicle provided in this embodiment is shown in Figure 2 Figure 2 The line segment a-b-c in the figure represents the charging uplink boundary, and the electric vehicle is charged with the maximum charging power from to , until the full battery level is reached at ; Figure 2 The line segment a-d-e-c in the figure represents the charging downlink boundary, and the electric vehicle is charged with the maximum charging power from to ​​The battery is discharged at maximum power during the specified period, when the electric vehicle's battery is nearing depletion. Stop discharging, then After a certain time, it will charge at maximum charging power until... Always fully charged ,in, In this embodiment, the initial charge of the electric vehicle is indicated. The value is the rated battery capacity. 20%.

[0058] Based on the above analysis, electric vehicles may not necessarily reach their maximum charging and discharging power during actual charging. Therefore, the actual change in their power level should fluctuate between the upper and lower charging boundaries. The closer the power level change is to the upper charging boundary, the simpler the charging behavior of the electric vehicle and the less interaction it has with the power grid, indicating a more stable power grid operation. Therefore, by analyzing the changes in charging and discharging power under each sample, the power boundary ratio is calculated as follows:

[0059] Curve fitting is performed on the charging and discharging power at all times for each sample in the training set to obtain the fitted curve;

[0060] In this embodiment, the least squares method is used for curve fitting. The least squares method is a well-known technique and will not be described in detail here.

[0061] Integrate the fitted curve with respect to time, and use the integrated curve as the actual power curve;

[0062] Calculate the total area enclosed between the charging upbound boundary and the charging downbound boundary;

[0063] It should be noted that the initial charge level during the charging process of each electric vehicle is... Rated battery capacity Charging pile access time Disconnection time with charging station Maximum charging power ,from Begin with constancy During charging, the battery level increases linearly over time, until it reaches a certain capacity. Charging stops, thus the charging upbound boundary is obtained; correspondingly, the maximum discharge power is reached. ,from Begin with constancy During discharge, the charge decreases linearly over time, until the charge reaches... Discharge stops, and the discharge stop time is obtained. Calculation with constant from Rise to how long it takes to charge, get the charging start time , and then linearly connect to the moment, get the charging downlink boundary.

[0064] Calculate the area enclosed between the charging uplink boundary and the actual power curve, denoted as the deviation area;

[0065] The ratio of the deviation area to the total area is taken as the power boundary ratio of each sample in the training set;

[0066] It should be noted that the larger the deviation area, the farther the actual power curve of the electric vehicle deviates from the simple mode of only charging during the current charging process, and the larger the obtained power boundary ratio, the more frequent and in-depth the interaction between the charging and discharging behavior of the electric vehicle and the power grid, and the more complex the behavior of the sample, and the more difficult the subsequent clustering.

[0067] Further, based on the instability and the power boundary ratio, a clustering evaluation value is determined, specifically:

[0068] The product of the instability and the power boundary ratio is taken as the clustering evaluation value of each sample in the training set;

[0069] It should be noted that the larger the clustering evaluation value, the poorer the power grid stability of the electric vehicle corresponding to the sample during the entire charging process, the more complex the charging behavior, and it is also more difficult to be accurately classified, and the subsequent need to increase the neighborhood radius of the SOM model, so that more neurons participate in the learning process, thereby improving the adaptability of the model to complex behavior.

[0070] At this point, the clustering evaluation value of each sample in the training set is obtained.

[0071] Step 3, determine the difference degree of each sample through the difference between the charging and discharging power of the sample and the remaining samples adjacent to its access time and the number of adjacent samples, and adjust the neighborhood radius of the classification model corresponding to each sample in the training set in combination with the clustering evaluation value.

[0072] Further, in the clustering process of the charging behavior of the electric vehicle, it is necessary to cluster the electric vehicles with similar distribution into a cluster, so the clustering result of a single electric vehicle not only depends on the data distribution of itself, but also depends on the distribution of the remaining electric vehicles adjacent to its access time.

[0073] Secondly, for the samples in the training set, the operation state of the power grid in adjacent time periods may be similar, and therefore, theoretically, in adjacent time periods, the charging behavior of a single sample and the remaining samples in the adjacent time period is relatively close. If there is a large difference between the charging behavior of a single sample and the remaining samples, it indicates that the specificity of the sample is high. Therefore, in the subsequent training process, a larger neighborhood radius needs to be used, which helps the neurons to update the weights in a larger range and learn more detailed information of the samples.

[0074] Based on the above analysis, the difference degree is calculated based on the difference between the charging and discharging power of the samples accessed in the local range in the training set, specifically as follows:

[0075] Taking the access time of each sample in the training set as the center, a local time period with a preset time length is constructed.

[0076] In this embodiment, the length of the local time period is 30 minutes. As other implementation manners, the implementer can set it according to the actual situation.

[0077] All samples with an access time in the local time period are obtained, denoted as adjacent samples.

[0078] The charging and discharging power of all samples at each time is normalized, and the mean of the difference between the normalized charging and discharging power of each sample and all adjacent samples at all times is calculated as the relative difference.

[0079] In this embodiment, the maximum and minimum normalization method is used for normalization, which is a known technology and will not be described here. As other implementation manners, the implementer can use other methods of existing technologies, such as Z-score standardization method, etc. This embodiment does not specially limit this. Secondly, the mean of the DTW distance of the normalized charging and discharging power between each sample and all adjacent samples at all times is calculated as the relative difference, wherein the calculation of the DTW distance is a known technology and will not be described here.

[0080] The number of all adjacent samples of each sample is counted and is positively mapped.

[0081] In this embodiment, the specific process of positive mapping is as follows: a logarithmic function is used for positive mapping. Assuming that the number is denoted as N, then As a result of positive mapping, wherein, represents the logarithmic function with 2 as the base.

[0082] The ratio between the relative difference and the result of positive mapping is taken as the difference degree of each sample in the training set.

[0083] It should be noted that the smaller the relative difference, the smaller the difference in charging and discharging power between the sample and its neighboring samples. The larger the positive mapping result, the higher the concentration of electric vehicles charging within the sample's charging segment. The smaller the difference, the more consistent its charging behavior is with the group of vehicles charging at the same time, and its charging behavior pattern is very typical, making it easy for clustering algorithms to identify and classify it into the corresponding cluster. Conversely, the smaller the difference, the higher the specificity of the electric vehicle's charging behavior, making it difficult to accurately classify it into any common category. It is necessary to increase the neighborhood radius of the SOM model to learn more detailed information in the sample.

[0084] Furthermore, based on the clustering evaluation value and the dissimilarity, an adjustment coefficient is calculated, specifically as follows:

[0085] The normalized result of the product of the cluster evaluation value and the dissimilarity is used as the adjustment coefficient for each sample in the training set.

[0086] In this embodiment, the sigmoid function is used for normalization. The sigmoid function is a well-known technique and will not be described in detail here.

[0087] It should be noted that the larger the adjustment coefficient, the higher the complexity and specificity of the charging behavior of the sample, and the more necessary it is to increase the neighborhood radius of the SOM model so that more neurons can participate in the learning process, thereby improving the model's adaptability to complex behaviors.

[0088] Furthermore, based on the adjustment coefficient, the neighborhood radius of the SOM model is adjusted as follows:

[0089]

[0090] in, For the training set Each sample corresponds to the adjusted neighborhood radius. The initial neighborhood radius is preset. For preset adjustment range, For the training set Adjustment factor for each sample;

[0091] In this embodiment, the preset initial neighborhood radius is set to 2. This determines the default neighborhood radius without considering sample complexity and variability. A smaller initial neighborhood radius means the model's influence on each sample is smaller in the early stages of training, focusing primarily on local features. In this embodiment, due to the large sample size, a larger grid size is needed to better capture the data distribution and features. Therefore, a large-size grid is used when setting the SOM grid structure. A larger grid can provide more neurons, thus mapping the data more finely. Specifically, with a grid size of 40, the maximum neighborhood radius is half the grid size, set to 20. This ensures... The value range is within 20, with a preset adjustment range. The value is 18, which is relatively large. It provides a large dynamic adjustment space, enabling the model to flexibly adjust the neighborhood radius according to the complexity and differences of the samples. As another implementation method, implementers can set it according to the actual situation.

[0092] It should be noted that the larger the adjustment coefficient, the larger the adjusted neighborhood radius of the sample. This will allow more neurons around the selected winning neuron to participate in the learning process in each iteration, enabling the model to perform extensive updates to the entire grid and ensure the capture of global features.

[0093] Thus, the adjusted neighborhood radius for each sample in the training set is obtained.

[0094] Step 4: Based on the adjusted neighborhood radius, iteratively train the classification model, evaluate the model using the test set to obtain the deployment model, classify the real-time charging information of electric vehicles, obtain the real-time schedulable resource capacity of each category, and use it as boundary conditions to establish and solve a multi-objective optimization function to adjust the output of distributed energy in the virtual power plant.

[0095] Furthermore, based on the adjusted neighborhood radius, the SOM model is used to train the training set, specifically as follows:

[0096] Based on the adjusted neighborhood radius corresponding to different samples in the training set, the training set is used as the input of the SOM model. The model is trained through multiple iterations, in which the adjusted neighborhood radius gradually decays in each iteration. The model after each iteration is used as a candidate model.

[0097] In this embodiment, the adjusted neighborhood radius is decayed using an exponential function with a decay rate of 2, i.e., the exponential decay formula is: ,in, When describing the m-th iteration, the first... Each sample corresponds to the adjusted neighborhood radius. For the first iteration Each sample corresponds to the adjusted neighborhood radius. The decay rate is set to 2; secondly, the learning rate of the SOM model is set to... Set the number of training rounds As another implementation method, the implementer can set it according to the actual situation.

[0098] It should be noted that the iterative training process of the SOM model is a well-known technique and will not be elaborated here.

[0099] Each candidate model is used to classify the test set. The Davies-Bouldin index is used to evaluate the clustering effect of each candidate model on the test set. The model with the smallest Davies-Bouldin index is selected as the deployment model.

[0100] It should be noted that the Davies-Bouldin index is a well-known technique and will not be elaborated upon here.

[0101] The deployment model is used to classify real-time charging information for electric vehicles and obtain multiple categories;

[0102] In this embodiment, the process of classifying real-time charging information of electric vehicles using a deployment model, namely the SOM model, is a well-known technique and will not be described in detail here.

[0103] Obtain the real-time rechargeable energy storage capacity, rechargeable power, dischargeable energy storage capacity, and dischargeable power of each electric vehicle under each category;

[0104] It should be noted that rechargeable energy storage capacity refers to how much electricity the electric vehicle can still be charged to reach full charge; dischargeable energy storage capacity refers to the electrical energy the electric vehicle can release from its minimum permissible charge level; and the rechargeable power is... ,in, For the first The rechargeable power of an electric vehicle at time t. For the first A number of electric vehicles Rechargeable energy storage capacity at any time For charging and discharging efficiency, The interval between adjacent time points. Maximum charging power, This expression describes the amount of charging power required from the grid to fully charge the electric vehicle in the next moment. Due to the limitation of the electric vehicle's rated charging power, the minimum of the two values ​​is selected. Correspondingly, the discharge power is... ,in, Let be the discharge power of the i-th electric vehicle at time t. For the first A number of electric vehicles Dischargeable energy storage capacity at any given time. For charging and discharging efficiency, For maximum discharge power, This indicates the minimum discharge power that can be provided to release the electric vehicle's charge in the next moment, chosen due to the limitation of the electric vehicle's rated discharge power.

[0105] In this embodiment, the charge / discharge efficiency The value is 0.92. As for other implementation methods, the implementer can set it according to the actual situation.

[0106] Using Minkowski summation, we obtain the real-time total rechargeable energy storage capacity, total rechargeable power, total discharge capacity, and total dischargeable power for each category.

[0107] It should be noted that Minkowski and his work are well-known technologies, and will not be elaborated upon here.

[0108] Furthermore, each type of electric vehicle is treated as a small energy storage unit. Based on other distributed energy sources, namely photovoltaic power generation, wind power generation, and thermal power generation, a multi-objective optimization function is established with the objectives of minimizing load variance, maximizing the absorption of photovoltaic and wind power, and minimizing the operating cost of thermal power generation. This function optimizes the output of wind power, photovoltaic power, thermal power, and electric vehicles. Specifically:

[0109] The objective function for minimizing the load variance is: ,in, ;

[0110] in, Net load, Let be the output power of the wind power generation at time t. Let be the output power of photovoltaic power generation at time t. Let be the output power of the thermal power plant at time t. For the first The output power of each category at time t, where L is the number of all categories. Let t be the normal load power at time t, which is the total power required for production and daily life, and T be all times, which means all times within a day, where T is 24 hours.

[0111] The objective function for minimizing the operating cost of thermal power generation is as follows: ;

[0112] in, This represents the cost coefficient for generating a unit of electricity during thermal power generation. In this embodiment, The value is 0.629. As for other implementation methods, the implementer can set it according to the actual situation.

[0113] The objective function for maximizing the photoelectric wind power consumption is: And the maximization of is converted into a minimization objective, so ;

[0114] Therefore, the final multi-objective optimization function is:

[0115]

[0116] The constraint conditions are:

[0117]

[0118] Wherein, is the minimum output power of thermal power generation, is the maximum output power of thermal power generation, , are the maximum output power of thermal power generation reduced and increased per unit time respectively, is the total dischargeable power of the th category at the th moment, is the total chargeable power of the th category at the th moment, is the power of the th category at the th moment, is the power of the th category at the th moment, is the minimum storage power of the th category, is the maximum storage power of the th category.

[0119] It should be noted that is the power balance constraint, which ensures that the load power and the supply-side power are balanced at a unit moment.

[0120] Solving the multi-objective optimization function outputs the corresponding real-time output power of photovoltaic power generation, wind power generation and thermal power generation. The virtual power plant adjusts the output power of the distributed energy according to the output result to ensure the coordinated operation of the power grid.

[0121] In this embodiment, the NSGA-II algorithm is used to solve the multi-objective optimization function, wherein the NSGA-II algorithm is a known technology and will not be described here. The maximum iteration number of the NSGA-II algorithm is set to 1000, the population size is 200, the mutation rate is 0.2, and the crossover rate is 0.8.

[0122] The embodiment of the present application further provides a virtual power plant coordinated operation device considering multi-energy complementation, wherein a computer program is stored, and the computer program is executed by a processor to realize the steps of any one of the above-mentioned virtual power plant coordinated operation methods considering multi-energy complementation.

[0123] Based on the same inventive concept as the above-mentioned method, the embodiment of the present application further provides a virtual power plant coordinated operation system considering multi-energy complementation, comprising a memory, a processor and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to realize the steps of any one of the above-mentioned virtual power plant coordinated operation methods considering multi-energy complementation.

[0124] It should be understood that, although Figure 1 The steps in the flowchart of the above-mentioned method are displayed in sequence according to the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, Figure 1 At least a part of the steps in the above-mentioned method can comprise a plurality of sub-steps or a plurality of stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with other steps or sub-steps or stages of other steps.

[0125] The technical features of the above-mentioned embodiments can be combined in any way, and to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described, but as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0126] The above-mentioned embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as the limitation of the present application. It should be pointed out that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, therefore, any simple modification, equivalent change and modification of the above-mentioned embodiments according to the technical essence of the present application, which does not depart from the technical solution of the present application, all belong to the protection scope of the technical solution of the present application.

Claims

1. A method for coordinating operation of a virtual power plant considering multi-energy complementarity, characterized in that, The method comprises the following steps: extracting charging information of different electric vehicles when charging from historical charging pile records, forming a data set, and dividing a training set and a test set, each record being taken as a sample, wherein the charging information comprises access time, charging and discharging state at each time and charging and discharging power; for the training set, analyzing fluctuation of the charging and discharging state and frequency of state change of each sample, calculating instability of each sample; based on cumulative deviation of the charging and discharging power of each sample, calculating a power boundary ratio of each sample, combining the instability to obtain a clustering evaluation value of each sample; determining difference of each sample through difference of the charging and discharging power between each sample and the remaining samples adjacent to the access time of each sample and the number of adjacent samples, combining the clustering evaluation value to adjust a neighborhood radius of the corresponding classification model of each sample in the training set, iteratively training the classification model, evaluating the model with the test set to obtain a deployment model, classifying real-time charging information of the electric vehicle, obtaining real-time schedulable resource capacity of each category, taking the same as a boundary condition, establishing a multi-objective optimization function and solving the same to adjust output of distributed energy in the virtual power plant; The iterative training of the classification model, the evaluation of the model with the test set and the obtaining of the deployment model comprise: based on the adjusted neighborhood radius of different samples in the training set, taking the training set as an input of a self-organizing mapping model, iteratively training the same for multiple times, wherein the adjusted neighborhood radius gradually decays in each iteration, and taking the model after completion of each iteration as a candidate model; classifying the test set with each candidate model, evaluating clustering effect of each candidate model on the test set by using a Davies-Bouldin index, and selecting a model with the smallest Davies-Bouldin index as the deployment model; The establishment of the multi-objective optimization function comprises: The objective function for minimizing the load variance is: , ,in, Net load, , , These represent the output power of wind power, photovoltaic power, and thermal power at time t, respectively. For the first The output power of each category at time t, where L is the number of all categories. Let be the normal load power at time t, and T be the load power at all times; The objective function of minimizing the operation cost of thermal power generation is established as: wherein, represents the cost coefficient of generating unit electric quantity in thermal power generation. The objective function of the maximum photoelectric wind power consumption is established as: .

2. The method of claim 1, wherein the multi-energy complementary virtual power plant coordinated operation method is characterized in that, The calculation of the instability of each sample comprises: For each sample in the training set, counting a number of inconsistent charging and discharging states of all adjacent two time points of each sample as a number of reversals of each sample; Calculating a discrete degree of the charging and discharging state of all time points of each sample; The instability is a product of the number of reversals and the discrete degree.

3. The method of claim 1, wherein the multi-energy complementary virtual power plant coordinated operation method is characterized by, The calculation of the power boundary ratio of each sample comprises: Based on the charging information of each sample in the training set, constructing a charging uplink boundary and a charging downlink boundary respectively; calculating a total area surrounded by the charging uplink boundary and the charging downlink boundary; Carrying out curve fitting on the charging and discharging power of all time points of each sample in the training set to obtain a fitting curve; taking the integral curve of the fitting curve with respect to time as an actual power curve; Calculating an area surrounded by the charging uplink boundary and the actual power curve, denoted as a deviation area; The power boundary ratio is a ratio of the deviation area to the total area.

4. The method of claim 1, wherein the multi-energy complementary virtual power plant coordinated operation method is characterized by, The clustering evaluation value is a product of the instability and the power boundary ratio.

5. The method of claim 1, wherein, The determination of the difference of each sample comprises: Taking the access time of each sample in the training set as a center, constructing a local period of a preset time length; obtaining all samples in the training set with the access time in the local period, denoted as adjacent samples; The charging and discharging power of all samples at each time is normalized, and the average of the difference between each sample and all its adjacent samples at all times corresponding to the normalized charging and discharging power is calculated as the relative difference; The number of all adjacent samples of each sample is counted and positively mapped; the ratio of the relative difference to the positively mapped result is taken as the difference degree of each sample in the training set.

6. The method for coordinated operation of a virtual power plant considering multi-energy complementarity according to claim 1, characterized in that, The first in the training set Each sample corresponds to the adjusted neighborhood radius. The calculation formula is: ,in, The initial neighborhood radius is preset. For preset adjustment range, For the training set The adjustment coefficient for each sample, wherein the adjustment coefficient is the normalized result of the product of the cluster evaluation value and the dissimilarity.

7. A virtual power plant coordinated operation device considering multi-energy complementation, wherein a computer program is stored in the device, and the device is characterized in that, The computer program, when executed by a processor, implements the steps of the method for coordinated operation of a virtual power plant considering multi-energy complementarity according to any one of claims 1-6. 8.A virtual power plant coordinated operation system considering multi-energy complementation, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, The processor executes the computer program to implement the steps of the method for coordinated operation of a virtual power plant considering multi-energy complementarity according to any one of claims 1-6.

Images