Virtual power plant regulation capability online detection method, system, equipment and medium

Abstract

The invention discloses a virtual power plant regulation capability online detection method, system and device, and a medium. The virtual power plant regulation capability online detection method comprises the following steps: collecting operation data of a virtual power plant and an agent user, and carrying out normalization processing; constructing and training a convolutional neural network model, and extracting key features of the operation data of the virtual power plant and the proxy user; constructing and training a virtual power plant regulation capability detection model based on a deep belief neural network, and predicting the confidence of the key features; inputting a daily operation curve of a to-be-detected virtual power plant, performing normalization processing, extracting key features through the convolutional neural network model, predicting confidence of the key features through the virtual power plant regulation capability detection model, and comparing the confidence with a preset confidence threshold to obtain a detection result of the virtual power plant regulation capability. According to the invention, the purpose of online detection of the adjustment capability of the virtual power plant is realized, the detection time limit is reduced, the manual operation is reduced, the detection report is output online, and the detection efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of virtual power plant regulation capability detection, and in particular to an online detection method, system, equipment and medium for virtual power plant regulation capability. Background Art

[0002] A virtual power plant is a smart energy system that aggregates demand-side resources such as adjustable loads, distributed power sources, and customer-side energy storage through advanced digital technology, control technology, Internet of Things technology, and information and communication technology, and conducts unified coordinated management and optimized scheduling. Virtual power plants can support demand-side management, enhance the grid's regulation capabilities, and promote the consumption of new energy. Virtual power plants need to undergo the first regulation capacity test, and only after all test indicators meet local policy regulations can they participate in demand response, ancillary services, and the spot market. During the operation of a virtual power plant, if there are changes in important factors or before the peak summer, the regulation capacity needs to be tested again.

[0003] At present, the regulation capacity detection of virtual power plants usually adopts the method of calculating regulation indicators one by one. During the detection period, the detection agency issues the corresponding detection and adjustment instructions, and needs to manually and accurately record the sampling time, power and other data of the regulation process, and calculate the regulation capacity detection items one by one according to the established formula, and compare them with the uploaded regulation capacity indicators one by one to verify whether the regulation capacity meets the standards. Conventional detection methods have disadvantages such as low detection efficiency and limited detection time. Summary of the invention

[0004] Purpose of the invention: The present invention aims to provide an efficient online detection method for the regulation capability of a virtual power plant based on a deep belief neural network. Another objective of the present invention is to provide an online detection system, equipment and medium for the regulation capability of a virtual power plant based on a deep belief neural network.

[0005] Technical solution: The online detection method of the regulation capability of a virtual power plant according to the present invention comprises the following steps:

[0006] Collect the operating data of virtual power plants and proxy users and perform min-max normalization processing;

[0007] Build and train a convolutional neural network model to extract key features of the operating data of virtual power plants and proxy users;

[0008] A virtual power plant regulation capability detection model is built and trained based on a deep belief neural network to predict the confidence of key features;

[0009] Preset the credibility threshold of the running data under each detection event;

[0010] The daily operation curve of the virtual power plant to be inspected is input and normalized, the key features are extracted through the convolutional neural network model, the confidence of the key features is predicted by the virtual power plant regulation capacity detection model, and it is compared with the credibility threshold to obtain the detection result of the virtual power plant regulation capacity.

[0011] Furthermore, the convolutional layer of the convolutional neural network model is

[0012]

[0013] Among them, the input signal u is the actual operation curve of the virtual power plant detection day, the weight of the convolution kernel v uses a zero-mean Gaussian distribution, (u*v)(i) represents the value of the convolution of the signal u and v at index i; uflip is the result of flipping the input signal u along each axis, * represents the convolution operation, u m is the value of signal u at index m, v i-m is the value of v at index im;

[0014] The activation function is the Sigmoid function;

[0015] The pooling layer uses the maximum pooling operation;

[0016] The formula of the fully connected layer is

[0017] z=w·x+b

[0018] Among them, z is the output vector; w is the weight matrix, which represents the parameters of the fully connected layer, and its dimension is n×m, where n is the number of output features and m is the number of input features; x is the input vector, which is the output of the previous layer, and its dimension is m; b is the bias vector, and its dimension is n.

[0019] Furthermore, the steps of constructing and training a virtual power plant regulation capability detection model based on a deep belief neural network are as follows:

[0020] The key features of the operation data of the virtual power plant and proxy users extracted by the convolutional neural network model are standardized:

[0021]

[0022] When denormalizing:

[0023]

[0024] Among them, x i is the corresponding standardized data; z i is the denormalized data; ω2 and ω1 correspond to the upper and lower boundaries respectively, They correspond to the maximum and minimum values ​​of the characteristic data respectively;

[0025] Initialize the weights and thresholds of the deep belief neural network, set the number of RBM training layers in the deep belief neural network, and perform RBM pre-training;

[0026] Select the ReLU function as the activation function of the deep belief network RBM:

[0027] ReLU(x)=max(0,x)

[0028] When x≥0, output x itself; when x<0, output 0;

[0029] The output layer uses the softmax function to calculate the probability distribution softmax(s i )for

[0030]

[0031] s i is the value of the i-th element in the input vector s; S is the total number of categories, For i from 1 to S and.

[0032] Furthermore, the key features of the operating data of virtual power plants and proxy users include the 96-point operating curve of virtual power plants, the 96-point operating curve of proxy users, and the regulation capacity, regulation accuracy, duration, response time and data delay time reported by virtual power plants and proxy users.

[0033] The virtual power plant regulation capability online detection system of the present invention comprises:

[0034] Data collection module, used to collect operating data of virtual power plants and proxy users;

[0035] A data input module is used to input the operating data of the virtual power plant and the proxy user and the daily operating curve of the virtual power plant to be inspected;

[0036] Normalization module, used to perform min-max normalization processing on the data and curves input by the data input module;

[0037] Convolutional neural network model module, used to build and train convolutional neural network models to extract key features of the operating data of virtual power plants and proxy users;

[0038] A virtual power plant regulation capability detection model module is used to build and train a virtual power plant regulation capability detection model based on a deep belief neural network to predict the confidence of key features;

[0039] The online detection report output module is used to output the detection results of the virtual power plant's regulation capability.

[0040] Furthermore, the convolutional layer of the convolutional neural network model of the convolutional neural network model module is

[0041]

[0042] Among them, the input signal u is the actual operation curve of the virtual power plant detection day, the weight of the convolution kernel v uses a zero-mean Gaussian distribution, (u*v)(i) represents the value of the convolution of the signal u and v at index i; uflip is the result of flipping the input signal u along each axis, * represents the convolution operation, u m is the value of signal u at index m, v i-m is the value of v at index im.

[0043] The activation function is the Sigmoid function;

[0044] The pooling layer uses the maximum pooling operation;

[0045] The formula of the fully connected layer is

[0046] z=w·x+b

[0047] Among them, z is the output vector; w is the weight matrix, which represents the parameters of the fully connected layer, and its dimension is n×m, where n is the number of output features and m is the number of input features; x is the input vector, which is the output of the previous layer, and its dimension is m; b is the bias vector, and its dimension is n.

[0048] Furthermore, the steps of constructing and training a virtual power plant regulation capability detection model based on a deep belief neural network are as follows:

[0049] The key features of the operation data of the virtual power plant and proxy users extracted by the convolutional neural network model are standardized:

[0050]

[0051] When denormalizing:

[0052]

[0053] Among them, x i is the corresponding standardized data; z i is the denormalized data; ω2 and ω1 correspond to the upper and lower boundaries respectively, They correspond to the maximum and minimum values ​​of the characteristic data respectively;

[0054] Initialize the weights and thresholds of the deep belief neural network, set the number of RBM training layers in the deep belief neural network, and perform RBM pre-training;

[0055] Select the ReLU function as the activation function of the deep belief network RBM:

[0056] ReLU(x)=max(0,x)

[0057] When x≥0, output x itself; when x<0, output 0;

[0058] The output layer uses the softmax function to calculate the probability distribution softmax(s i )for

[0059]

[0060] s i is the value of the i-th element in the input vector s; S is the total number of categories, For i from 1 to S and.

[0061] Furthermore, the key features of the operating data of virtual power plants and proxy users include the 96-point operating curve of virtual power plants, the 96-point operating curve of proxy users, and the regulation capacity, regulation accuracy, duration, response time and data delay time reported by virtual power plants and proxy users.

[0062] The computer device of the present invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the steps of the above method are implemented when the processor executes the computer program.

[0063] The computer-readable storage medium of the present invention stores a computer program thereon, and the computer program implements the steps of the above method when executed by a processor.

[0064] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: 1. The present invention extracts key features from the operating data of virtual power plants and proxy users through convolutional neural networks, inputs the virtual power plant regulation capacity detection model built and trained based on deep confidence neural networks, predicts the confidence of key features, and obtains the detection results of the regulation capacity of the virtual power plant by comparing the confidence with the credibility threshold; 2. The present invention achieves the purpose of online detection of the regulation capacity of virtual power plants, alleviates the detection period restrictions, reduces manual operations, outputs detection reports online, and improves detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a flow chart of the online detection method of the regulation capability of a virtual power plant of the present invention;

[0066] Figure 2 It is a schematic diagram of the structure of a convolutional neural network;

[0067] Figure 3It is a structural schematic diagram of the virtual power plant regulation capacity detection model;

[0068] Figure 4 This is a structural diagram of the online detection system for the regulation capacity of the virtual power plant. DETAILED DESCRIPTION

[0069] The present invention will be further described below in conjunction with the accompanying drawings.

[0070] The online detection method of the regulation capability of a virtual power plant according to the present invention comprises the following steps:

[0071] Step S1, collecting actual operation data of the virtual power plant on the day of inspection.

[0072] Step S2: preprocess the data using the mix-max standardization method:

[0073]

[0074] The mix-max normalization method is used to normalize the virtual power plant operation data to the numerical range of [0,1].

[0075] Step S3, using the convolutional neural network (CNN) algorithm to extract the key features of the operation curves of the virtual power plant and the proxy user, first select the convolutional layer formula:

[0076]

[0077] Among them, the input signal u is the actual operation curve of the virtual power plant detection day, the weight of the convolution kernel v uses a zero-mean Gaussian distribution, (u*v)(i) represents the value of the convolution of the signal u and v at index i; uflip is the result of flipping the input signal u along each axis, * represents the convolution operation, u m is the value of signal u at index m, v i-m is the value of v at index im.

[0078] Since the convolutional neural network algorithm is used to extract features, the activation function uses the Sigmoid function, which will be classified according to multiple labels:

[0079]

[0080] z is the input value, and e is the base of the natural logarithm. To retain important features, the pooling layer uses the maximum pooling operation. The formula for the fully connected layer is:

[0081] z=w·x+b

[0082] z is the output vector, which is the original output of the fully connected layer (without activation function processing). w is the weight matrix, which represents the parameters of the fully connected layer, and its dimension is n×m, where n is the number of output features and m is the number of input features. x is the input vector, which is the output of the previous layer, and its dimension is m. b is the bias vector, which is also a parameter, and its dimension is n. The output results are used as the key features of the operating data of the virtual power plant and the proxy users.

[0083] Step S4, using a deep confidence neural network to evaluate the confidence of the predicted operating data results, includes:

[0084] Determine the sample set required for the deep belief neural network, including the 96-point operating curve of the virtual power plant output by the CNN model, the 96-point operating curve of the proxy user, and key features such as the regulation capacity, regulation accuracy, duration, response time, and data delay reported by the virtual power plant and proxy users.

[0085] A mathematical model for detecting the regulation capacity of virtual power plants is established using a deep belief neural network. The key features of the CNN model output are imported into the deep belief neural network as training samples. A deep belief neural network training model is established to train the training samples and the parameters of the model are continuously adjusted to obtain the best training model.

[0086] Establishing a deep belief neural network training model to learn and train the training samples includes the following steps:

[0087] The virtual power plant operation characteristics extracted by the CNN model are standardized, and the formula is:

[0088]

[0089] The denormalization formula is:

[0090]

[0091] In the formula, x i is the corresponding standardized data; z i is the denormalized data; ω2 and ω1 correspond to the upper and lower boundaries respectively, They correspond to the maximum and minimum values ​​of the feature data respectively.

[0092] Initialize the parameters of the deep belief neural network, including weights, thresholds, and RBM training layers, and perform RBM pre-training;

[0093] In order to effectively alleviate the gradient vanishing problem and accelerate the network training process, the ReLU function is selected as the activation function of the deep belief mental network RBM. The formula is:

[0094] ReLU(x)=max(0,x)

[0095] When x ≥ 0, output x itself; when x < 0, output 0. Ensure that the output result remains linear in the positive interval and the output is constant at 0 in the negative interval.

[0096] The output layer uses the softmax function to calculate the probability distribution, and the calculation formula is:

[0097]

[0098] s i is the value of the i-th element in the input vector s; S is the total number of categories, For i from 1 to S Sum, ensure that the sum of the output probabilities is 1. The output result is the confidence level of the operating data results of each detection indicator of the virtual power plant.

[0099] Step S5, preset the credibility threshold of the operating data results under each detection indicator of the virtual power plant. When the confidence output of the deep belief network for the operating curve is not within the preset credibility threshold, it is considered that the detection content does not meet the standard. Otherwise, the detection content meets the standard.

[0100] Step S6, input the daily operation curve of the virtual power plant to be tested, and based on the comparison result between the confidence output by the deep belief network and the preset credibility threshold, online detect the regulation capacity of the virtual power plant and output the test result.

[0101] The virtual power plant regulation capability online detection system of the present invention comprises:

[0102] Data collection module, used to collect operating data of virtual power plants and proxy users;

[0103] A data input module is used to input the operating data of the virtual power plant and the proxy user and the daily operating curve of the virtual power plant to be inspected;

[0104] Normalization module, used to perform min-max normalization processing on the data and curves input by the data input module;

[0105] Convolutional neural network model module, used to build and train convolutional neural network models to extract key features of the operating data of virtual power plants and proxy users;

[0106] A virtual power plant regulation capability detection model module is used to build and train a virtual power plant regulation capability detection model based on a deep belief neural network to predict the confidence of key features;

[0107] The online detection report output module is used to output the detection results of the virtual power plant's regulation capability.

[0108] Furthermore, the convolutional layer of the convolutional neural network model of the convolutional neural network model module is

[0109]

[0110] Among them, the input signal u is the actual operation curve of the virtual power plant detection day, the weight of the convolution kernel v uses a zero-mean Gaussian distribution, (u*v)(i) represents the value of the convolution of the signal u and v at index i; uflip is the result of flipping the input signal u along each axis, * represents the convolution operation, u m is the value of signal u at index m, v i-m is the value of v at index im.

[0111] The activation function is the Sigmoid function;

[0112] The pooling layer uses the maximum pooling operation;

[0113] The formula of the fully connected layer is

[0114] z=w·x+b

[0115] Among them, z is the output vector; w is the weight matrix, which represents the parameters of the fully connected layer, and its dimension is n×m, where n is the number of output features and m is the number of input features; x is the input vector, which is the output of the previous layer, and its dimension is m; b is the bias vector, and its dimension is n.

[0116] Furthermore, the steps of constructing and training a virtual power plant regulation capability detection model based on a deep belief neural network are as follows:

[0117] The key features of the operation data of the virtual power plant and proxy users extracted by the convolutional neural network model are standardized:

[0118]

[0119] When denormalizing:

[0120]

[0121] Among them, x i is the corresponding standardized data; z i is the denormalized data; ω2 and ω1 correspond to the upper and lower boundaries respectively, They correspond to the maximum and minimum values ​​of the characteristic data respectively;

[0122] Initialize the weights and thresholds of the deep belief neural network, set the number of RBM training layers in the deep belief neural network, and perform RBM pre-training;

[0123] Select the ReLU function as the activation function of the deep belief network RBM:

[0124] ReLU(x)=max(0,x)

[0125] When x≥0, output x itself; when x<0, output 0;

[0126] The output layer uses the softmax function to calculate the probability distribution softmax(s i )for

[0127]

[0128] s i is the value of the i-th element in the input vector s; S is the total number of categories, For i from 1 to S and.

[0129] Furthermore, the key features of the operating data of virtual power plants and proxy users include the 96-point operating curve of virtual power plants, the 96-point operating curve of proxy users, and the regulation capacity, regulation accuracy, duration, response time and data delay time reported by virtual power plants and proxy users.

[0130] The computer device of the present invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the steps of the above method are implemented when the processor executes the computer program.

[0131] The computer-readable storage medium of the present invention stores a computer program thereon, and the computer program implements the steps of the above method when executed by a processor.

Claims

1. A method for online detection of regulation capability of a virtual power plant, characterized in that: The following steps are involved: Collect the operating data of virtual power plants and proxy users and perform min-max normalization processing; Build and train a convolutional neural network model to extract key features of the operating data of virtual power plants and proxy users; A virtual power plant regulation capability detection model is built and trained based on a deep belief neural network to predict the confidence of key features; Preset the credibility threshold of the running data under each detection event; The daily operation curve of the virtual power plant to be inspected is input and normalized, the key features are extracted through the convolutional neural network model, the confidence of the key features is predicted by the virtual power plant regulation capacity detection model, and it is compared with the credibility threshold to obtain the detection result of the virtual power plant regulation capacity.

2. The online detection method for regulating capability of a virtual power plant according to claim 1 is characterized in that: The convolutional layer of the convolutional neural network model is Among them, the input signal u is the actual operation curve of the virtual power plant detection day, the weight of the convolution kernel v uses a zero-mean Gaussian distribution, (u*v)(i) represents the value of the convolution of the signal u and v at index i; uflip is the result of flipping the input signal u along each axis, * represents the convolution operation, u m is the value of signal u at index m, v i-m is the value of v at index im; The activation function is the Sigmoid function; The pooling layer uses the maximum pooling operation; The formula for the fully connected layer is z = w x + b Among them, z is the output vector; w is the weight matrix, which represents the parameters of the fully connected layer, and its dimension is n×m, where n is the number of output features and m is the number of input features; x is the input vector, which is the output of the previous layer, and its dimension is m; b is the bias vector, and its dimension is n.

3. The online detection method for regulating capability of a virtual power plant according to claim 2 is characterized in that: The steps to build and train a virtual power plant regulation capability detection model based on a deep belief neural network are as follows: The key features of the operation data of the virtual power plant and proxy users extracted by the convolutional neural network model are standardized: When denormalizing: Among them, x i is the corresponding standardized data; z i is the denormalized data; ω2 and ω1 correspond to the upper and lower boundaries respectively, They correspond to the maximum and minimum values ​​of the characteristic data respectively; Initialize the weights and thresholds of the deep belief neural network, set the number of RBM training layers in the deep belief neural network, and perform RBM pre-training; Select the ReLU function as the activation function of the deep belief network RBM: ReLU(x)=max(0,x) When x≥0, output x itself; when x<0, output 0; The output layer uses the softmax function to calculate the probability distribution softmax(s i )for s i is the value of the i-th element in the input vector s; S is the total number of categories, For i from 1 to S and.

4. The online detection method for regulating capability of a virtual power plant according to claim 3 is characterized in that: The key features of the operating data of virtual power plants and proxy users include the 96-point operating curve of virtual power plants, the 96-point operating curve of proxy users, the regulation capacity, regulation accuracy, duration, response time and data delay time reported by virtual power plants and proxy users.

5. An online detection system for regulating capacity of a virtual power plant, characterized in that: include: Data collection module, used to collect operating data of virtual power plants and proxy users; A data input module is used to input the operating data of the virtual power plant and the proxy user and the daily operating curve of the virtual power plant to be inspected; Normalization module, used to perform min-max normalization processing on the data and curves input by the data input module; Convolutional neural network model module, used to build and train convolutional neural network models to extract key features of the operating data of virtual power plants and proxy users; A virtual power plant regulation capability detection model module is used to build and train a virtual power plant regulation capability detection model based on a deep belief neural network to predict the confidence of key features; The online detection report output module is used to output the detection results of the virtual power plant's regulation capability.

6. The online detection system for regulating capability of a virtual power plant according to claim 5 is characterized in that: The convolutional layer of the convolutional neural network model of the convolutional neural network model module is Among them, the input signal u is the actual operation curve of the virtual power plant detection day, the weight of the convolution kernel v uses a zero-mean Gaussian distribution, (u*v)(i) represents the value of the convolution of the signal u and v at index i; uflip is the result of flipping the input signal u along each axis, * represents the convolution operation, u m is the value of signal u at index m, v i-m is the value of v at index im; The activation function is the Sigmoid function; The pooling layer uses the maximum pooling operation; The formula for the fully connected layer is z = w x + b Among them, z is the output vector; w is the weight matrix, which represents the parameters of the fully connected layer, and its dimension is n×m, where n is the number of output features and m is the number of input features; x is the input vector, which is the output of the previous layer, and its dimension is m; b is the bias vector, and its dimension is n.

7. The online detection system for regulating capability of a virtual power plant according to claim 6 is characterized in that: In the virtual power plant regulation capability detection model module, the steps of building and training the virtual power plant regulation capability detection model based on the deep belief neural network are as follows: The key features of the operation data of the virtual power plant and proxy users extracted by the convolutional neural network model are standardized: When denormalizing: Among them, x i is the corresponding standardized data; z i is the denormalized data; ω2 and ω1 correspond to the upper and lower boundaries respectively, They correspond to the maximum and minimum values ​​of the characteristic data respectively; Initialize the weights and thresholds of the deep belief neural network, set the number of RBM training layers in the deep belief neural network, and perform RBM pre-training; Select the ReLU function as the activation function of the deep belief network RBM: ReLU(x)=max(0,x) When x≥0, output x itself; when x<0, output 0; The output layer uses the softmax function to calculate the probability distribution softmax(s i )for s i is the value of the i-th element in the input vector s; S is the total number of categories, For i from 1 to S and.

8. The online detection system for regulating capability of a virtual power plant according to claim 7 is characterized in that: The key features of the operating data of virtual power plants and proxy users include the 96-point operating curve of virtual power plants, the 96-point operating curve of proxy users, the regulation capacity, regulation accuracy, duration, response time and data delay time reported by virtual power plants and proxy users.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images