WO2021243930A1 - Method for identifying composition of bus load, and machine-readable storage medium - Google Patents
Method for identifying composition of bus load, and machine-readable storage medium Download PDFInfo
- Publication number
- WO2021243930A1 WO2021243930A1 PCT/CN2020/124511 CN2020124511W WO2021243930A1 WO 2021243930 A1 WO2021243930 A1 WO 2021243930A1 CN 2020124511 W CN2020124511 W CN 2020124511W WO 2021243930 A1 WO2021243930 A1 WO 2021243930A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- sub
- neural network
- load data
- bus load
- convolution
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 37
- 239000000203 mixture Substances 0.000 title claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 52
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 49
- 238000012360 testing method Methods 0.000 claims description 39
- 238000012545 processing Methods 0.000 claims description 22
- 238000012937 correction Methods 0.000 claims description 20
- 238000013528 artificial neural network Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 10
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 7
- 238000010586 diagram Methods 0.000 description 23
- 239000011159 matrix material Substances 0.000 description 8
- 230000006870 function Effects 0.000 description 7
- 238000003062 neural network model Methods 0.000 description 6
- 238000013178 mathematical model Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 230000004044 response Effects 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 229910000831 Steel Inorganic materials 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000012067 mathematical method Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000013102 re-test Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 239000004753 textile Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Definitions
- the present disclosure relates to the field of power system automation, for example, to a method for identifying bus load composition and a machine-readable storage medium.
- the bus load is usually composed of a large number of sub-line load resources of different types and characteristics.
- controlling the operation of the grid according to the flexible load response gathered at the busbar is an important means to adjust the balance of market supply and demand and optimize the allocation of resources.
- there may be heavy transmission and transformation equipment during peak power loads For the problems of load and overload, if dispatchers can grasp the overall aggregation characteristics of multiple flexible load responses at the corresponding busbars in time, they can guide these flexible load responses to participate in the operation control of the grid by formulating reasonable incentive mechanisms or electricity price policies to reduce Necessary power grid transformation and expansion.
- the present disclosure provides a method for identifying the bus load composition and a machine-readable storage medium to solve the problem that the power system cannot collect the characteristics of the subordinate composition load of the bus load.
- a method for identifying bus load composition includes:
- Construct a deep convolutional neural network model and train the deep convolutional neural network model through the training set, so that when the bus load data is input into the trained deep convolutional neural network model, the training
- the deep convolutional neural network model outputs the sub-line load label quantity corresponding to the input bus load data.
- a machine-readable storage medium stores an instruction that causes a machine to execute the above-mentioned bus load composition identification method.
- FIG. 1 is a schematic flowchart of a method for identifying a bus load composition according to an embodiment of the present invention
- FIG. 2 is a schematic diagram of a deep convolutional neural network structure provided by an embodiment of the present invention.
- FIG. 3(a) is a schematic diagram after clustering processing of original sub-line load data according to an embodiment of the present invention
- Fig. 3(b) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention.
- Fig. 3(c) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention.
- FIG. 3(d) is a schematic diagram of another original sub-line load data after clustering processing according to an embodiment of the present invention.
- Fig. 3(e) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention.
- FIG. 3(f) is a schematic diagram of another original sub-line load data after clustering processing according to an embodiment of the present invention.
- Fig. 3(g) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention.
- FIG. 3(h) is a schematic diagram of another original sub-line load data after clustering processing according to an embodiment of the present invention.
- Fig. 3(i) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention.
- Fig. 3(j) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention.
- Fig. 3(k) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention.
- Fig. 3(l) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention.
- Fig. 3(m) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention.
- Fig. 3(n) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention.
- FIG. 3(o) is a schematic diagram of another original sub-line load data after clustering processing according to an embodiment of the present invention.
- Fig. 3(p) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention.
- FIG. 4 is a schematic diagram of another deep convolutional neural network structure provided by an embodiment of the present invention.
- FIG. 5 is a schematic diagram of another deep convolutional neural network structure provided by an embodiment of the present invention.
- FIG. 6 is a schematic diagram of a deep convolutional neural network structure after adding a convolution residual block according to an embodiment of the present invention.
- Fig. 1 is a schematic flow chart of a method for identifying a bus load composition according to an embodiment of the present invention. As shown in Fig. 1, the method for identifying a bus load composition includes:
- Step S100 Determine the first relationship formed by the bus load data, the sub-line load data corresponding to the bus load data, and the sub-line load label amount.
- the first relationship in the embodiment of the present invention is a mathematical model composed of bus load.
- the mathematical model is determined by the vector of bus load active power, the row vector group of typical sub-line load active power, and the label quantity representing the number of typical sub-line loads. of.
- the mathematical model composed of the bus load can be expressed by the following formula:
- L total(1*T) is a vector representing the active power of a bus load
- L total(1*T) is a 1*T vector
- T is the number of sampling points (such as 24 points, 96 points, etc.)
- L sub (n*T) is the row vector group representing the active power of the typical sub-line load.
- the typical sub-line load is for example industrial load, residential load, commercial load, etc.
- the row vector group of the typical sub-line load active power is n*T Matrix, n is the total number of types of typical sub-line loads in the area; A (1*n) is the label quantity representing the number of typical sub-line loads, A (1*n) is a 1*n vector, A (1*n) Is the quantity to be solved, and the characteristics of A (1*n) can express the composition characteristics of the bus load.
- prinv() is the operation to solve the pseudo-inverse matrix. If prinv(L sub(n*T) ) has a solution, enter L total(1*T) and perform mathematical calculations to obtain A (1*n) .
- the first relationship may have some problems: for example, the types of sub-line load data that can be collected in formula (1) are limited, and the actual bus load data may have sub-line load data that has not been considered; (2) collection The received sub-line load data and the bus load data to be identified are often under different weather and temperature conditions; (3) There may be a certain difference between the typical sub-line load data and the actual sub-line load data.
- ⁇ (1*T) is the error caused by the above problem. If the traditional mathematical method is used to solve the formula (3), it will often make prinv(L sub(n*T) ) more or no solution. Therefore, it is necessary to try to use a method with strong nonlinear mapping ability to solve this problem.
- Deep learning is applied to the solution of a variety of nonlinear problems.
- Various documents, reports and experiments show that it has excellent effects and strong generalization ability in dealing with nonlinear mapping problems.
- deep convolutional neural network learning is used to obtain the composition characteristics of the bus load, and the learning process of the deep convolutional neural network learning task is as follows:
- Step 200 Establish a training set of the bus load data and the sub-line load label amount according to the first relationship.
- the input quantity is the vector of bus load active power
- Step S300 Construct a learning model of a deep convolutional neural network structure, train the learning model through the training set, so that any bus load data is input to the learning model, and the learning model outputs the corresponding sub-line load label quantity.
- the deep learning method selected in the embodiment of the present invention is a convolutional neural network (Convolutional Neural Network, CNN), which can transform the learning task into a neural network multi-label classification and multi-output regression problem.
- CNN convolutional Neural Network
- the embodiment of the present invention constructs a deep convolutional neural network structure.
- the deep convolutional neural network structure contains multiple convolutional hidden layers, and each convolutional hidden layer consists of a set of convolution kernels.
- the structure extracts the characteristics of the input data by performing convolution operation on the input data. In theory, as the number of convolutional layers in the deep convolutional neural network structure increases, more features of the input data can be analyzed and extracted.
- the learning model of the deep convolutional neural network structure in the embodiment of the present invention includes a plurality of convolutional hidden layers, and each convolutional hidden layer includes a set of convolution kernels, and each convolution kernel has a convolution kernel length and a convolution kernel.
- Four-dimensional parameters of kernel width, convolution kernel depth, and number of convolution kernels are shown in Figure 2.
- the first two dimensions represent the length and width of the convolution kernel
- the third dimension represents the depth of the convolution kernel
- the fourth The dimension represents the number of convolution kernels.
- the learning model for constructing a deep convolutional neural network structure includes: the bus load data is used as an input, and convolution calculation and function processing are performed through each convolution hidden layer to obtain a corresponding output; and each output Compress to a preset dimension and fully connect all output quantities to obtain the sub-line load label quantity.
- the batch normalization (Batch Normalization, BatchNorm) function can be applied to batch normalize the network hidden layer
- the Rectified Linear Unit (ReLU) function can be applied as an activation function
- a Flatten function is used to compress the output dimension of the convolutional layer into one dimension, which is convenient for connection with the following fully connected layer.
- the performing convolution calculation through each of the convolution hidden layers includes: for each convolution kernel of each convolution layer, performing convolution calculation through a second relationship, where the second relationship is in the convolution
- the convolution kernel length and the convolution kernel width of the convolution kernel are the weighted sum of the convolution kernel weight and the input amount under a two-dimensional benchmark.
- the second relationship is expressed by the following formula:
- X new (i, j) represents the output data
- X(i, j) represents the input amount
- c represents the length of the convolution kernel
- d represents the width of the convolution kernel
- ⁇ (u, v) represents the weight of the convolution kernel
- b Represents paranoid parameters.
- Each output quantity can be expressed as the weight sum of the input quantity X and the c*d unit convolution kernel, that is, the feature of the c*d unit input quantity is extracted, and the characteristic of the entire input quantity is the aggregate quantity of these local features. The more the number of convolution kernels, the more local features of the input data can be extracted, so as to provide more information for the regression of the deep convolution model.
- a test set of the bus load data and the load label amount of the sub-line is established; The accuracy of the load label volume is tested.
- a training set of the bus load data and the sub-line load label amount is established, wherein the bus load data includes: acquiring multiple sub-line load data for a set time length at a set time frequency And after performing clustering processing on the multiple sub-line load data, synthesize bus-line load data through the first relationship.
- the ideal training set sample data is the actually collected bus load data and its corresponding sub-line load data, but due to reasons such as confidentiality, the embodiment of the present invention failed to obtain the bus load data of a region, and only a certain number of sub-line load data were obtained. Line load data, and even if the actual bus load data can be obtained, the data volume is limited and cannot meet the data volume requirements of the neural network. Therefore, a large amount of training set sample data needs to be produced.
- the 61 sub-line load data (time length is a day, the sampling frequency is one point per hour) curve obtained in the embodiment of the present application is taken as an example, and the dimensionality reduction processing is performed on the types of the 61 sub-line load data curves.
- the embodiment of the present invention uses the k-means method to cluster these sub-line load data curves, which can be clustered into 16 categories, and the clustering results are shown in Figure 3(a) to Figure 3(p) , And then synthesize the bus load data according to formula (1) in step S100.
- step S300 After obtaining the training set, after training the network model according to step S300, test the accuracy of the predicted load label amounts of multiple sub-lines for different types of sub-line load label amounts:
- the elements of the sub-line load label amount matrix A (1*n) are all 0 and 1.
- this method embodiment can be transformed to solve the multi-label classification problem of the neural network.
- the sub-line load data is data formed by selecting one sub-line load data from each type of sub-line load data in the multiple types of sub-line load data after clustering. Refer to the schematic diagram of the structure of the deep convolutional neural network in FIG. 4 to test the trained neural network model.
- the loss function of the neural network model is as follows:
- I the actual sub-line load label amount
- a i is the calculated sub-line load label amount
- N r is the number of samples in the training set.
- Table 1 The realization results of the recognition accuracy of the training set data volume, the data volume of the test set and the bus load of the test set are shown in Table 1:
- the calculation formula of the accuracy rate is as follows:
- N s is the number of samples in the test set
- n is the number of types of sub-line loads.
- the sub-line load label quantity matrix A (1*n) is a number within a certain numerical range, and this number represents the number of sub-line loads. Take the number range of sub-line load (0, 10) as an example.
- the embodiment of the present invention can be transformed to solve the multi-output regression problem of the neural network.
- the sub-line load data is data composed of one sub-line load data selected from each type of sub-line load data in the multiple types of sub-line load data after clustering. Refer to the schematic diagram of the structure of the deep convolutional neural network in Fig. 5 to test the trained neural network model.
- the loss function expression of the neural network model of the embodiment of the present invention is as follows:
- I the actual sub-line load label amount
- a i is the calculated sub-line load label amount
- N r is the number of samples in the training set.
- the synthesized bus load data is used as input data, and output data is obtained through the neural network model. Determine whether the label quantity matrix (0,...,3,9) of the sub-line load quantity exists.
- the realization results of the recognition accuracy of the training set data volume, the data volume of the test set and the bus load of the test set are shown in Table 2:
- Table 3 shows the proportion of different sub-line load data that constitute the bus load data.
- the bus load composition identification method provided by the embodiment of the present invention can realize the identification of the bus load composition and has a higher identification rate.
- the method for identifying the bus load composition further includes performing error correction on the first relationship.
- performing error correction on the first relationship includes: performing error correction on the first relationship based on the clustering and deviation of the bus load data in the training set; or performing error correction on the first relationship based on differences caused by external factors. According to the deviation of the bus load data, error correction is performed on the first relationship.
- the typical sub-line load data is extracted.
- the sub-line load data that constitutes the bus load data and the typical sub-line load data in the embodiment of the present invention There is a difference between the load data, or these sub-line load data are not completely included in the typical sub-line load data.
- the method further includes: adding error data to the training set; and adding error data to the depth Convolutional hidden layers are added to the convolutional neural network structure, and convolutional residual blocks are added.
- w (1*T) represents the error caused by the clustered typical sub-line load data and the actual sub-line load data and the load data that is not considered by the sub-line load data sample database.
- This embodiment is based on normally distributed data To simulate the error data, the expression is shown in equation (9).
- the training set of step S200 is improved, and the learning model is retrained.
- the deep convolutional neural network structure in Fig. 4 On the basis of the deep convolutional neural network structure in Fig. 4, several layers of convolutional hidden layers are added, and a convolutional residual block is added. As the depth of the convolutional neural network increases, the phenomenon of gradient disappearance appears, and the training effect of the neural network becomes worse. After adding the convolution residual block, it will not add additional parameters and calculations to the original neural network, but also It can increase the training speed of the mathematical model and improve the training effect.
- the structure of the deep convolutional neural network with the convolution residual block added is shown in Figure 6.
- the test 1 is retested. Compared with the above test 1, the data volume of the training set, the data volume of the test set and the bus load of the test set after error correction are performed.
- the composition recognition accuracy rate is shown in Table 4.
- the training set data of the embodiment of the present invention is for clustering processing and sub-lines that are not considered under the learning model of the convolutional neural network structure after adding a convolutional hidden layer and adding a convolutional residual block.
- the error correction caused by the load data has a strong generalization ability.
- the sample data used when training the neural network and the bus load data during actual identification may not be under exactly the same weather and temperature conditions.
- the training set data was collected in May 2019.
- the bus load data that needs to be identified is the data on May 20, 2020.
- the weather and temperature of these two days cannot be exactly the same.
- the sub-line load data used for training and the actual structure of the sub-line load data must exist A certain deviation, which is likely to increase the identification error of the bus load, so the error correction of the first relationship is required.
- the revised first relationship is expressed by the following formula:
- w y(1*T) represents the error caused by the sub-line load data under different weather and temperature conditions.
- This embodiment simulates the error based on uniformly distributed data.
- the expression is shown in equation (11):
- the bus load data of the test set and the bus load data of the training set are under different weather and temperature. Because the actual bus load data cannot be obtained, the simulation method is used to synthesize the bus load data, that is, the bus load data of the test set is composed of The sub-line load data of the training set is synthesized after adding a certain error correction. It is input into the deep convolutional neural network structure of Fig. 4, and the accuracy of the output bus identification result is 83.68%.
- the training set of step S200 is improved, and the learning model is retrained, so that the identification accuracy of the bus can be improved.
- Re-testing test one compared with the original test one, the data volume of the training set after error correction, the data volume of the test set, and the bus load composition of the test set are identified as accurately as shown in Table 5 and Table 6.
- the training set data and the deep convolutional neural network structure learning model of the embodiment of the present invention also has strong generalization ability for the identification error correction of the bus load composition caused by differences in weather, temperature, etc.
- the present disclosure proposes a method for identifying bus load composition.
- the bus load composition identification problem is designed as two scenarios of multi-label classification and multi-output regression of deep convolutional neural networks, and different training set data are designed for different identification scenarios
- the learning model the experimental results show that the method proposed in the present disclosure has a high recognition rate for the components of the bus load, and it also has a strong generalization for clustering, sub-line load data sample deviation and deviation caused by weather and temperature. ability.
- An embodiment of the present invention also provides a machine-readable storage medium, which stores instructions on the machine-readable storage medium, and the instructions cause a machine to execute the bus load composition identification method described in the foregoing embodiment.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Economics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Public Health (AREA)
- Probability & Statistics with Applications (AREA)
- Water Supply & Treatment (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Supply And Distribution Of Alternating Current (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A method for identifying the composition of a bus load, and a machine-readable storage medium. The method for identifying the composition of a bus load comprises: determining bus load data and a first relationship composed of sub-line load data corresponding to the bus load data and sub-line load label quantities (S100); according to the first relationship, establishing a training set of the bus load data and the sub-line load label quantities (S200); constructing a learning model having a deep convolutional neural network structure, and training the learning model by using the training set so that, once any bus load data is inputted into the learning model, the learning model outputs a corresponding sub-line load label quantity (S300).
Description
本申请要求在2020年06月05日提交中国专利局、申请号为202010507240.6的中国专利申请的优先权,该申请的全部内容通过引用结合在本申请中。This application claims the priority of a Chinese patent application filed with the Chinese Patent Office with an application number of 202010507240.6 on June 5, 2020, and the entire content of this application is incorporated into this application by reference.
本公开涉及电力***自动化领域,例如涉及一种母线负荷构成辨识方法及机器可读存储介质。The present disclosure relates to the field of power system automation, for example, to a method for identifying bus load composition and a machine-readable storage medium.
母线负荷通常由大量不同种类、特性迥异的子线负荷资源汇聚而成。随着电力市场工作的推进,根据母线处汇聚的柔性负荷响应量对电网运行进行控制是调节市场供需平衡、优化资源配置的一种重要手段,例如,电力负荷高峰时段可能存在输变电设备重载和过载问题,调度人员如果能够及时掌握相应母线处多个柔性负荷响应量的整体聚合特性,就可通过制定合理的激励机制或电价政策引导这些柔性负荷响应量参与电网的运行控制,减少不必要的电网改造和扩建。根据母线处汇聚的柔性负荷响应量对电网运行进行控制所产生的好处逐渐引起了电网调度部门对于母线负荷的构成成分的关注,但是由于***归属权、经济性等原因,调度部门只能采集到母线负荷的整体特性,不能采集到母线负荷的下属构成负荷特性。The bus load is usually composed of a large number of sub-line load resources of different types and characteristics. With the advancement of the power market, controlling the operation of the grid according to the flexible load response gathered at the busbar is an important means to adjust the balance of market supply and demand and optimize the allocation of resources. For example, there may be heavy transmission and transformation equipment during peak power loads For the problems of load and overload, if dispatchers can grasp the overall aggregation characteristics of multiple flexible load responses at the corresponding busbars in time, they can guide these flexible load responses to participate in the operation control of the grid by formulating reasonable incentive mechanisms or electricity price policies to reduce Necessary power grid transformation and expansion. The benefits of controlling the operation of the power grid based on the flexible load response gathered at the bus bar gradually aroused the attention of the power grid dispatching department to the composition of the bus load. However, due to system ownership and economic reasons, the dispatching department can only collect The overall characteristics of the bus load cannot be collected from the subordinate load characteristics of the bus load.
发明内容Summary of the invention
本公开提供一种母线负荷构成辨识方法及机器可读存储介质,以解决电力***不能采集到母线负荷的下属构成负荷特性的问题。The present disclosure provides a method for identifying the bus load composition and a machine-readable storage medium to solve the problem that the power system cannot collect the characteristics of the subordinate composition load of the bus load.
提供了一种母线负荷构成辨识方法,所述母线负荷构成辨识方法包括:A method for identifying bus load composition is provided, and the method for identifying bus load composition includes:
确定母线负荷数据、所述母线负荷数据对应的子线负荷数据和所述母线负荷数据对应的子线负荷标签量构成的第一关系;Determining a first relationship constituted by bus load data, sub-line load data corresponding to the bus load data, and sub-line load label quantities corresponding to the bus load data;
根据所述第一关系,建立所述母线负荷数据和所述母线负荷数据对应的子线负荷标签量的训练集;According to the first relationship, establish a training set of the bus load data and the sub-line load label quantity corresponding to the bus load data;
构建深度卷积神经网络模型,通过所述训练集对所述深度卷积神经网络模型进行训练,以使在将母线负荷数据输入训练后的深度卷神经网络积模型的情况下,所述训练后的深度卷积神经网络模型输出与输入的母线负荷数据对应的子线负荷标签量。Construct a deep convolutional neural network model, and train the deep convolutional neural network model through the training set, so that when the bus load data is input into the trained deep convolutional neural network model, the training The deep convolutional neural network model outputs the sub-line load label quantity corresponding to the input bus load data.
还提供一种机器可读存储介质,该机器可读存储介质上存储有指令,该指令使得机器执行上述的母线负荷构成辨识方法。A machine-readable storage medium is also provided, the machine-readable storage medium stores an instruction that causes a machine to execute the above-mentioned bus load composition identification method.
图1为本发明实施例提供的一种母线负荷构成辨识方法的流程示意图;FIG. 1 is a schematic flowchart of a method for identifying a bus load composition according to an embodiment of the present invention;
图2为本发明实施例提供的一种深度卷积神经网络结构的示意图;2 is a schematic diagram of a deep convolutional neural network structure provided by an embodiment of the present invention;
图3(a)为本发明实施例提供的一种原始子线负荷数据聚类处理后的示意图;FIG. 3(a) is a schematic diagram after clustering processing of original sub-line load data according to an embodiment of the present invention;
图3(b)为本发明实施例提供的另一种原始子线负荷数据聚类处理后的示意图;Fig. 3(b) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention;
图3(c)为本发明实施例提供的另一种原始子线负荷数据聚类处理后的示意图;Fig. 3(c) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention;
图3(d)为本发明实施例提供的另一种原始子线负荷数据聚类处理后的示意图;FIG. 3(d) is a schematic diagram of another original sub-line load data after clustering processing according to an embodiment of the present invention;
图3(e)为本发明实施例提供的另一种原始子线负荷数据聚类处理后的示意图;Fig. 3(e) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention;
图3(f)为本发明实施例提供的另一种原始子线负荷数据聚类处理后的示意图;FIG. 3(f) is a schematic diagram of another original sub-line load data after clustering processing according to an embodiment of the present invention;
图3(g)为本发明实施例提供的另一种原始子线负荷数据聚类处理后的示意图;Fig. 3(g) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention;
图3(h)为本发明实施例提供的另一种原始子线负荷数据聚类处理后的示意图;FIG. 3(h) is a schematic diagram of another original sub-line load data after clustering processing according to an embodiment of the present invention;
图3(i)为本发明实施例提供的另一种原始子线负荷数据聚类处理后的示意图;Fig. 3(i) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention;
图3(j)为本发明实施例提供的另一种原始子线负荷数据聚类处理后的示意图;Fig. 3(j) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention;
图3(k)为本发明实施例提供的另一种原始子线负荷数据聚类处理后的示意图;Fig. 3(k) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention;
图3(l)为本发明实施例提供的另一种原始子线负荷数据聚类处理后的示意图;Fig. 3(l) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention;
图3(m)为本发明实施例提供的另一种原始子线负荷数据聚类处理后的示 意图;Fig. 3(m) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention;
图3(n)为本发明实施例提供的另一种原始子线负荷数据聚类处理后的示意图;Fig. 3(n) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention;
图3(o)为本发明实施例提供的另一种原始子线负荷数据聚类处理后的示意图;FIG. 3(o) is a schematic diagram of another original sub-line load data after clustering processing according to an embodiment of the present invention;
图3(p)为本发明实施例提供的另一种原始子线负荷数据聚类处理后的示意图;Fig. 3(p) is a schematic diagram of another original sub-line load data after clustering processing provided by an embodiment of the present invention;
图4为本发明实施例提供的另一种深度卷积神经网络结构的示意图;4 is a schematic diagram of another deep convolutional neural network structure provided by an embodiment of the present invention;
图5为本发明实施例提供的另一种深度卷积神经网络结构的示意图;5 is a schematic diagram of another deep convolutional neural network structure provided by an embodiment of the present invention;
图6为本发明实施例提供的一种添加卷积残差块后的深度卷积神经网络结构的示意图。FIG. 6 is a schematic diagram of a deep convolutional neural network structure after adding a convolution residual block according to an embodiment of the present invention.
下面将参考附图并结合实施例来说明本公开。Hereinafter, the present disclosure will be described with reference to the drawings and in conjunction with the embodiments.
以下说明均是示例性的说明,旨在对本公开提供说明。除非另有指明,本公开所采用的所有技术术语与本申请所属领域的一般技术人员的通常理解的含义相同。The following descriptions are all exemplary descriptions, and are intended to provide descriptions of the present disclosure. Unless otherwise specified, all technical terms used in this disclosure have the same meanings as commonly understood by those skilled in the art to which this application belongs.
图1为本发明实施例提供的一种母线负荷构成辨识方法的流程示意图,如图1所述,所述母线负荷构成辨识方法包括:Fig. 1 is a schematic flow chart of a method for identifying a bus load composition according to an embodiment of the present invention. As shown in Fig. 1, the method for identifying a bus load composition includes:
步骤S100:确定母线负荷数据、所述母线负荷数据对应的子线负荷数据和子线负荷标签量构成的第一关系。Step S100: Determine the first relationship formed by the bus load data, the sub-line load data corresponding to the bus load data, and the sub-line load label amount.
本发明实施例中的第一关系是母线负荷构成的数学模型,该数学模型是通过母线负荷有功功率的向量、典型子线负荷有功功率的行向量组和代表典型子线负荷数量的标签量确定的。所述母线负荷构成的数学模型可以用下式表示:The first relationship in the embodiment of the present invention is a mathematical model composed of bus load. The mathematical model is determined by the vector of bus load active power, the row vector group of typical sub-line load active power, and the label quantity representing the number of typical sub-line loads. of. The mathematical model composed of the bus load can be expressed by the following formula:
L
total(1*T)=A
(1*n)*L
sub(n*T) (1)
L total(1*T) = A (1*n) *L sub(n*T) (1)
其中,L
total(1*T)为表示一条母线负荷的有功功率的向量,L
total(1*T)为1*T的向量,T为采样点数(例如24点,96点等);L
sub(n*T)为表示典型子线负荷有功功率的行向量组,典型子线负荷例如是工业负荷、居民负荷、商业负荷等,该典型子线负荷有功功率的行向量组为n*T的矩阵,n为该地区典型子线负荷的类型总数;A
(1*n)为表示典型子线负荷数量的标签量,A
(1*n)为1*n的向量,A
(1*n)为待求解的量,且A
(1*n)的特征可以表示出母线负荷的构成特征。
Among them, L total(1*T) is a vector representing the active power of a bus load, L total(1*T) is a 1*T vector, and T is the number of sampling points (such as 24 points, 96 points, etc.); L sub (n*T) is the row vector group representing the active power of the typical sub-line load. The typical sub-line load is for example industrial load, residential load, commercial load, etc. The row vector group of the typical sub-line load active power is n*T Matrix, n is the total number of types of typical sub-line loads in the area; A (1*n) is the label quantity representing the number of typical sub-line loads, A (1*n) is a 1*n vector, A (1*n) Is the quantity to be solved, and the characteristics of A (1*n) can express the composition characteristics of the bus load.
通过式(1),可以将待求解的量A
(1*n)通过下式表示:
Through formula (1), the quantity to be solved A (1*n) can be expressed by the following formula:
L
total(1*T)*prinv(L
sub(n*T))=A
(1*n) (2)
L total(1*T) *prinv(L sub(n*T) )=A (1*n) (2)
其中,prinv()为求解伪逆矩阵的运算,若prinv(L
sub(n*T))存在解,则输入L
total(1*T),进行数学计算便可得到A
(1*n)。
Among them, prinv() is the operation to solve the pseudo-inverse matrix. If prinv(L sub(n*T) ) has a solution, enter L total(1*T) and perform mathematical calculations to obtain A (1*n) .
然而应用于实际中,第一关系可能存在一些问题:例如式(1)中能够采集到的子线负荷数据种类有限,实际母线负荷数据可能存在未考虑到的子线负荷数据;(2)采集到的子线负荷数据与要辨识的母线负荷数据往往处于不同的气象、温度条件下;(3)典型子线负荷数据与真实子线负荷数据之间可能存在一定差值。However, in practice, the first relationship may have some problems: for example, the types of sub-line load data that can be collected in formula (1) are limited, and the actual bus load data may have sub-line load data that has not been considered; (2) collection The received sub-line load data and the bus load data to be identified are often under different weather and temperature conditions; (3) There may be a certain difference between the typical sub-line load data and the actual sub-line load data.
考虑上述问题的影响,可以在式(1)中添加误差变量得到下列式(3),Considering the influence of the above problems, the error variable can be added to the formula (1) to obtain the following formula (3),
L
total(1*T)=A
(1*n)*L
sub(n*T)+σ
(1*T) (3)
L total(1*T) = A (1*n) *L sub(n*T) +σ (1*T) (3)
σ
(1*T)为上述问题带来的误差,若采用传统数学方法求解式(3),往往会使prinv(L
sub(n*T))多解或无解。因此,需要尝试采用有强大非线性映射能力的方法来求解该问题。
σ (1*T) is the error caused by the above problem. If the traditional mathematical method is used to solve the formula (3), it will often make prinv(L sub(n*T) ) more or no solution. Therefore, it is necessary to try to use a method with strong nonlinear mapping ability to solve this problem.
深度学习被应用于多种非线性问题的求解上,多种文献、报告以及实验表明其在处理非线性映射问题上,具有优良的效果和强大的泛化能力。本发明实施例应用深度卷积神经网络学习得到母线负荷的构成特征,深度卷积神经网络学习任务的学习过程如下:Deep learning is applied to the solution of a variety of nonlinear problems. Various documents, reports and experiments show that it has excellent effects and strong generalization ability in dealing with nonlinear mapping problems. In the embodiment of the present invention, deep convolutional neural network learning is used to obtain the composition characteristics of the bus load, and the learning process of the deep convolutional neural network learning task is as follows:
步骤200:根据所述第一关系,建立所述母线负荷数据和所述子线负荷标签量的训练集。Step 200: Establish a training set of the bus load data and the sub-line load label amount according to the first relationship.
根据式(1)或式(2)建立输入量是母线负荷有功功率的向量,输出量是典型子线负荷数量的向量的训练集,例如是{(L
total(1*T)1,A
(1*n)1),(L
total(1*T)2,A
(1*n)2),…,(L
total(1*T)m,A
(1*n)m)},其中(L
total(1*T)k,A
(1*n)k)为第k个训练样本,k=1,2,…,m,L
total(1*T)k为第k个样本的母线负荷数据,A
(1*n)k为第k个样本中的n种子线负荷标签量。
According to formula (1) or formula (2), the input quantity is the vector of bus load active power, and the output quantity is the training set of the vector of typical sub-line load quantity, for example {(L total(1*T)1 , A ( 1*n)1 ), (L total(1*T)2 , A (1*n)2 ),..., (L total(1*T)m , A (1*n)m )}, where ( L total(1*T)k , A (1*n)k ) is the kth training sample, k=1, 2,...,m, L total(1*T)k is the bus load of the kth sample Data, A (1*n)k is the n seed line load label amount in the k-th sample.
步骤S300:构建深度卷积神经网络结构的学习模型,通过所述训练集对所述学习模型进行训练,以使任意母线负荷数据输入所述学习模型,所述学习模型输出对应的子线负荷标签量。Step S300: Construct a learning model of a deep convolutional neural network structure, train the learning model through the training set, so that any bus load data is input to the learning model, and the learning model outputs the corresponding sub-line load label quantity.
本发明实施例所选择的深度学习方法为卷积神经网络(Convolutional Neural Network,CNN),可以将该学习任务转化为神经网络的多标签分类、多输出回归问题。The deep learning method selected in the embodiment of the present invention is a convolutional neural network (Convolutional Neural Network, CNN), which can transform the learning task into a neural network multi-label classification and multi-output regression problem.
本发明实施例构建了一种深度卷积神经网络结构,相比普通卷积神经网络结构,深度卷积神经网络结构含有多个卷积隐层,每个卷积隐层由一组卷积核 构成,通过对输入数据进行卷积运算进而提取输入数据的特征。理论上讲,随着深度卷积神经网络结构中的卷积层层数增加,可以分析和提取到输入数据的更多特征。The embodiment of the present invention constructs a deep convolutional neural network structure. Compared with the ordinary convolutional neural network structure, the deep convolutional neural network structure contains multiple convolutional hidden layers, and each convolutional hidden layer consists of a set of convolution kernels. The structure extracts the characteristics of the input data by performing convolution operation on the input data. In theory, as the number of convolutional layers in the deep convolutional neural network structure increases, more features of the input data can be analyzed and extracted.
本发明实施例中的深度卷积神经网络结构的学习模型包括多个卷积隐层,每个卷积隐层包含一组卷积核,且每个卷积核有卷积核长度、卷积核宽度、卷积核深度和卷积核数量四维参数。深度卷积神经网络结构的示意图如图2所示,卷积核对应的四维矩阵中,前两个维度表示卷积核的长度和宽度,第三个维度表示卷积核的深度,第四个维度表示卷积核的数量。The learning model of the deep convolutional neural network structure in the embodiment of the present invention includes a plurality of convolutional hidden layers, and each convolutional hidden layer includes a set of convolution kernels, and each convolution kernel has a convolution kernel length and a convolution kernel. Four-dimensional parameters of kernel width, convolution kernel depth, and number of convolution kernels. The schematic diagram of the deep convolutional neural network structure is shown in Figure 2. In the four-dimensional matrix corresponding to the convolution kernel, the first two dimensions represent the length and width of the convolution kernel, the third dimension represents the depth of the convolution kernel, and the fourth The dimension represents the number of convolution kernels.
所述构建深度卷积神经网络结构的学***坦化(Flatten)函数将卷积层的输出量维度压缩为1维,方便与后面的全连接层连接。The learning model for constructing a deep convolutional neural network structure includes: the bus load data is used as an input, and convolution calculation and function processing are performed through each convolution hidden layer to obtain a corresponding output; and each output Compress to a preset dimension and fully connect all output quantities to obtain the sub-line load label quantity. As shown in Figure 2, after the convolution calculation of each convolutional hidden layer, the batch normalization (Batch Normalization, BatchNorm) function can be applied to batch normalize the network hidden layer, and the Rectified Linear Unit (ReLU) function can be applied As an activation function, a Flatten function is used to compress the output dimension of the convolutional layer into one dimension, which is convenient for connection with the following fully connected layer.
所述通过每个所述卷积隐层进行卷积计算,包括:针对每个卷积层的每个卷积核,通过第二关系进行卷积计算,所述第二关系是在所述卷积核的卷积核长度和卷积核宽度二维基准下的卷积核权重与输入量的加权和。所述第二关系用下式表示:The performing convolution calculation through each of the convolution hidden layers includes: for each convolution kernel of each convolution layer, performing convolution calculation through a second relationship, where the second relationship is in the convolution The convolution kernel length and the convolution kernel width of the convolution kernel are the weighted sum of the convolution kernel weight and the input amount under a two-dimensional benchmark. The second relationship is expressed by the following formula:
其中,X
new(i,j)表示输出数据,X(i,j)表示输入量,c表示卷积核长度,d表示卷积核宽度,ω(u,v)表示卷积核权重,b表示偏执参数。每个输出量可以表达为输入量X与c*d单元卷积核的权重和,即提取了c*d单元输入量的特征,整个输入量的特征为这些局部特征的聚合量。卷积核数量越多,就能够提取到输入数据越多的局部特征,以此为深度卷积模型的回归提供更多信息。
Among them, X new (i, j) represents the output data, X(i, j) represents the input amount, c represents the length of the convolution kernel, d represents the width of the convolution kernel, ω(u, v) represents the weight of the convolution kernel, b Represents paranoid parameters. Each output quantity can be expressed as the weight sum of the input quantity X and the c*d unit convolution kernel, that is, the feature of the c*d unit input quantity is extracted, and the characteristic of the entire input quantity is the aggregate quantity of these local features. The more the number of convolution kernels, the more local features of the input data can be extracted, so as to provide more information for the regression of the deep convolution model.
得到该深度卷积神经网络结构的学习模型后,通过训练集对该学习模型进行训练,以使该学习模型可对测试集上的每一个例如L
total(1*T),输出其对应的n种子线负荷标签量例如A
(1*n)。
After obtaining the learning model of the deep convolutional neural network structure, train the learning model through the training set, so that the learning model can output the corresponding n for each item on the test set, such as L total(1*T) The amount of seed line load label, for example, A (1*n) .
本发明实施例中,建立所述母线负荷数据和所述子线负荷标签量的测试集;以及通过所述学习模型,针对不同类别的所述子线负荷标签量,对预测出的多个子线负荷标签量的准确性进行测试。建立与训练集数据结构一样的测试集,例如是{(L
total(1*T)m+1,A
(1*n)m+1),(L
total(1*T)m+2,A
(1*n)m+2),…,(L
total(1*T)m+k,A
(1*n)m+k)},其中(L
total(1*T)m+k,A
(1*n)m+k)为第k 个测试样本,k=1,2,…,L
total(1*T)m+k为第k个样本的母线负荷数据,A
(1*n)m+k为第k个样本中的n种子线负荷标签量。
In the embodiment of the present invention, a test set of the bus load data and the load label amount of the sub-line is established; The accuracy of the load label volume is tested. Create a test set with the same data structure as the training set, for example {(L total(1*T)m+1 , A (1*n)m+1 ), (L total(1*T)m+2 , A (1*n)m+2 ),...,(L total(1*T)m+k , A (1*n)m+k )}, where (L total(1*T)m+k , A (1*n)m+k ) is the kth test sample, k=1, 2,..., L total(1*T)m+k is the bus load data of the kth sample, A (1*n) m+k is the load label amount of n seed lines in the k-th sample.
通过上述步骤,基于实际数据,得到本申请实施例的母线负荷构成辨识方法。Through the above steps and based on actual data, the bus load composition identification method of the embodiment of the present application is obtained.
针对步骤S200,建立所述母线负荷数据和所述子线负荷标签量的训练集,其中,所述母线负荷数据包括:以设定的时间频率获取设定的时间长度下的多个子线负荷数据;以及对所述多个子线负荷数据进行聚类处理后,通过所述第一关系合成母线负荷数据。For step S200, a training set of the bus load data and the sub-line load label amount is established, wherein the bus load data includes: acquiring multiple sub-line load data for a set time length at a set time frequency And after performing clustering processing on the multiple sub-line load data, synthesize bus-line load data through the first relationship.
理想的训练集样本数据为实际采集到的母线负荷数据及其对应的子线负荷数据,但由于保密等原因,本发明实施例未能获得一地区的母线负荷数据,仅获得了一定数量的子线负荷数据,且即使能获得实际的母线负荷数据,其数据量也是有限的,不能够满足神经网络的数据量要求,因此,需要制造大量的训练集样本数据。The ideal training set sample data is the actually collected bus load data and its corresponding sub-line load data, but due to reasons such as confidentiality, the embodiment of the present invention failed to obtain the bus load data of a region, and only a certain number of sub-line load data were obtained. Line load data, and even if the actual bus load data can be obtained, the data volume is limited and cannot meet the data volume requirements of the neural network. Therefore, a large amount of training set sample data needs to be produced.
实际的子线负荷数据曲线成千上万,如果直接以这些子线负荷数据来构成子线负荷数据样本,并不利于本发明实施例的实施。首先,从数学模型的角度分析,由于这些子线负荷数据不是完全线性无关的,易导致子线负荷标签量矩阵A存在多解或无解的情形。其次,从网络模型训练的角度分析,若直接以这些子线负荷数据来构成子线负荷数据样本会导致输出数据维度大大增加,进而使神经网络的权重值的数量成倍增加,需要训练的样本数量也会呈指数级增加,进而降低网络模型训练的收敛效率和收敛精度。最后,从实际用户的需求角度分析,用户的需求一般为辨识出母线负荷是否含有一些特定行业的负荷,该特定行业例如为重工业(如钢铁、化工、纺织等),往往不会苛求于辨识出哪家企业的负荷数据。There are thousands of actual sub-line load data curves. If these sub-line load data are directly used to form the sub-line load data samples, it is not conducive to the implementation of the embodiment of the present invention. First, from the perspective of mathematical model analysis, since these sub-line load data are not completely linearly independent, it is easy to cause the sub-line load label quantity matrix A to have multiple solutions or no solutions. Secondly, from the perspective of network model training, if these sub-line load data are directly used to form the sub-line load data samples, the output data dimension will be greatly increased, and the number of weight values of the neural network will increase exponentially. The samples that need training The number will also increase exponentially, thereby reducing the convergence efficiency and convergence accuracy of network model training. Finally, from the perspective of actual user demand, the user’s demand is generally to identify whether the bus load contains the load of some specific industries, such as heavy industries (such as steel, chemicals, textiles, etc.), which often do not require identification. Which company’s load data.
因此,本申请实施例以获得的一地区的61条子线负荷数据(时间长度为日,采样频率为每小时一个点)曲线为例,对这61条子线负荷数据曲线的类型进行降维处理,本发明实施例采用k-均值(k-means)方法对这些子线负荷数据曲线进行聚类处理,可聚类为16类,聚类结果如图3(a)至图3(p)所示,然后根据步骤S100中的式(1)进行母线负荷数据的合成。Therefore, the 61 sub-line load data (time length is a day, the sampling frequency is one point per hour) curve obtained in the embodiment of the present application is taken as an example, and the dimensionality reduction processing is performed on the types of the 61 sub-line load data curves. The embodiment of the present invention uses the k-means method to cluster these sub-line load data curves, which can be clustered into 16 categories, and the clustering results are shown in Figure 3(a) to Figure 3(p) , And then synthesize the bus load data according to formula (1) in step S100.
得到训练集后,根据步骤S300对网络模型进行训练后,针对不同类别的子线负荷标签量,对预测出的多个子线负荷标签量的准确性进行测试:After obtaining the training set, after training the network model according to step S300, test the accuracy of the predicted load label amounts of multiple sub-lines for different types of sub-line load label amounts:
测试一,例如子线负荷标签量矩阵A
(1*n)的元素均为0、1,此时本方法实施例可转化为解决神经网络的多标签分类的问题。例如,子线负荷数据为从聚类处理后的多类子线负荷数据中的每类子线负荷数据选取一条子线负荷数据组 成的数据。参考图4的深度卷积神经网络的结构示意图,对训练后的神经网络模型进行测试。
Test one, for example, the elements of the sub-line load label amount matrix A (1*n) are all 0 and 1. At this time, this method embodiment can be transformed to solve the multi-label classification problem of the neural network. For example, the sub-line load data is data formed by selecting one sub-line load data from each type of sub-line load data in the multiple types of sub-line load data after clustering. Refer to the schematic diagram of the structure of the deep convolutional neural network in FIG. 4 to test the trained neural network model.
在该测试环境下,神经网络模型的损失函数式如下:In this test environment, the loss function of the neural network model is as follows:
其中,
为实际子线负荷标签量,A
i为计算子线负荷标签量,N
r为训练集的样本数量。
in, Is the actual sub-line load label amount, A i is the calculated sub-line load label amount, and N r is the number of samples in the training set.
参考图4,合成的母线负荷数据作为输入数据,经过标准定标器(StandardScaler)方法标准化,假设样本的每个数据服从正态分布,然后按照x=(x-μ)/σ将母线负荷数据转化为标准正态分布,通过该神经网络模型得到输出数据。判断子线负荷标签量矩阵(0,…,1,1)是否存在。测试一的训练集的数据量、测试集的数据量及测试集的母线负荷构成辨识准确率的实现结果如表1所示:Referring to Figure 4, the synthesized bus load data is used as input data, standardized by the StandardScaler method, assuming that each data of the sample obeys a normal distribution, and then the bus load data is calculated according to x=(x-μ)/σ It is transformed into a standard normal distribution, and the output data is obtained through the neural network model. Determine whether the sub-line load label amount matrix (0,...,1,1) exists. The realization results of the recognition accuracy of the training set data volume, the data volume of the test set and the bus load of the test set are shown in Table 1:
表1测试一的实验结果Table 1 Experimental results of test one
一实施例中,准确率的计算公式如下:In one embodiment, the calculation formula of the accuracy rate is as follows:
其中,N
s为测试集的样本数量,n为子线负荷的类型数量。
Among them, N s is the number of samples in the test set, and n is the number of types of sub-line loads.
测试二,例如子线负荷标签量矩阵A
(1*n)为一定数值范围内的数字,该数字即表示子线负荷的数量,以子线负荷的数量范围是(0,10)为例,此时本发明实施例可转化为解决神经网络的多输出回归的问题。同测试一类似,子线负荷数据为从聚类处理后的多类子线负荷数据中的每类子线负荷数据选取一条子线负荷数据组成的数据。参考图5的深度卷积神经网络的结构示意图,对训练后的神经网络模型进行测试。
Test two, for example, the sub-line load label quantity matrix A (1*n) is a number within a certain numerical range, and this number represents the number of sub-line loads. Take the number range of sub-line load (0, 10) as an example. At this time, the embodiment of the present invention can be transformed to solve the multi-output regression problem of the neural network. Similar to Test 1, the sub-line load data is data composed of one sub-line load data selected from each type of sub-line load data in the multiple types of sub-line load data after clustering. Refer to the schematic diagram of the structure of the deep convolutional neural network in Fig. 5 to test the trained neural network model.
在该测试环境下,本发明实施例的神经网络模型的损失函数表达式如下:In this test environment, the loss function expression of the neural network model of the embodiment of the present invention is as follows:
其中,
为实际子线负荷标签量,A
i为计算子线负荷标签量,N
r为训练集的样本数量。
in, Is the actual sub-line load label amount, A i is the calculated sub-line load label amount, and N r is the number of samples in the training set.
参考图5,合成的母线负荷数据作为输入数据,通过该神经网络模型得到输出数据。判断子线负荷数量的标签量矩阵(0,…,3,9)是否存在。测试二的训练集的数据量、测试集的数据量及测试集的母线负荷构成辨识准确率的实现结果如表2所示:Referring to Figure 5, the synthesized bus load data is used as input data, and output data is obtained through the neural network model. Determine whether the label quantity matrix (0,...,3,9) of the sub-line load quantity exists. The realization results of the recognition accuracy of the training set data volume, the data volume of the test set and the bus load of the test set are shown in Table 2:
表2测试二的实验结果Table 2 Experimental results of test two
表3是构成母线负荷数据的不同子线负荷数据的占比情况。Table 3 shows the proportion of different sub-line load data that constitute the bus load data.
表3子线负荷在母线负荷中的占比及占比对应的识别准确率Table 3 The proportion of sub-line load in the bus load and the corresponding recognition accuracy rate
通过表1和表2可知,测试二中的母线负荷的平均辨识准确率没有测试一中的母线负荷的平均辨识准确率高,神经网络在处理测试二的回归问题上,较处理测试一的多标签分类问题,整体效果差一些。但通过表3可知,对于母线负荷数据中占比较高的子线负荷数据,其辨识率仍比较高。It can be seen from Table 1 and Table 2 that the average identification accuracy rate of the bus load in Test 2 is not as high as that of the bus load in Test 1. The neural network handles the regression problem of Test 2 more than that in Test 1. The overall effect of label classification is poor. However, it can be seen from Table 3 that the recognition rate is still relatively high for the sub-line load data which accounts for a relatively high proportion of the bus load data.
基于上述测试可知,本发明实施例提供的母线负荷构成辨识方法,能够实现对母线负荷的构成辨识,且具有较高的辨识率。Based on the above test, it can be known that the bus load composition identification method provided by the embodiment of the present invention can realize the identification of the bus load composition and has a higher identification rate.
因为实际的母线负荷场景与训练集中的母线负荷生成场景之间存在一定偏差,包括:(1)能够采集到的L
sub(n*T)种类有限,实际母线负荷构成中可能存在未考虑到的子线负荷;(2)采集到的子线负荷数据与要辨识的母线负荷数据往往处于不同的气象、温度条件下;(3)典型子线负荷与真实子线负荷间存在一定差值。因此,所述母线负荷构成辨识方法还包括对所述第一关系进行误差修正。
Because there are certain deviations between the actual bus load scene and the bus load generation scene in the training set, including: (1 ) The types of L sub(n*T) that can be collected are limited, and there may be unconsidered in the actual bus load composition Sub-line load; (2) The collected sub-line load data and the bus load data to be identified are often under different weather and temperature conditions; (3) There is a certain difference between the typical sub-line load and the actual sub-line load. Therefore, the method for identifying the bus load composition further includes performing error correction on the first relationship.
一实施例中,对所述第一关系进行误差修正,包括:基于所述训练集中所述母线负荷数据的聚类和偏差,对所述第一关系进行误差修正;或基于外因差异引起的所述母线负荷数据的偏差,对所述第一关系进行误差修正。In an embodiment, performing error correction on the first relationship includes: performing error correction on the first relationship based on the clustering and deviation of the bus load data in the training set; or performing error correction on the first relationship based on differences caused by external factors. According to the deviation of the bus load data, error correction is performed on the first relationship.
1)基于所述训练集中所述母线负荷数据的聚类和偏差,对所述第一关系进行误差修正。1) Perform error correction on the first relationship based on the clustering and deviation of the bus load data in the training set.
上述实施例中,在对实际子线负荷数据聚类处理后,提取的是典型子线负荷数据,但本发明实施例在实际应用场景中,构成母线负荷数据的子线负荷数据与典型子线负荷数据之间存在差值,或者这些子线负荷数据不完全包含于典型子线负荷数据中。In the above embodiment, after the actual sub-line load data is clustered, the typical sub-line load data is extracted. However, in the actual application scenario, the sub-line load data that constitutes the bus load data and the typical sub-line load data in the embodiment of the present invention There is a difference between the load data, or these sub-line load data are not completely included in the typical sub-line load data.
一实施例中,在基于所述训练集中所述母线负荷数据的聚类和偏差,对所述第一关系进行误差修正之后,还包括:在所述训练集中加入误差数据;以及在所述深度卷积神经网络结构中添加卷积隐层,并添加卷积残差块。In an embodiment, after performing error correction on the first relationship based on the clustering and deviation of the bus load data in the training set, the method further includes: adding error data to the training set; and adding error data to the depth Convolutional hidden layers are added to the convolutional neural network structure, and convolutional residual blocks are added.
若测试集的母线负荷数据不是上述16种典型子线负荷数据的叠加,而是实际子线负荷数据的叠加,通过测试一的深度卷积神经网络,输出的母线负荷构成辨识结果平均准确率为85.43%,因此需要对第一关系进行误差修正。修正后的第一关系用下式表示:If the bus load data of the test set is not the superposition of the above-mentioned 16 typical sub-line load data, but the superposition of the actual sub-line load data, through the deep convolutional neural network of test 1, the average accuracy of the output bus load composition identification result is 85.43%, so the error correction of the first relationship is needed. The revised first relationship is expressed by the following formula:
L
total(1*T)=A
(1*n)*L
sub(n*T)+w
j(1*T) (8)
L total(1*T) = A (1*n) *L sub(n*T) + w j(1*T) (8)
其中,w
(1*T)代表聚类后的典型子线负荷数据与实际子线负荷数据引起的误差以及子线负荷数据样本库未考虑到的负荷数据,本实施例基于正态分布的数据来模拟误差数据,表达式如式(9)所示。
Among them, w (1*T) represents the error caused by the clustered typical sub-line load data and the actual sub-line load data and the load data that is not considered by the sub-line load data sample database. This embodiment is based on normally distributed data To simulate the error data, the expression is shown in equation (9).
w
j~N(μ,σ
2) (9)
w j ~N(μ,σ 2 ) (9)
根据式(8)对步骤S200的训练集进行改进,重新训练学习模型。在图4的深度卷积神经网络结构的基础上多添加了几层卷积隐层,并添加了卷积残差块。随着卷积神经网络深度的增加,出现了梯度消失现象,神经网络的训练效果反而变差,添加卷积残差块后,其不但不会给原神经网络增加额外的参数和计算量,而且可以提高数学模型的训练速度,并提高训练效果。添加了卷积残差块的深度卷积神经网络结构如图6所示。According to formula (8), the training set of step S200 is improved, and the learning model is retrained. On the basis of the deep convolutional neural network structure in Fig. 4, several layers of convolutional hidden layers are added, and a convolutional residual block is added. As the depth of the convolutional neural network increases, the phenomenon of gradient disappearance appears, and the training effect of the neural network becomes worse. After adding the convolution residual block, it will not add additional parameters and calculations to the original neural network, but also It can increase the training speed of the mathematical model and improve the training effect. The structure of the deep convolutional neural network with the convolution residual block added is shown in Figure 6.
以式(9)的μ=0,σ=50为例,对测试一重新测试,相比上述测试一,进行误差修正后的训练集的数据量、测试集的数据量及测试集的母线负荷构成辨识准确率如表4所示。Taking the μ=0 and σ=50 of formula (9) as an example, the test 1 is retested. Compared with the above test 1, the data volume of the training set, the data volume of the test set and the bus load of the test set after error correction are performed. The composition recognition accuracy rate is shown in Table 4.
表4误差修正后的实验结果Table 4 Experimental results after error correction
通过实验结果可知,本发明实施例的训练集数据在添加卷积隐层并添加了卷积残差块后的卷积神经网络结构的学习模型下,对于聚类处理和未考虑到的子线负荷数据引起的误差修正具有较强的泛化能力。It can be seen from the experimental results that the training set data of the embodiment of the present invention is for clustering processing and sub-lines that are not considered under the learning model of the convolutional neural network structure after adding a convolutional hidden layer and adding a convolutional residual block. The error correction caused by the load data has a strong generalization ability.
2)基于外因差异引起的所述母线负荷数据的偏差,对所述第一关系进行误差修正。2) Perform error correction on the first relationship based on the deviation of the bus load data caused by the difference of external factors.
气象、温度一直是影响负荷数据的一个重要因素,训练神经网络时采用的样本数据与实际辨识时的母线负荷数据可能不处于完全相同的气象、温度条件下,例如训练集数据采集于2019年5月20日,而需要辨识的母线负荷数据为2020年5月20日的数据,这两天的气象、温度不可能完全相同,训练用的子线负荷数据与真实构成的子线负荷数据势必存在一定偏差,该偏差很可能会增大母线负荷的辨识误差,因此需要对第一关系进行误差修正。修正后的第一关系用下式表示:Meteorology and temperature have always been an important factor affecting load data. The sample data used when training the neural network and the bus load data during actual identification may not be under exactly the same weather and temperature conditions. For example, the training set data was collected in May 2019. On the 20th, the bus load data that needs to be identified is the data on May 20, 2020. The weather and temperature of these two days cannot be exactly the same. The sub-line load data used for training and the actual structure of the sub-line load data must exist A certain deviation, which is likely to increase the identification error of the bus load, so the error correction of the first relationship is required. The revised first relationship is expressed by the following formula:
L
total(1*T)=A
(1*n)*L
sub(n*T)+w
y(1*T) (10)
L total(1*T) = A (1*n) *L sub(n*T) +w y(1*T) (10)
其中,w
y(1*T)代表处于不同气象、温度条件下的子线负荷数据引起的误差,本实施例基于均匀分布的数据来模拟该误差,表达式如式(11)所示:
Among them, w y(1*T) represents the error caused by the sub-line load data under different weather and temperature conditions. This embodiment simulates the error based on uniformly distributed data. The expression is shown in equation (11):
测试集的母线负荷数据与训练集的母线负荷数据处于不同的气象、温度下,因为不能获得实际的母线负荷数据,还是采用仿真的方法进行母线负荷数据合成,即测试集的母线负荷数据是由训练集的子线负荷数据添加了一定误差修正后合成的。将其输入图4的深度卷积神经网络结构,输出的母线辨识结果的准确率为83.68%。The bus load data of the test set and the bus load data of the training set are under different weather and temperature. Because the actual bus load data cannot be obtained, the simulation method is used to synthesize the bus load data, that is, the bus load data of the test set is composed of The sub-line load data of the training set is synthesized after adding a certain error correction. It is input into the deep convolutional neural network structure of Fig. 4, and the accuracy of the output bus identification result is 83.68%.
根据式(10)对步骤S200的训练集进行改进,重新训练学习模型,使该母线辨识精度能够提高。对测试一重新测试,相比原始测试一,进行误差修正后的训练集的数据量、测试集的数据量及测试集的母线负荷构成辨识准确率如表5和表6所示。According to formula (10), the training set of step S200 is improved, and the learning model is retrained, so that the identification accuracy of the bus can be improved. Re-testing test one, compared with the original test one, the data volume of the training set after error correction, the data volume of the test set, and the bus load composition of the test set are identified as accurately as shown in Table 5 and Table 6.
表5误差修正后的子线负荷为有名值的实验结果Table 5 The experimental results of the sub-line load after the error correction is a famous value
表6误差修正后的子线负荷为标幺值的实验结果Table 6 The experimental results of the sub-line load after the error correction is the standard unit value
通过实验结果可知,本发明实施例的训练集数据和深度卷积神经网络结构学习模型,对于气象、温度等差异引起的母线负荷构成辨识误差修正,也具有较强的泛化能力。It can be known from the experimental results that the training set data and the deep convolutional neural network structure learning model of the embodiment of the present invention also has strong generalization ability for the identification error correction of the bus load composition caused by differences in weather, temperature, etc.
本公开提出了一种母线负荷构成辨识方法,将母线负荷构成辨识问题设计为深度卷积神经网络的多标签分类和多输出回归两大情景,并针对不同的辨识情景设计了不同的训练集数据和学习模型,实验结果表明,本公开所提方法对于母线负荷的构成成分具有较高的辨识率,针对聚类、子线负荷数据样本偏差和气象、温度引起的偏差也具有较强的泛化能力。The present disclosure proposes a method for identifying bus load composition. The bus load composition identification problem is designed as two scenarios of multi-label classification and multi-output regression of deep convolutional neural networks, and different training set data are designed for different identification scenarios And the learning model, the experimental results show that the method proposed in the present disclosure has a high recognition rate for the components of the bus load, and it also has a strong generalization for clustering, sub-line load data sample deviation and deviation caused by weather and temperature. ability.
本发明实施例还提供一种机器可读存储介质,该机器可读存储介质上存储有指令,该指令使得机器执行上述实施例所述的母线负荷构成辨识方法。An embodiment of the present invention also provides a machine-readable storage medium, which stores instructions on the machine-readable storage medium, and the instructions cause a machine to execute the bus load composition identification method described in the foregoing embodiment.
Claims (10)
- 一种母线负荷构成辨识方法,包括:A method for identifying bus load composition, including:确定母线负荷数据、所述母线负荷数据对应的子线负荷数据和所述母线负荷数据对应的子线负荷标签量构成的第一关系;Determining a first relationship constituted by bus load data, sub-line load data corresponding to the bus load data, and sub-line load label quantities corresponding to the bus load data;根据所述第一关系,建立所述母线负荷数据和所述母线负荷数据对应的子线负荷标签量的训练集;According to the first relationship, establish a training set of the bus load data and the sub-line load label quantity corresponding to the bus load data;构建深度卷积神经网络模型,通过所述训练集对所述深度卷积神经网络模型进行训练,以使在将母线负荷数据输入训练后的深度卷积神经网络模型的情况下,所述训练后的深度卷积神经网络模型输出与输入的母线负荷数据对应的子线负荷标签量。Construct a deep convolutional neural network model, and train the deep convolutional neural network model through the training set, so that when the bus load data is input to the trained deep convolutional neural network model, the trained deep convolutional neural network model The deep convolutional neural network model outputs the sub-line load label quantity corresponding to the input bus load data.
- 根据权利要求1所述的方法,其中,所述深度卷积神经网络模型包括多个卷积隐层,每个卷积隐层包含一组卷积核,且每个卷积核的参数包括卷积核长度、卷积核宽度、卷积核深度和卷积核数量。The method according to claim 1, wherein the deep convolutional neural network model includes a plurality of convolution hidden layers, each convolution hidden layer includes a set of convolution kernels, and the parameters of each convolution kernel include convolution The length of the core, the width of the core, the depth of the core, and the number of the core.
- 根据权利要求2所述的方法,其中,所述构建深度卷积神经网络模型,包括:The method according to claim 2, wherein said constructing a deep convolutional neural network model comprises:通过每个卷积隐层对所述深度卷积神经网络模型的输入量进行卷积计算和函数处理,得到每个卷积隐层对应的输出量;Performing convolution calculation and function processing on the input of the deep convolutional neural network model through each convolution hidden layer to obtain the output corresponding to each convolution hidden layer;将每个卷积隐层对应的输出量压缩成预设维度的输出量,并通过全连接神经网络输出所述预设维度的输出量。The output corresponding to each convolution hidden layer is compressed into an output of a preset dimension, and the output of the preset dimension is output through a fully connected neural network.
- 根据权利要求3所述的方法,其中,所述通过每个卷积隐层对所述深度卷积神经网络模型的输入量进行卷积计算,包括:The method according to claim 3, wherein the performing convolution calculation on the input of the deep convolutional neural network model through each convolution hidden layer comprises:针对每个卷积层的每个卷积核,通过第二关系对所述深度卷积神经网络模型的输入量进行卷积计算,其中,所述第二关系表示在所述卷积核的卷积核长度和卷积核宽度构成的二维基准下的卷积核权重与所述输入量的加权和。For each convolution kernel of each convolution layer, convolution calculation is performed on the input of the deep convolutional neural network model through a second relationship, where the second relationship represents the convolution of the convolution kernel The weighted sum of the weight of the convolution kernel under the two-dimensional reference formed by the length of the convolution kernel and the width of the convolution kernel and the input amount.
- 根据权利要求4所述的方法,其中,所述第二关系用下式表示:The method according to claim 4, wherein the second relationship is expressed by the following formula:其中,X new(i,j)表示输出数据,X(i,j)表示所述输入量,c表示所述卷积核长度,d表示所述卷积核宽度,ω(u,v)表示所述卷积核权重,b表示偏置参数。 Wherein, X new (i, j) represents output data, X(i, j) represents the input amount, c represents the length of the convolution kernel, d represents the width of the convolution kernel, and ω(u, v) represents The weight of the convolution kernel, b represents a bias parameter.
- 根据权利要求1所述的方法,还包括:The method according to claim 1, further comprising:针对不同类别的子线负荷标签量,建立母线负荷数据和所述母线负荷数据 对应的子线负荷标签量的测试集;For different types of sub-line load label amounts, establish a test set of bus load data and sub-line load label amounts corresponding to the bus load data;将所述测试集中的母线负荷数据输入至所述训练后的深度卷积神经网络模型中,根据所述测试集中的子线负荷标签量对所述训练后的深度卷积神经网络模型输出的子线负荷标签量的准确性进行测试。The bus load data in the test set is input into the trained deep convolutional neural network model, and the sub-line load label quantity in the test set is used for the output of the trained deep convolutional neural network model. The accuracy of the line load label volume is tested.
- 根据权利要求1所述的方法,其中,所述母线负荷数据通过以下方式生成:The method according to claim 1, wherein the bus load data is generated in the following manner:以设定的时间频率获取设定的时间长度下的多个子线负荷数据;Obtain multiple sub-line load data under the set time length at the set time frequency;对所述多个子线负荷数据进行聚类处理后,通过所述第一关系合成所述母线负荷数据。After performing clustering processing on the multiple sub-line load data, synthesize the bus-line load data through the first relationship.
- 根据权利要求1所述的方法,还包括:The method according to claim 1, further comprising:基于所述训练集中的母线负荷生成场景与实际母线负荷场景的偏差,对所述第一关系进行误差修正。Perform error correction on the first relationship based on the deviation between the bus load generation scenario in the training set and the actual bus load scenario.
- 根据权利要求8所述的方法,在所述基于训练集中的母线负荷生成场景与实际母线负荷场景的偏差,对所述第一关系进行误差修正之后,还包括:The method according to claim 8, after the error correction is performed on the first relationship based on the deviation between the bus load generation scene in the training set and the actual bus load scene, the method further comprises:根据修正后的第一关系,在所述训练集中加入误差数据,得到新的训练集;According to the revised first relationship, adding error data to the training set to obtain a new training set;在所述训练后的深度卷积神经网络模型中添加卷积隐层,并添加卷积残差块,得到新的深度卷积神经网络模型;Adding a convolutional hidden layer to the trained deep convolutional neural network model, and adding a convolutional residual block to obtain a new deep convolutional neural network model;通过所述新的训练集对所述新的深度卷积神经网络模型进行训练。Training the new deep convolutional neural network model through the new training set.
- 一种机器可读存储介质,存储有指令,所述指令使得机器执行如权利要求1至9中任一项所述的母线负荷构成辨识方法。A machine-readable storage medium storing instructions that cause a machine to execute the bus load composition identification method according to any one of claims 1 to 9.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010507240.6A CN111680744A (en) | 2020-06-05 | 2020-06-05 | Bus load composition identification method and machine readable storage medium |
CN202010507240.6 | 2020-06-05 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2021243930A1 true WO2021243930A1 (en) | 2021-12-09 |
Family
ID=72453946
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2020/124511 WO2021243930A1 (en) | 2020-06-05 | 2020-10-28 | Method for identifying composition of bus load, and machine-readable storage medium |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN111680744A (en) |
WO (1) | WO2021243930A1 (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111680744A (en) * | 2020-06-05 | 2020-09-18 | 中国电力科学研究院有限公司 | Bus load composition identification method and machine readable storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110490220A (en) * | 2019-07-02 | 2019-11-22 | 中国电力科学研究院有限公司 | A kind of bus load discrimination method and system |
WO2019227273A1 (en) * | 2018-05-28 | 2019-12-05 | Alibaba Group Holding Limited | Hierarchical concept based neural network model for data center power usage effectiveness prediction |
CN110689162A (en) * | 2019-08-12 | 2020-01-14 | 国电南瑞科技股份有限公司 | Bus load prediction method, device and system based on user side classification |
CN110932274A (en) * | 2019-12-18 | 2020-03-27 | 辽宁工业大学 | Power system measurement and load parameter analysis and identification method |
CN111680744A (en) * | 2020-06-05 | 2020-09-18 | 中国电力科学研究院有限公司 | Bus load composition identification method and machine readable storage medium |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106599417A (en) * | 2016-11-30 | 2017-04-26 | 中国电力科学研究院 | Method for identifying urban power grid feeder load based on artificial neural network |
CN108416695B (en) * | 2018-02-24 | 2020-07-07 | 合肥工业大学 | Power load probability density prediction method, system and medium based on deep learning |
CN110443302B (en) * | 2019-08-02 | 2023-06-09 | 天津相和电气科技有限公司 | Load identification method based on feature fusion and deep learning and application thereof |
-
2020
- 2020-06-05 CN CN202010507240.6A patent/CN111680744A/en active Pending
- 2020-10-28 WO PCT/CN2020/124511 patent/WO2021243930A1/en active Application Filing
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019227273A1 (en) * | 2018-05-28 | 2019-12-05 | Alibaba Group Holding Limited | Hierarchical concept based neural network model for data center power usage effectiveness prediction |
CN110490220A (en) * | 2019-07-02 | 2019-11-22 | 中国电力科学研究院有限公司 | A kind of bus load discrimination method and system |
CN110689162A (en) * | 2019-08-12 | 2020-01-14 | 国电南瑞科技股份有限公司 | Bus load prediction method, device and system based on user side classification |
CN110932274A (en) * | 2019-12-18 | 2020-03-27 | 辽宁工业大学 | Power system measurement and load parameter analysis and identification method |
CN111680744A (en) * | 2020-06-05 | 2020-09-18 | 中国电力科学研究院有限公司 | Bus load composition identification method and machine readable storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN111680744A (en) | 2020-09-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111460001B (en) | Power distribution network theoretical line loss rate evaluation method and system | |
CN116760772B (en) | Control system and method for converging flow divider | |
CN110942084A (en) | Loss reduction measure making method based on synchronous line loss abnormity identification | |
CN104050547A (en) | Non-linear optimization decision-making method of planning schemes for oilfield development | |
Li et al. | Grey relational decision making model of three-parameter interval grey number based on AHP and DEA | |
CN112884590A (en) | Power grid enterprise financing decision method based on machine learning algorithm | |
CN111339167A (en) | Method for analyzing influence factors of transformer area line loss rate based on K-means and principal component linear regression | |
WO2021243930A1 (en) | Method for identifying composition of bus load, and machine-readable storage medium | |
CN109599866A (en) | A kind of power system state estimation method of prediction auxiliary | |
Gao et al. | Real-time long-term voltage stability assessment based on eGBDT for large-scale power system with high renewables penetration | |
CN114219177A (en) | Computer room environment regulation and control method and device, electronic equipment and storage medium | |
CN112508254A (en) | Method for determining investment prediction data of transformer substation engineering project | |
CN109767037A (en) | A kind of prediction technique of summer Daily treatment cost | |
CN114372835B (en) | Comprehensive energy service potential customer identification method, system and computer equipment | |
CN106548199A (en) | A kind of large user's load characteristics clustering method based on demand response | |
He et al. | Prediction of listed companies' revenue based on model-fused | |
CN114329805A (en) | Connecting piece multidisciplinary collaborative design optimization method based on self-adaptive agent model | |
CN106886851A (en) | Electric energy meter standby redundancy Forecasting Methodology based on electric energy meter repair based on condition of component | |
Bowala et al. | Superiority of the neural network dynamic regression models for ontario electricity demand forecasting | |
Urama et al. | Application of Random Matrix Theory in Estimating Realistic Implied Correlation Matrix from Option Prices | |
CN117408573B (en) | UPFC equipment portrait display and performance analysis method and system | |
Haoru et al. | Risk Assessment of Internet Credit Based on Big Data Analysis | |
Zhou et al. | Sequential minimal optimization algorithm applied in short-term load forecasting | |
Li et al. | Accurate investment forecasting of power transmission and transformation project based on BP neural network and regression analysis | |
Peng et al. | A research on stock index prediction based on multiple linear regression and ELM neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 20938859 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 24.03.2023) |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 20938859 Country of ref document: EP Kind code of ref document: A1 |