Method, device and system for predicting flexibility of electric vehicle type virtual power plant based on neural network, and medium

Abstract

The invention discloses a method for predicting the flexibility of an electric vehicle type virtual power plant based on a neural network. The method comprises the following steps: constructing a flexibility model of an electric vehicle; acquiring original data of the electric vehicle in the virtual power plant; inputting the original data of each electric vehicle into a flexibility model of the electric vehicle, and calculating to obtain the adjusting power and the adjusting electric quantity of each electric vehicle; taking the original data, the adjustment power and the adjustment electric quantity of each single electric vehicle as a sample library, and constructing a training set; inputting the training set into a prediction algorithm based on a convolutional neural network and a gate loop unit for training, and performing probability prediction on the flexibility of each single electric vehicle in combination with a quantile regression loss function; and aggregating the probability prediction results of the existing electric vehicles in the virtual power plant to obtain a flexibility probability prediction result of the electric vehicle type virtual power plant. The method can accurately predict the flexibility of the electric vehicle type virtual power plant under the influence of multivariate uncertain factors such as electricity prices and user willingness.

Description

Technical Field

[0001] The present invention relates to the field of power system prediction, and in particular to a method, device, system and medium for predicting the flexibility of an electric vehicle type virtual power plant based on a neural network. Background Art

[0002] In the context of energy transformation, the proportion of renewable energy such as wind and light is continuing to rise. Due to the uncertainty and volatility of their output, new challenges have been brought to the safe and stable operation of the power system. By the end of 2023, the number of new energy vehicles in my country will be 20.41 million, and it is expected that by 2030, the number of EVs will reach 80 million. As a load-side flexibility resource, electric vehicles can tap their own flexible adjustment potential by adjusting their own charging and discharging power. Although the flexible adjustment capacity of a single electric vehicle is relatively small, virtual power plants can provide greater flexibility to the power system by aggregating and optimizing electric vehicle clusters through advanced information and communication technologies. Since electric vehicle-based virtual power plants have both fast and flexible adjustment capabilities and superior economic efficiency, they are expected to achieve resource aggregation management, promote local consumption of new energy, and ensure safe and economic operation of the power grid in the process of participating in the power market and power grid operation.

[0003] Existing research mainly focuses on power demand forecasting, rather than directly predicting the flexibility of adjustable resources on the load side. However, flexibility forecasting is of unique importance to improving the flexibility and reliability of power systems. Flexibility forecasting has important applications in virtual power plants, demand response, renewable energy integration, and grid balancing. It can help power systems better cope with load fluctuations, optimize energy utilization, and increase the penetration of renewable energy. An accurate flexibility forecasting model can effectively reduce the impact of various uncertainties on the resource allocation of electric vehicle-based virtual power plants, thereby contributing to the improvement of virtual power plant operating efficiency and grid stability.

[0004] Existing flexibility prediction methods are also mainly based on point prediction. However, since electric vehicle charging behavior is affected by many factors such as electricity prices and user willingness, there are many uncertainties. Point prediction alone cannot accurately predict the flexibility of electric vehicle-based virtual power plants. Summary of the invention

[0005] The purpose of the present invention is to solve the deficiencies of the above-mentioned prior art, thereby providing a flexibility prediction method for electric vehicle type virtual power plants based on neural networks, which can accurately predict the flexibility of electric vehicle type virtual power plants under the influence of multiple uncertain factors.

[0006] The present invention also aims to provide a flexibility prediction device, system and medium for an electric vehicle type virtual power plant based on a neural network.

[0007] A flexibility prediction method for an electric vehicle type virtual power plant based on a neural network comprises the following steps:

[0008] Analyze the regulation power and regulation power characteristics of a single electric vehicle and construct the flexibility model of the electric vehicle;

[0009] Collect raw data of electric vehicles in virtual power plants;

[0010] Inputting the original data of each electric vehicle into the flexibility model of the electric vehicle to calculate the adjustment power and adjustment power of each electric vehicle;

[0011] The original data, regulated power and regulated electricity of each single electric vehicle are used as the sample library to construct a training set;

[0012] The training set is input into the prediction algorithm based on convolutional neural network and gated recurrent unit for training, and the probability prediction of the flexibility of each single electric vehicle is made in combination with the quantile regression loss function;

[0013] The probability prediction results of existing electric vehicles in the virtual power plant are aggregated to obtain the flexibility probability prediction results of the electric vehicle type virtual power plant.

[0014] The regulated power refers to the power that the electric vehicle can instantly increase or decrease, and the regulated power refers to the cumulative regulated power of the electric vehicle within a time interval.

[0015] The regulation power and regulation power characteristics of a single electric vehicle are analyzed, and the flexibility model of the electric vehicle is constructed. Specifically:

[0016] The power increase or decrease of a single electric vehicle is determined by the difference between the net load of the day-ahead dispatch plan and the maximum power increase or decrease, taking into account the charging and discharging efficiency:

[0017]

[0018]

[0019] Minimum battery level The following three constraints should be met:

[0020] Ensure that the battery power is not less than the battery power at the maximum discharge power, ensure that the battery power is not less than the minimum battery power required for departure, and the battery power is not less than the reserve power;

[0021]

[0022] When the demand response starts, the battery power is lower than the reserve power. Before the reserve power is reached, the battery is charged at the maximum charging power.

[0023] Maximum battery charge The following two constraints should be satisfied:

[0024] Ensure that the battery charge is not greater than the battery charge at the maximum charging power, and the battery charge cannot exceed the maximum capacity of the battery soc max ,

[0025]

[0026] According to the state at the time of departure, the downward adjustment ability of a single electric vehicle is determined by the difference between the current battery power and the lowest battery power that can be achieved, and the upward energy adjustment ability is determined by the difference between the current battery power and the highest battery power that can be achieved.

[0027]

[0028]

[0029] Where: They represent the upward and downward power regulation of the electric vehicle at time t, respectively. They represent the lower and upper limits of the electric vehicle's power regulation at time t, They represent the lower and upper limits of the battery power of the electric vehicle at time t, respectively, d , η c Respectively represent the discharge and charge efficiency, p m,DA (t) is the net load of electric vehicles in the day-ahead dispatch plan, t s ,t e , They represent the start and end time of demand response and the time when the electric vehicle leaves, and A(t) represents the time at [t s ,t e ]A collection of electric vehicles that are charged and discharged within a time interval, soc m,DA is the battery capacity of the electric vehicle in the day-ahead dispatch plan, Indicates the battery capacity. soc resv They represent the minimum battery capacity and reserve capacity of electric vehicles respectively. They represent the maximum discharge power and the maximum charging power respectively, m represents the mth vehicle, and t represents the t time.

[0030] The prediction algorithm based on convolutional neural network and gated recurrent unit is composed of convolutional neural network and gated recurrent unit;

[0031] The convolutional neural network contains three one-dimensional convolutional layers and one one-dimensional maximum pooling layer. The activation functions of the three one-dimensional convolutional layers are scaled exponential linear units.

[0032] The gated recurrent unit consists of a gated recurrent network and a fully connected layer.

[0033] The training set and test set are input into the prediction algorithm based on convolutional neural network and gated recurrent unit for training, and the probability prediction of the flexibility of each single electric vehicle is made in combination with the quantile regression loss function, specifically:

[0034] The training set is used as the input of a one-dimensional convolutional layer, followed by two other one-dimensional convolutional layers, and then a one-dimensional maximum pooling layer for training to obtain feature information;

[0035] The feature information output by the one-dimensional maximum pooling layer is used as the input of the gated recurrent network, and its output is passed through the fully connected layer and the sigmoid activation function to obtain the prediction result of the electric vehicle flexibility at the next moment;

[0036] Based on the prediction results of EV flexibility at the next moment, a quantile regression loss function is applied to generate a prediction interval from 5% to 95%.

[0037] The quantile regression loss function is as follows:

[0038]

[0039] Where: is the predicted value, y i is the true value, q is the quantile value between 0 and 1,

[0040] Among them, the q value determines the degree of penalty of the loss function for over-prediction and under-prediction. The larger the q value, the greater the penalty for over-prediction and the smaller the penalty for under-prediction.

[0041] Aggregate the probability prediction results of existing electric vehicles in the virtual power plant to obtain the flexibility probability prediction results of the electric vehicle type virtual power plant, specifically:

[0042] Aggregating the probability prediction results of the flexibility of each single electric vehicle in the virtual power plant, the total adjustment power and adjustment power formula are as follows:

[0043]

[0044]

[0045] In the formula, They represent the downward / upward power regulation and the lower / upper limit of the regulated power of the virtual power plant, Indicates the arrival time of the electric vehicle.

[0046] A flexibility prediction device for an electric vehicle type virtual power plant based on a neural network, comprising:

[0047] A model construction module, wherein the model construction module is used to analyze the regulation power and regulation power characteristics of a single electric vehicle and construct a flexibility model of the electric vehicle;

[0048] A data acquisition module, wherein the data acquisition module is used to collect raw data of electric vehicles in the virtual power plant;

[0049] A data processing module, wherein the data processing module is used to input the original data of each electric vehicle into the flexibility model of the electric vehicle to calculate the regulated power and regulated electricity of each electric vehicle; and to construct a training set using the original data, regulated power and regulated electricity of each single electric vehicle as a sample library;

[0050] A prediction module, wherein the prediction module is used to input the training set into a prediction algorithm based on a convolutional neural network and a gated recurrent unit for training, and to perform a probability prediction on the flexibility of each single electric vehicle in combination with a quantile regression loss function;

[0051] An aggregation module is used to aggregate the probability prediction results of existing electric vehicles in the virtual power plant to obtain the flexibility probability prediction results of the electric vehicle type virtual power plant.

[0052] The original data of electric vehicles include the time when the electric vehicle is connected and the battery power when it is connected, the time when the electric vehicle is driven away and the battery power when it is driven away, the battery capacity, and the maximum charge and discharge power.

[0053] The model construction module is specifically used to:

[0054] The power increase or decrease of a single electric vehicle is determined by the difference between the net load of the day-ahead dispatch plan and the maximum power increase or decrease, taking into account the charging and discharging efficiency:

[0055]

[0056]

[0057] Minimum battery level The following three constraints should be met:

[0058] Ensure that the battery power is not less than the battery power at the maximum discharge power, ensure that the battery power is not less than the minimum battery power required for departure, and the battery power is not less than the reserve power;

[0059]

[0060] When the demand response starts, the battery power is lower than the reserve power. Before the reserve power is reached, the battery is charged at the maximum charging power.

[0061] Maximum battery charge The following two constraints should be satisfied:

[0062] Ensure that the battery charge is not greater than the battery charge at the maximum charging power, and the battery charge cannot exceed the maximum capacity of the battery soc max ,

[0063]

[0064] According to the state at the time of departure, the downward adjustment ability of a single electric vehicle is determined by the difference between the current battery power and the lowest battery power that can be achieved, and the upward energy adjustment ability is determined by the difference between the current battery power and the highest battery power that can be achieved.

[0065]

[0066]

[0067] Where: They represent the upward and downward power regulation of the electric vehicle at time t, respectively. They represent the lower and upper limits of the electric vehicle's power regulation at time t, They represent the lower and upper limits of the battery power of the electric vehicle at time t, respectively, d , η c Respectively represent the discharge and charge efficiency, p m,DA (t) is the net load of electric vehicles in the day-ahead dispatch plan, t s ,t e , They represent the start and end time of demand response and the time when the electric vehicle leaves, and A(t) represents the time at [t s ,t e ]A collection of electric vehicles that are charged and discharged within a time interval, soc m,DA is the battery capacity of the electric vehicle in the day-ahead dispatch plan, Indicates the battery capacity. soc resv They represent the minimum battery capacity and reserve capacity of electric vehicles respectively. They represent the maximum discharge power and the maximum charging power respectively, m represents the mth vehicle, and t represents the t time.

[0068] The prediction module is specifically used for:

[0069] The training set is used as the input of a one-dimensional convolutional layer, followed by two other one-dimensional convolutional layers, and then a one-dimensional maximum pooling layer for training to obtain feature information;

[0070] The feature information output by the one-dimensional maximum pooling layer is used as the input of the gated recurrent network, and its output is passed through the fully connected layer and the sigmoid activation function to obtain the prediction result of the electric vehicle flexibility at the next moment;

[0071] Based on the prediction results of EV flexibility at the next moment, a quantile regression loss function is applied to generate a prediction interval from 5% to 95%.

[0072] The aggregation module is specifically used for:

[0073] Aggregating the probability prediction results of the flexibility of each single electric vehicle in the virtual power plant, the total adjustment power and adjustment power formula are as follows:

[0074]

[0075]

[0076] In the formula, They represent the downward / upward power regulation and the lower / upper limit of the regulated power of the virtual power plant, Indicates the arrival time of the electric vehicle.

[0077] A flexibility prediction system for an electric vehicle type virtual power plant based on a neural network, comprising: a computer readable storage medium and a processor;

[0078] The computer-readable storage medium is used to store executable instructions;

[0079] The processor is used to read the executable instructions stored in the computer-readable storage medium to execute the flexibility prediction method described in the above technical solution.

[0080] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the flexibility prediction method described in the above technical solution is implemented.

[0081] Compared with the prior art, the present invention has the following beneficial effects: the proposed probabilistic prediction algorithm combining convolutional neural network and gated recurrent unit, in which the convolutional neural network can extract useful information from highly uncertain power demand data, and the gated recurrent unit has advantages in time series prediction, and the prediction uncertainty is quantified by calculating the prediction interval through quantile regression. The proposed hybrid framework is superior to the traditional prediction neural network. Under the influence of multiple uncertain factors such as electricity prices and user willingness, the flexibility of electric vehicle-based virtual power plants can be accurately predicted. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 It is the overall flow chart of the present invention;

[0083] Figure 2 It is a system principle block diagram of the present invention. DETAILED DESCRIPTION

[0084] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0085] In a first aspect, an embodiment of the present invention provides a method for predicting flexibility of an electric vehicle type virtual power plant based on a neural network, such as Figure 1 As shown, the method includes:

[0086] Analyze the regulation power and regulation power characteristics of a single electric vehicle, and construct a flexibility model of the electric vehicle.

[0087] In some embodiments, the flexibility of an electric vehicle is often characterized by regulating power and regulating electricity. The regulating power refers to the power that the electric vehicle can instantly increase or decrease; the regulating electricity refers to the cumulative amount of electricity that the electric vehicle can regulate within a time interval.

[0088] Specifically, the power increase or decrease of a single electric vehicle is determined by the difference between the net load of the day-ahead scheduling plan and the maximum power increase or decrease, taking into account the charging and discharging efficiency.

[0089]

[0090]

[0091] Where: They represent the upward and downward power regulation of the electric vehicle at time t, respectively, d , η c Respectively represent the discharge and charge efficiency, p m,DA (t) is the net load of electric vehicles in the day-ahead dispatch plan, t s ,t e They represent the start and end time of demand response respectively, and A(t) represents the s ,t e ]A collection of electric vehicles that are charged and discharged within a time interval, Represent the maximum discharge power and maximum charging power respectively.

[0092] The regulation power of a single electric vehicle is affected by the battery power and the time when the electric vehicle left the vehicle. The downward regulation capacity is determined by the difference between the current battery power and the lowest battery power that can be achieved, and the upward energy regulation capacity is determined by the difference between the current battery power and the highest battery power that can be achieved. In reality, vehicles that have not left can continue to provide power regulation, while vehicles that have already left cannot participate in demand response, so their regulation power should be calculated based on the state when they left.

[0093]

[0094]

[0095] Where: They represent the lower and upper limits of the electric vehicle's power regulation at time t, They represent the lower and upper limits of the battery power of the electric vehicle at time t, Indicates the time when the electric vehicle leaves, soc m, DA is the battery capacity of the electric vehicle in the day-ahead dispatch plan, Indicates the battery capacity.

[0096] The highest battery charge that electric vehicles can currently achieve and minimum battery level It depends on the current battery power, charging and discharging capabilities, and the arrival and departure time of the vehicle. The maximum battery power should meet two constraints: ensure that the battery power is not greater than the battery power charged at the maximum charging power; the battery power cannot exceed the maximum capacity of the battery. The minimum battery power should meet three constraints: ensure that the battery is not lower than the battery power at the maximum discharge power; ensure that the battery power cannot be lower than the minimum battery power required when leaving; the battery power cannot be lower than the reserve power. If the battery power at the start of demand response is lower than the reserve power, it should be charged at the maximum charging power before the reserve power is reached.

[0097]

[0098]

[0099] Where: soc resv They represent the minimum battery capacity and reserve capacity of electric vehicles respectively.

[0100] Collect raw data of electric vehicles in virtual power plants;

[0101] The original data of each electric vehicle is input into the flexibility model of the electric vehicle to calculate the regulated power and regulated electricity of each electric vehicle, and the original data, regulated power and regulated electricity of each single electric vehicle are used as a sample library to construct a training set;

[0102] The training set input is based on a prediction algorithm that combines a convolutional neural network and a gated recurrent unit, combined with a quantile regression loss function to make probabilistic predictions of EV flexibility.

[0103] Aiming at the multivariate uncertainty problem in the prediction of electric vehicle flexibility, the present invention proposes a prediction algorithm combining convolutional neural network and gated recurrent unit, and combines quantile regression loss function to make probabilistic prediction of electric vehicle flexibility. By inputting information such as the access time and access battery power, departure time and departure battery power, battery capacity, and maximum charge and discharge power of the electric vehicle, the prediction interval of the electric vehicle's regulated power and regulated power is output.

[0104] In some embodiments, a prediction algorithm combining a convolutional neural network and a gated recurrent unit is designed to perform a probability prediction of electric vehicle flexibility in combination with a quantile regression loss function, wherein the prediction algorithm includes:

[0105] The prediction algorithm consists of a convolutional neural network and a gated recurrent unit. By constructing an accurate point prediction model, it attempts to find a single point prediction value that is as close to the true value as possible.

[0106] The convolutional neural network contains three one-dimensional convolutional layers and one one-dimensional maximum pooling layer, and uses the scaled exponential linear unit as the activation function of the three one-dimensional convolutional layers. The original data of electric vehicles is used as the input of the one-dimensional convolutional layer, followed by two one-dimensional convolutional layers and then a one-dimensional maximum pooling layer. This part mainly extracts features from the original electric vehicle data and extracts the most useful feature information.

[0107] Among them, there are two main benefits of choosing the scaled exponential linear unit as the activation function: compared with other activation functions, the scaled exponential linear unit has better convergence ability; secondly, it can avoid the problem of gradient disappearance. In addition, considering that the convolution layer can make neurons share weights and reduce the number of parameters, the pooling layer can ensure the translation invariance of the learned features.

[0108] The gated recurrent unit consists of a gated recurrent network and a fully connected layer. The feature information output by the one-dimensional maximum pooling layer is used as the input of the gated recurrent network, and its output is passed through the fully connected layer and the sigmoid activation function to obtain the prediction results of the electric vehicle's flexibility. This part is used to analyze the features extracted by the convolutional neural network and estimate the prediction results of the electric vehicle's regulated power and regulated power at the next moment.

[0109] Among them, compared with traditional recurrent neural networks, gated recurrent networks avoid the problem of gradient disappearance and are able to store time series data.

[0110] In some embodiments, a prediction algorithm combining a convolutional neural network and a gated recurrent unit is designed, and a quantile regression loss function is combined to perform a probability prediction of the flexibility of an electric vehicle, wherein the probability prediction includes:

[0111] Based on the above prediction results of the electric vehicle's regulated power and regulated power at the next moment, a 5% to 95% prediction interval is generated using the quantile regression loss function. The prediction interval is a quantification of the uncertainty of the prediction. It provides upper and lower bounds on the probability. The quantile refers to the proportion of observations in a group that are below a certain value. The quantile regression loss function is as follows:

[0112]

[0113] Where: is the predicted value, y i is the true value and q is the quantile value between 0 and 1.

[0114] Among them, the q value determines the degree of penalty imposed by the loss function on over-prediction (prediction value is higher than the true value) and under-prediction (prediction value is lower than the true value). The larger the value of q, the greater the penalty for over-prediction and the smaller the penalty for under-prediction. Traditional least squares regression assumes that the variance of the residual (the difference between the predicted value and the true value) is constant under all values ​​of the independent variable. This is often not true in reality because in many cases, the variance of the residual will change with the change of the independent variable. The quantile loss function does not require this assumption, so it can better handle the problem of non-constant residual variance.

[0115] Aggregate the probability prediction results of existing electric vehicles in the virtual power plant to form the flexibility probability prediction results of the electric vehicle type virtual power plant.

[0116] In some embodiments, based on the prediction results of the regulated electricity and regulated power of electric vehicles in the virtual power plant, the existing electric vehicles in the virtual power plant are aggregated, and the total regulated power and regulated electricity formulas are as follows:

[0117]

[0118]

[0119] In the formula, They represent the downward / upward power regulation and the lower / upper limit of the regulated power of the virtual power plant, Indicates the arrival time of the electric vehicle.

[0120] Among them, when calculating the amount of electricity regulated by the virtual power plant, the starting time must be determined based on the actual situation of each electric vehicle (i.e., the arrival time).

[0121] In a second aspect, the present invention provides a flexibility prediction device for an electric vehicle type virtual power plant based on a neural network, such as Figure 2 As shown, including:

[0122] A model construction module, wherein the model construction module is used to analyze the regulation power and regulation power characteristics of a single electric vehicle and construct a flexibility model of the electric vehicle;

[0123] The data acquisition module is used to collect the original data of electric vehicles in the virtual power plant; the original data of electric vehicles include the connection time and connection battery power of the electric vehicle, the departure time and departure battery power, battery capacity, and maximum charge and discharge power.

[0124] A data processing module, wherein the data processing module is used to input the original data of each electric vehicle into the flexibility model of the electric vehicle to calculate the regulated power and regulated electricity of each electric vehicle; and to construct a training set using the original data, regulated power and regulated electricity of each single electric vehicle as a sample library;

[0125] A prediction module, wherein the prediction module is used to input the training set into a prediction algorithm based on a convolutional neural network and a gated recurrent unit for training, and to perform a probability prediction on the flexibility of each single electric vehicle in combination with a quantile regression loss function;

[0126] An aggregation module is used to aggregate the probability prediction results of existing electric vehicles in the virtual power plant to obtain the flexibility probability prediction results of the electric vehicle type virtual power plant.

[0127] The model construction module is specifically used to:

[0128] The power increase or decrease of a single electric vehicle is determined by the difference between the net load of the day-ahead dispatch plan and the maximum power increase or decrease, taking into account the charging and discharging efficiency:

[0129]

[0130]

[0131] Minimum battery level The following three constraints should be met:

[0132] Ensure that the battery power is not less than the battery power at the maximum discharge power, ensure that the battery power is not less than the minimum battery power required for departure, and the battery power is not less than the reserve power;

[0133]

[0134] When the demand response starts, the battery power is lower than the reserve power. Before the reserve power is reached, the battery is charged at the maximum charging power.

[0135] Maximum battery charge The following two constraints should be satisfied:

[0136] Ensure that the battery charge is not greater than the battery charge at the maximum charging power, and the battery charge cannot exceed the maximum capacity of the battery soc max ,

[0137]

[0138] According to the state at the time of departure, the downward adjustment ability of a single electric vehicle is determined by the difference between the current battery power and the lowest battery power that can be achieved, and the upward energy adjustment ability is determined by the difference between the current battery power and the highest battery power that can be achieved.

[0139]

[0140]

[0141] Where: They represent the upward and downward power regulation of the electric vehicle at time t, respectively. They represent the lower and upper limits of the electric vehicle's power regulation at time t, They represent the lower and upper limits of the battery power of the electric vehicle at time t, respectively, d , η c Respectively represent the discharge and charge efficiency, p m,DA (t) is the net load of electric vehicles in the day-ahead dispatch plan, t s ,t e , They represent the start and end time of demand response and the time when the electric vehicle leaves, and A(t) represents the time at [t s ,t e ]A collection of electric vehicles that are charged and discharged within a time interval, soc m,DA is the battery capacity of the electric vehicle in the day-ahead dispatch plan, Indicates the battery capacity. soc resv They represent the minimum battery capacity and reserve capacity of electric vehicles respectively. They represent the maximum discharge power and the maximum charging power respectively, m represents the mth vehicle, and t represents the t time.

[0142] The prediction module is specifically used for:

[0143] The training set is used as the input of a one-dimensional convolutional layer, followed by two other one-dimensional convolutional layers, and then a one-dimensional maximum pooling layer for training to obtain feature information;

[0144] The feature information output by the one-dimensional maximum pooling layer is used as the input of the gated recurrent network, and its output is passed through the fully connected layer and the sigmoid activation function to obtain the prediction result of the electric vehicle flexibility at the next moment;

[0145] Based on the prediction results of EV flexibility at the next moment, a quantile regression loss function is applied to generate a prediction interval from 5% to 95%.

[0146] The aggregation module is specifically used for:

[0147] Aggregating the probability prediction results of the flexibility of each single electric vehicle in the virtual power plant, the total adjustment power and adjustment power formula are as follows:

[0148]

[0149]

[0150] In the formula, They represent the downward / upward power regulation and the lower / upper limit of the regulated power of the virtual power plant, Indicates the arrival time of the electric vehicle.

[0151] In a third aspect, the present invention provides a flexibility prediction system for an electric vehicle type virtual power plant based on a neural network, comprising: a computer readable storage medium and a processor;

[0152] The computer-readable storage medium is used to store executable instructions;

[0153] The processor is used to read the executable instructions stored in the computer-readable storage medium to execute the flexibility prediction method according to any one of claims 1 to 7.

[0154] In a fourth aspect, the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0155] The above are only preferred implementations of the present invention. It should be pointed out that a number of modifications and improved technical solutions made by those skilled in the art without departing from the technical solution should also be deemed to fall within the scope of protection required by the claims.

Claims

1. A flexibility prediction method for electric vehicle type virtual power plant based on neural network, characterized in that: The following steps are involved: Analyze the regulation power and regulation power characteristics of a single electric vehicle and construct the flexibility model of the electric vehicle; Collect raw data of electric vehicles in virtual power plants; Inputting the original data of each electric vehicle into the flexibility model of the electric vehicle to calculate the adjustment power and adjustment power of each electric vehicle; The training set is constructed using the original data, regulated power and regulated electricity of each single electric vehicle as the sample library; The training set is input into the prediction algorithm based on convolutional neural network and gated recurrent unit for training, and the quantile regression loss function is used to make probability predictions on the flexibility of each single electric vehicle. The probability prediction results of existing electric vehicles in the virtual power plant are aggregated to obtain the flexibility probability prediction results of the electric vehicle type virtual power plant.

2. The flexibility prediction method of electric vehicle type virtual power plant based on neural network as claimed in claim 1 is characterized in that: The regulated power refers to the power that the electric vehicle can instantly adjust up or down, and the regulated power refers to the cumulative regulated power of the electric vehicle within a time interval.

3. The flexibility prediction method of electric vehicle type virtual power plant based on neural network as claimed in claim 1 is characterized in that: The regulation power and regulation power characteristics of a single electric vehicle are analyzed, and the flexibility model of the electric vehicle is constructed. Specifically: The power increase or decrease of a single electric vehicle is determined by the difference between the net load of the day-ahead dispatch plan and the maximum power increase or decrease, taking into account the charging and discharging efficiency: Minimum battery level The following three constraints should be met: Ensure that the battery power is not less than the battery power at the maximum discharge power, ensure that the battery power is not less than the minimum battery power required for departure, and the battery power is not less than the reserve power; When the demand response starts, the battery capacity is lower than the reserve capacity. Before the reserve capacity is reached, the battery is charged at the maximum charging power. Maximum battery charge The following two constraints should be satisfied: Ensure that the battery charge is not greater than the battery charge at the maximum charging power, and the battery charge cannot exceed the maximum capacity of the battery soc max , According to the state at the time of departure, the downward adjustment ability of a single electric vehicle is determined by the difference between the current battery power and the lowest battery power that can be achieved, and the upward energy adjustment ability is determined by the difference between the current battery power and the highest battery power that can be achieved. Where: They represent the upward and downward power regulation of the electric vehicle at time t, respectively. They represent the lower and upper limits of the amount of electricity that the electric vehicle can regulate at time t, They represent the lower and upper limits of the battery power of the electric vehicle at time t, respectively, d , η c Respectively represent the discharge and charge efficiency, p m,DA (t) is the net load of electric vehicles in the day-ahead dispatch plan, t s ,t e , They represent the start and end time of demand response and the time when the electric vehicle leaves, and A(t) represents the time at [t s ,t e ]A collection of electric vehicles that are charged and discharged within a time interval, soc m,DA is the battery capacity of the electric vehicle in the day-ahead dispatch plan, Indicates the battery capacity. soc resv They represent the minimum battery capacity and reserve capacity of electric vehicles respectively. They represent the maximum discharge power and the maximum charging power respectively, m represents the mth vehicle, and t represents time t.

4. The method for predicting flexibility of an electric vehicle type virtual power plant based on a neural network as claimed in claim 1, characterized in that: The prediction algorithm based on convolutional neural network and gated recurrent unit is composed of convolutional neural network and gated recurrent unit; The convolutional neural network contains three one-dimensional convolutional layers and one one-dimensional maximum pooling layer. The activation functions of the three one-dimensional convolutional layers are scaled exponential linear units. The gated recurrent unit consists of a gated recurrent network and a fully connected layer.

5. The method for predicting flexibility of an electric vehicle type virtual power plant based on a neural network as claimed in claim 1 or 4, characterized in that: The training set and test set are input into the prediction algorithm based on convolutional neural network and gated recurrent unit for training, and the probability prediction of the flexibility of each single electric vehicle is made in combination with the quantile regression loss function, specifically: The training set is used as the input of a one-dimensional convolutional layer, followed by two other one-dimensional convolutional layers, and then a one-dimensional maximum pooling layer for training to obtain feature information; The feature information output by the one-dimensional maximum pooling layer is used as the input of the gated recurrent network, and its output is passed through the fully connected layer and the sigmoid activation function to obtain the prediction result of the electric vehicle flexibility at the next moment; Based on the prediction results of EV flexibility at the next moment, a quantile regression loss function is applied to generate a prediction interval from 5% to 95%.

6. The method for predicting flexibility of an electric vehicle type virtual power plant based on a neural network as claimed in claim 5, characterized in that: The quantile regression loss function is as follows: Where: is the predicted value, y i is the true value, q is the quantile value between 0 and 1, Among them, the q value determines the degree of penalty of the loss function for over-prediction and under-prediction. The larger the q value, the greater the penalty for over-prediction and the smaller the penalty for under-prediction.

7. The method for predicting flexibility of an electric vehicle type virtual power plant based on a neural network as claimed in claim 1, characterized in that: Aggregate the probability prediction results of existing electric vehicles in the virtual power plant to obtain the flexibility probability prediction results of the electric vehicle type virtual power plant, specifically: Aggregating the probability prediction results of the flexibility of each single electric vehicle in the virtual power plant, the total adjustment power and adjustment power formula are as follows: In the formula, They represent the downward / upward power regulation and the lower / upper limit of the regulated power of the virtual power plant, Indicates the arrival time of the electric vehicle.

8. A flexibility prediction device for an electric vehicle type virtual power plant based on a neural network, characterized in that: include: A model construction module, wherein the model construction module is used to analyze the regulation power and regulation power characteristics of a single electric vehicle and construct a flexibility model of the electric vehicle; A data acquisition module, wherein the data acquisition module is used to collect raw data of electric vehicles in the virtual power plant; A data processing module, wherein the data processing module is used to input the original data of each electric vehicle into the flexibility model of the electric vehicle to calculate the regulated power and regulated electricity of each electric vehicle; and to construct a training set using the original data, regulated power and regulated electricity of each single electric vehicle as a sample library; A prediction module, wherein the prediction module is used to input the training set into a prediction algorithm based on a convolutional neural network and a gated recurrent unit for training, and to perform a probability prediction on the flexibility of each single electric vehicle in combination with a quantile regression loss function; An aggregation module is used to aggregate the probability prediction results of existing electric vehicles in the virtual power plant to obtain the flexibility probability prediction results of the electric vehicle type virtual power plant.

9. The flexibility prediction device of an electric vehicle type virtual power plant based on a neural network according to claim 8 is characterized in that: The model construction module is specifically used to: The power increase or decrease of a single electric vehicle is determined by the difference between the net load of the day-ahead dispatch plan and the maximum power increase or decrease, taking into account the charging and discharging efficiency: Minimum battery level The following three constraints should be met: Ensure that the battery power is not less than the battery power at the maximum discharge power, ensure that the battery power is not less than the minimum battery power required for departure, and the battery power is not less than the reserve power; When the demand response starts, the battery capacity is lower than the reserve capacity. Before the reserve capacity is reached, the battery is charged at the maximum charging power. Maximum battery charge The following two constraints should be satisfied: Ensure that the battery charge is not greater than the battery charge at the maximum charging power, and the battery charge cannot exceed the maximum capacity of the battery soc max , According to the state at the time of departure, the downward adjustment ability of a single electric vehicle is determined by the difference between the current battery power and the lowest battery power that can be achieved, and the upward energy adjustment ability is determined by the difference between the current battery power and the highest battery power that can be achieved. Where: They represent the upward and downward power regulation of the electric vehicle at time t, respectively. They represent the lower and upper limits of the amount of electricity that the electric vehicle can regulate at time t, They represent the lower and upper limits of the battery power of the electric vehicle at time t, respectively, d , η c Respectively represent the discharge and charge efficiency, p m,DA (t) is the net load of electric vehicles in the day-ahead dispatch plan, t s ,t e , They represent the start and end time of demand response and the time when the electric vehicle leaves, and A(t) represents the time at [t s ,t e ]A collection of electric vehicles that are charged and discharged within a time interval, soc m,DA is the battery capacity of the electric vehicle in the day-ahead dispatch plan, Indicates the battery capacity. soc resv They represent the minimum battery capacity and reserve capacity of electric vehicles respectively. They represent the maximum discharge power and the maximum charging power respectively, m represents the mth vehicle, and t represents time t.

10. The flexibility prediction device of an electric vehicle type virtual power plant based on a neural network according to claim 8, characterized in that: The prediction module is specifically used for: The training set is used as the input of a one-dimensional convolutional layer, followed by two other one-dimensional convolutional layers, and then a one-dimensional maximum pooling layer for training to obtain feature information; The feature information output by the one-dimensional maximum pooling layer is used as the input of the gated recurrent network, and its output is passed through the fully connected layer and the sigmoid activation function to obtain the prediction result of the electric vehicle flexibility at the next moment; Based on the prediction results of EV flexibility at the next moment, a quantile regression loss function is applied to generate a prediction interval from 5% to 95%.

11. The flexibility prediction device of an electric vehicle type virtual power plant based on a neural network according to claim 8, characterized in that: The aggregation module is specifically used for: Aggregating the probability prediction results of the flexibility of each single electric vehicle in the virtual power plant, the total adjustment power and adjustment power formula are as follows: In the formula, They represent the downward / upward power regulation and the lower / upper limit of the regulated power of the virtual power plant, Indicates the arrival time of the electric vehicle.

12. A flexibility prediction system for electric vehicle type virtual power plant based on neural network, characterized in that: include: A computer readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is used to read the executable instructions stored in the computer-readable storage medium to execute the flexibility prediction method according to any one of claims 1 to 7.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the flexibility prediction method according to any one of claims 1 to 7 is implemented.

Images