CN113780679A - Load prediction method and device based on ubiquitous power Internet of things - Google Patents
Load prediction method and device based on ubiquitous power Internet of things Download PDFInfo
- Publication number
- CN113780679A CN113780679A CN202111138775.1A CN202111138775A CN113780679A CN 113780679 A CN113780679 A CN 113780679A CN 202111138775 A CN202111138775 A CN 202111138775A CN 113780679 A CN113780679 A CN 113780679A
- Authority
- CN
- China
- Prior art keywords
- data
- load prediction
- production element
- production
- long
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000004519 manufacturing process Methods 0.000 claims abstract description 167
- 238000000605 extraction Methods 0.000 claims abstract description 50
- 230000007787 long-term memory Effects 0.000 claims abstract description 21
- 230000006403 short-term memory Effects 0.000 claims abstract description 21
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 239000000284 extract Substances 0.000 claims abstract description 6
- 230000015654 memory Effects 0.000 claims description 27
- 230000006870 function Effects 0.000 claims description 19
- 230000004913 activation Effects 0.000 claims description 17
- 238000006243 chemical reaction Methods 0.000 claims description 14
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 9
- 230000006399 behavior Effects 0.000 claims description 8
- 230000005611 electricity Effects 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 4
- 238000012546 transfer Methods 0.000 claims description 4
- 230000007246 mechanism Effects 0.000 abstract description 9
- 230000007774 longterm Effects 0.000 abstract 1
- 230000008569 process Effects 0.000 description 9
- 238000012549 training Methods 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 5
- 238000004891 communication Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 238000010219 correlation analysis Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000013277 forecasting method Methods 0.000 description 2
- 230000000717 retained effect Effects 0.000 description 2
- 230000002411 adverse Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000013439 planning Methods 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000005612 types of electricity Effects 0.000 description 1
Images
Classifications
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Abstract
The invention discloses a load prediction method and a device based on a ubiquitous power Internet of things, wherein the method comprises the following steps: respectively acquiring data sets of a target user and other users in the ubiquitous power Internet of things range, and preprocessing the data sets; wherein the data set comprises electrical data, inter-user association data, production element data and load related data within a user; constructing a feature extraction model based on a depth residual error network, wherein the feature extraction model extracts correlation features between a first data set and electrical data of a target user; and constructing a load prediction model based on the long-term and short-term memory network, and performing load prediction on the load prediction model according to a second data set to obtain the load data of the target user. The load prediction method and the load prediction system can reduce the uncertainty of the load prediction and improve the load prediction accuracy of the target user through the load prediction model of the long-term and short-term memory network based on the attention mechanism and the long-term and short-term jump connection.
Description
Technical Field
The invention relates to the technical field of power systems, in particular to a load prediction method and device based on a ubiquitous power internet of things.
Background
The 5G technology provides technical support for communication of mass data, high frequency data, low time delay data and low energy consumption data, and is based on the fact that a data acquisition system and a data communication system are gradually installed on each device, each user of a power grid and even each electric device in each user in the future. Meanwhile, in recent years, the blowout type development of technologies such as big data technology, deep learning method, optimization control technology, cloud computing and edge computing also provides technical support for data mining.
When load prediction is performed on electrical data of a target user, some of the prior art schemes improve the accuracy of load prediction by using mathematical modeling or artificial intelligence technology, and some schemes improve by introducing new influence factors, such as common data types of electricity price, weather, temperature, humidity and the like.
In the load prediction model in the prior art, due to the fact that the considered influence factors are low in dimensionality, input information is incomplete, complete data characteristics are difficult to cover, the solution of the load prediction model is high in uncertainty, the accuracy of predicted power grid load is not enough, and operation management and scheduling planning of a power grid are not facilitated.
Disclosure of Invention
The invention aims to provide a load prediction method and a load prediction device based on a ubiquitous power Internet of things, and the load prediction method and the load prediction device are used for solving the technical problem that the accuracy of a load prediction result of a target user is insufficient in the prior art.
The purpose of the invention can be realized by the following technical scheme:
a load prediction method based on a ubiquitous power Internet of things comprises the following steps:
respectively acquiring data sets of a target user and other users in the ubiquitous power Internet of things range, and preprocessing the data sets; wherein the data set comprises electrical data, inter-user association data, production element data and load related data within a user;
constructing a feature extraction model based on a depth residual error network, wherein the feature extraction model extracts correlation features between a first data set and electrical data of a target user; wherein the first data set comprises electrical data of other users, the inter-user correlation data and the load-related data;
and constructing a load prediction model based on a long-term and short-term memory network, wherein the load prediction model carries out load prediction according to a second data set to obtain the load data of the target user, the second data set comprises the first data set, the associated characteristics, the residual production element data and the residual production element type, and the residual production element data and the residual production element type are determined according to the production element data.
Optionally, before constructing the load prediction model based on the long-term and short-term memory network, the method further includes:
and calculating the similarity among all the production elements according to the production element data, removing the production elements with the similarity exceeding a preset threshold value, taking the data of other production elements as residual production element data, and clustering the residual production element data to obtain the residual production element category.
Optionally, calculating the similarity between all production elements according to the production element data includes:
using formulas
Calculating the similarity between the two production elements;
where I (X, Y) is the similarity between production elements X, Y, X, Y are production elements, X is the data of production element X, Y is the data of production element Y, p (X) is the probability of X occurring in all events including production element X, p (Y) is the probability of Y occurring in all events including production element Y, and p (X, Y) is the probability of X and Y occurring simultaneously in all events.
Optionally, the inter-user association data includes:
network topology association relation, user service association degree and electricity utilization behavior similarity;
the user service association degree represents the association degree of service traffic among users, and the electricity utilization behavior similarity is obtained by clustering the electrical data among the users.
Optionally, the extracting the associated features between the first data set and the electrical data of the target user by the feature extraction model comprises:
arranging data in the first data set into a multi-channel matrix form, extracting shallow layer correlation features through a convolutional layer and an activation function layer, extracting deep layer correlation features through a plurality of residual blocks, and performing weighted summation on the shallow layer correlation features and the deep layer correlation features through a first global long jump connection to obtain the correlation features; the first global long-hop connection spans several residual blocks and one convolutional layer.
Optionally, the load predicting by the load prediction model according to the second data set to obtain the load data of the target user includes:
arranging the data in the second data set into a multi-channel matrix form, carrying out nonlinear conversion through a convolution layer and an activation function layer to obtain a nonlinear conversion result, carrying out characteristic learning and transfer through a plurality of long-term and short-term memory network layers, and outputting the load data of the target user through a batch normalization layer, the activation function layer and a full connection layer.
Optionally, the learning and transferring of features through several long-short term memory network layers comprises:
the long and short term memory network basic unit group keeps learned knowledge or new learning knowledge through local short hop connection;
the long-short term memory network layer reserves the learned knowledge or the new learned knowledge through local long-jump connection;
superposing the nonlinear conversion result and the learning result of the long-short term memory network layer through a second global long-jump connection;
wherein the local short-hop connection spans a set of long-short term memory network primitives, the local long-hop connection spans a long-short term memory network layer, the second global long-hop connection spans multiple long-short term memory network layers, and the long-short term memory network layer comprises multiple sets of long-short term memory network primitives.
The invention also provides a load prediction device based on the ubiquitous power internet of things, which comprises the following components:
the data set acquisition module is used for respectively acquiring data sets of a target user and other users in the ubiquitous power Internet of things range and preprocessing the data sets; wherein the data set comprises electrical data, inter-user association data, production element data and load related data within a user;
the characteristic extraction module is used for constructing a characteristic extraction model based on a depth residual error network, and the characteristic extraction model is used for extracting the correlation characteristics between the first data set and the electrical data of the target user; wherein the first data set comprises electrical data of other users, the inter-user correlation data and the load-related data;
and the load prediction module is used for constructing a load prediction model based on a long-term and short-term memory network, the load prediction model carries out load prediction according to a second data set to obtain the load data of the target user, the second data set comprises the first data set, the association characteristics, the residual production element data and the residual production element type, and the residual production element data and the residual production element type are determined according to the production element data.
Optionally, the method further comprises:
and the production element data processing module is used for calculating the similarity among all the production elements according to the production element data, removing the production elements with the similarity exceeding a preset threshold value, taking the data of other production elements as residual production element data, and clustering the residual production element data to obtain the residual production element category.
Optionally, the calculating, by the production element data processing module, the similarity between all production elements according to the production element data includes:
using formulas
Calculating the similarity between the two production elements;
where I (X, Y) is the similarity between production elements X, Y, X, Y are production elements, X is the data of production element X, Y is the data of production element Y, p (X) is the probability of X occurring in all events including production element X, p (Y) is the probability of Y occurring in all events including production element Y, and p (X, Y) is the probability of X and Y occurring simultaneously in all events.
The invention provides a load prediction method and a device based on a ubiquitous power Internet of things, wherein the method comprises the following steps: respectively acquiring data sets of a target user and other users in the ubiquitous power Internet of things range, and preprocessing the data sets; wherein the data set comprises electrical data, inter-user association data, production element data and load related data within a user; constructing a feature extraction model based on a depth residual error network, wherein the feature extraction model extracts correlation features between a first data set and electrical data of a target user; wherein the first data set comprises electrical data of other users, the inter-user correlation data and the load-related data; and constructing a load prediction model based on a long-term and short-term memory network, wherein the load prediction model carries out load prediction according to a second data set to obtain the load data of the target user, the second data set comprises the first data set, the associated characteristics, the residual production element data and the residual production element type, and the residual production element data and the residual production element type are determined according to the production element data.
In view of this, the technical scheme of the invention brings the following beneficial effects:
(1) according to the load forecasting method, the load forecasting of the target user is carried out according to various data of multiple users in the range of the ubiquitous Internet of things, and the potential characteristics among various data in the Internet of things are fully excavated by utilizing high-dimensional characteristics, so that the uncertainty of the load forecasting can be reduced;
(2) influence of other users on the target user is fully mined through the inter-user associated data, a feature extraction model is built to extract associated features between the first data set and the target user electrical data, and load prediction accuracy of the target user can be improved;
(3) the influence of the production elements on the load is fully mined through the production element data in the user, and the production element data are classified, so that the knowledge extraction of the production element data can be improved, and the efficiency of load prediction is improved;
(4) through the load prediction model based on the long-term and short-term memory network, the knowledge learning and memory capabilities of data in different periods can be improved, and the load prediction can be accurately performed on the target user.
Drawings
FIG. 1 is a schematic flow chart of a load forecasting method according to the present invention;
FIG. 2 is a schematic structural diagram of a feature extraction model based on a depth residual error network according to the present invention;
FIG. 3 is a schematic structural diagram of a load prediction model based on a long-term and short-term memory network according to the present invention;
FIG. 4 is a schematic diagram of the structure of the LSTM basic unit in the present invention;
fig. 5 is a schematic structural diagram of the load prediction apparatus of the present invention.
Detailed Description
The embodiment of the invention provides a load prediction method and device based on a ubiquitous power Internet of things, and aims to solve the technical problem that the accuracy of a load prediction result of a target user is insufficient in the prior art.
To facilitate an understanding of the invention, the invention will now be described more fully with reference to the accompanying drawings. Preferred embodiments of the present invention are shown in the drawings. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
Referring to fig. 1, an embodiment of a load prediction method based on a ubiquitous power internet of things according to the present invention includes:
s1: respectively acquiring data sets of a target user and other users in the ubiquitous power Internet of things range, and preprocessing the data sets; wherein the data set comprises electrical data, inter-user association data, production element data and load related data within a user;
s2: constructing a feature extraction model based on a depth residual error network, wherein the feature extraction model extracts correlation features between a first data set and electrical data of a target user; wherein the first data set comprises electrical data of other users, the inter-user correlation data and the load-related data;
s3: and constructing a load prediction model based on a long-term and short-term memory network, wherein the load prediction model carries out load prediction according to a second data set to obtain the load data of the target user, the second data set comprises the first data set, the associated characteristics, the residual production element data and the residual production element type, and the residual production element data and the residual production element type are determined according to the production element data.
In this embodiment, the range division method of the ubiquitous power internet of things is various, and mainly includes the following five methods: the system comprises a same geographical position, a same administrative plate, a same business circle, a building group with the same attribute and a same platform area.
In step S1, users in the range of the power internet of things are classified into target users and other users, the target users are targets for load prediction, data sets of the target users and the other users in the range of the power internet of things are respectively obtained, the data sets include electrical data, inter-user association data, production factor data and load related data inside the users, and the obtained data sets are preprocessed.
Wherein, the electrical data mainly includes: active power, reactive power, electric energy, frequency, voltage, current, harmonic wave, opening and closing states and the like; the associated data among users mainly comprises: network topology association relation, user service association degree and electricity utilization behavior similarity; the user internal production element data includes: the method comprises the following steps of (1) user investment condition, user profit and loss condition, user internal object purchase condition, user internal object use condition, user internal object transfer condition, user internal human resource allocation condition, user main service completion condition, power utilization condition of different user internal electric appliances and the like; the load related data refers to other data related to user load, and mainly comprises the following data: weather, temperature, humidity, electricity prices, working days, holidays, and the like.
The user service association degree represents the association degree of service traffic between users. For example, when there is a business relationship between users A, B, the number of business calls between users A, B and the importance of each business call are weighted and summed to obtain the user B's importance to user ADegree of quasi-association CB(ii) a The same method can obtain the quasi-association degree of other users except the user B to the user A, and further obtain the sum C of the quasi-association degrees of other users to the user Asum(ii) a Calculating the degree of quasi-association CBSum of degree of quasi-correlation CsumThe ratio is used as the service association degree of the user B to the user A.
Wherein, the similarity of the electricity utilization behaviors is obtained by analyzing the electrical data of the user. Firstly, a clustering center curve of the electrical data of the users is obtained through a clustering method, then the distance of the clustering center curve of the unused users is calculated, the similarity degree of the user behaviors is measured by utilizing the distance, and the smaller the distance is, the more similar the electricity utilization behaviors among the users are.
Specifically, in this embodiment, a K-means clustering method is used, and the number of clusters K is first given1Then randomly selecting K data as clustering centers, distributing the rest data to the nearest clustering centers, wherein each clustering center represents a cluster; when one piece of data is distributed, the clustering center recalculates the clustering center according to the existing data in the corresponding cluster, and the calculation target is to minimize the sum of squares of distances from other data in the cluster to the clustering center; the convergence condition of the k-means method is that the cluster center is not changed any more, i.e., the sum of the squares of the distances from each data to the corresponding cluster center is minimized. In this embodiment, the distance of the clustering center curves of different users can be calculated by using the cosine distance.
In this embodiment, because the production elements in the power industry are of a plurality of types and have a large data volume, the similarity between all the production elements is calculated according to the production element data, the production elements with the similarity exceeding a preset threshold are removed, the data of other production elements are used as the remaining production element data, and the remaining production element data are clustered to obtain the category of the remaining production elements.
Specifically, first, the formula is used:
calculating the similarity between the two production elements;
wherein X, Y are production elements, I (X, Y) is the similarity between production elements X, Y, X is the data of production element X, Y is the data of production element Y, p (X) is the probability of X occurring in all events containing production element X, p (Y) is the probability of Y occurring in all events containing production element Y, and p (X, Y) is the probability of X and Y occurring simultaneously in all events.
Then, performing correlation analysis by using a mutual information method, and eliminating production elements with high similarity to reduce information redundancy; because linear or nonlinear correlation exists among production elements in the industry, the mutual information method can capture any type of correlation, so that the mutual information method is used for performing correlation analysis on the production elements in the industry, removing the production elements with similarity exceeding a preset threshold, retaining data of the remaining production elements, and clustering the data of the remaining production elements by using a clustering algorithm to obtain the category of the remaining production elements.
It should be noted that, in step S3, clustering may be performed using a K-means method, and the number of clusters is set to K2The clustering process is the same as the k-means method of step S1, and is not described herein again.
Referring to fig. 2, in step S2, a feature extraction model based on a deep residual error network is constructed, the first data set is used as input data of the feature extraction model, and the electrical data of the target user is used as output data of the feature extraction model, so that the feature extraction model is trained. Wherein the first data set comprises electrical data of other users, inter-user correlation data and load related data.
Specifically, input data of the feature extraction model is arranged into a multi-channel matrix form according to types, each channel corresponds to one type of data, and the input data is rearranged into the matrix form in a time sequence form according to the sequence of the front row and the rear row. It should be noted that the type refers to data of a lower layer across which the input data crosses, for example, the input data is "load-related data", and then the type specifically refers to weather, temperature, humidity, and the like in the load-related data.
In the embodiment, when the feature extraction model is trained, shallow layer correlation features are extracted through a convolution layer and an activation function layer of a deep residual error network, and in the preferred embodiment, the number of layers of the convolution layer and the activation function layer is 1-3; each residual block internally comprises a convolution layer, an activation function layer and a batch normalization layer, the first end and the last end of each residual block are provided with short jump connections, and deep layer characteristics are extracted through n residual blocks; the first global long jump connection spans n residual blocks and one convolutional layer, and the shallow layer feature and the deep layer feature are subjected to weighted summation through the first global long jump connection to obtain an associated feature; and the correlation characteristics output the electrical data of the target user through the batch normalization layer and the activation function layer.
It is worth to be noted that the first global long-jump connection and the short-jump connection both have the function of improving the learning ability of the feature extraction model in order to avoid the disappearance of the gradient. The number n of the residual blocks is a hyper-parameter of the feature extraction model, is artificially set before the model training and is artificially adjusted according to a plurality of iterative training results. In each iteration of the feature extraction model training process, a convolution kernel in convolution operation is a parameter of the feature extraction model, after each iteration training, output target user electrical data is compared with obtained target user data to obtain an error between the output target user electrical data and the obtained target user data, the convolution kernel in the convolution operation is updated according to error feedback, finally the error is minimized, and the feature extraction model is trained well at the moment.
And extracting shallow correlation characteristics and deep correlation characteristics between the input data and the target user electrical data by using the trained characteristic extraction model, and taking the obtained shallow correlation characteristics and deep correlation characteristics as one of the inputs of the load prediction model in the step S3.
Similarly, the output data of the feature extraction model is arranged in a multi-channel matrix form according to types, and each channel corresponds to one type of electrical data. The type of the electrical data of the target user is output by the characteristic extraction model, and the type of the electrical data is active power, reactive power, electric energy, frequency, voltage, current, harmonic wave, opening and closing state and the like spanned by the output.
It is worth mentioning that the shallow correlation feature represents a shallow correlation between the input data (i.e. the first data set) and the output data (i.e. the electrical data of the target user) of the feature extraction model, and the deep correlation feature represents a deep correlation between the input data and the output data of the feature extraction model.
It is worth noting that the deeper the number of network layers of the neural network model, the more knowledge is learned from the input data, so shallow or deep correlations are only related to the number of network layers. In the feature extraction model of this embodiment, the shallow layer network is composed of 1-3 convolutional layers and an activation function layer, the intermediate output of the last layer of the shallow layer network is a shallow layer correlation feature, the deep layer network is composed of a plurality of residual blocks and one convolutional layer, and the intermediate output of the last layer of the deep layer network is a deep layer correlation feature.
Referring to fig. 3, in step S3, a load prediction model based on a Long-Short Term Memory network (LSTM) is constructed, input data of the load prediction model is a second data set, and the second data set includes: the input data of the feature extraction model is the first data set, the shallow and deep associated features extracted by the feature extraction model, the residual production element data and the residual production element category. Specifically, similar to the feature extraction model in step S2, the input data of the load prediction model in step S3 is arranged in a matrix form of multiple channels, each channel corresponds to one type of data, and the data in each channel is rearranged in the matrix form in the order of the preceding and following rows.
The output data of the load prediction model in step S3 is the load data of the target user, and the arrangement form of the output data is a time-series form.
In this embodiment, the process of training the load prediction model is as follows:
firstly, carrying out nonlinear conversion by utilizing a shallow network to convert input data into a nonlinear space so as to more effectively form mapping with output data;
then, a deep network comprising a plurality of LSTM basic unit groups and three kinds of jump connection is constructed, the jump connection reserves the learning result of the middle process to the rear level by crossing multiple layers, and the effect is that the learning result of the starting point of the jump connection can be reserved even if the part spanned by the starting point and the end point of the jump connection cannot learn useful knowledge, so that the problem that the prediction precision is difficult to improve because the LSTM basic unit groups are stacked in the traditional method is solved;
and finally, the knowledge is retained and fused by using the global long jump connection, and finally, the mapping is formed with the output.
It is worth noting that m, p in FIG. 3mThe hyper-parameters of the load prediction model are artificially set before training and are adjusted according to the result of multiple iterative training. And the parameters to be adjusted in the training process are the weights of the internal parameters of a forgetting gate, an input gate and an output gate in the lstm basic unit and convolution kernels in convolution operation, and after each iterative training, the parameters are updated in a feedback mode according to the error between the output load prediction result and the actual load, so that the error is minimum finally.
Specifically, when the load prediction model performs nonlinear conversion by using a shallow network, the nonlinear conversion of the input data is realized by using a plurality of convolutional layers and activation function layers which form the shallow network, and in a preferred embodiment, the shallow network is composed of 1-3 convolutional layers and activation function layers, and the nonlinear conversion of the input data is realized by using 1-3 convolutional layers and activation function layers. Because the input data types are multiple and the feature dimension is high, the load prediction model utilizes an attention mechanism to realize effective discrimination and weighting processing of important features and non-important features, and the weight is determined by the contribution of the features to the output, namely the gradient of the features to the output.
It should be noted that, for the load prediction model, the input of each layer may be referred to as a feature of the data, the input data is referred to as an input feature, the data of the intermediate process is referred to as an intermediate feature or a potential feature, and a plurality of local features are also included in the input feature and the potential feature. Therefore, in this embodiment, the feature of attention mechanism processing not only refers to the shallow-layer related feature and the deep-layer related feature extracted by the feature extraction model, but also includes all the input features and the potential features.
Specifically, after the nonlinear conversion, the embodiment constructs a deep network including a plurality of LSTM base unit groups and three types of hopping connections (second global long-hopping connection, local long-hopping connection, and local short-hopping connection), and implements learning and transfer of features through m LSTM layers.
Furthermore, the starting point and the end point of the second global long jump connection span m LSTM layers, the nonlinear conversion result of the shallow layer network can be superposed with the learning results of the m LSTM layers, and the superposed intermediate result is ensured not to lose the shallow layer characteristics due to deepening of the network level;
furthermore, the starting point and the end point of the local long jump connection span 1 LSTM layer, so that each LSTM layer can not cause adverse effect on a subsequent network, namely learned knowledge can be retained or new knowledge can be learned. Each LSTM layer comprises pmAn LSTM basic unit group, and pmEach LSTM elementary unit group is divided into several groups, i.e. each LSTM layer comprises several groups of LSTM elementary unit groups.
Further, the starting point and the end point of the local short-hop connection span each group of the LSTM elementary units, ensuring that the part covered by the local short-hop connection can retain learned knowledge or learn new knowledge. Each LSTM base unit group includes an LSTM base unit layer, an attention mechanism layer, an activation function layer, and a Dropout layer.
It should be noted that, the load prediction model in this embodiment needs to perform a convolution operation after m LSTM layers, and its purpose is to map the learning results of the m LSTM layers to the nonlinear space after the nonlinear conversion of the shallow network, so that the nonlinear conversion result of the shallow network and the learning results of the m LSTM layers are superimposed, and it is ensured that the intermediate result obtained after the superimposition does not lose the shallow feature due to deepening of the network hierarchy.
It should be noted that fig. 2 and fig. 3 are similar in overall structure, but the core portions are different (fig. 2 is a residual block, fig. 3 is an LSTM layer), resulting in different functions (fig. 2 is feature extraction, fig. 3 is load prediction). In the neural network, adding convolutional layer is equivalent to performing nonlinear transformation, so the front part of fig. 2 performs nonlinear transformation to obtain shallow features, and the front part of fig. 3 performs nonlinear transformation to convert the input data into a nonlinear space first, so as to form an effective mapping with the output data, thereby improving the prediction capability of the conventional LSTM method.
The LSTM base unit has a certain memory capacity by the mutual cooperation of the forgetting gate, the input gate, and the output gate, and can learn data for a certain period of time, and therefore, the LSTM base unit is widely used for load prediction. Conventional approaches implement load prediction by either a limited number of LSTM primitives or stacking multiple sets of LSTM primitives (deepening hierarchy). However, as the hierarchy deepens and the data period becomes longer, the prediction accuracy is improved to meet the bottleneck for two reasons: firstly, the data with overlong processing time interval of the LSTM basic unit can forget the previous part; secondly, as the hierarchy is deepened, the problem that the gradient disappears can occur, namely, the knowledge learned by the front part of the network is difficult to store in the rear part of the network.
Referring to fig. 4, the LSTM basic unit provided in this embodiment specifically includes a forgetting gate, an input gate, and an output gate, where x (t) is input data at time t; s (t-1) and S (t) are state memory units at t-1 and t moments respectively; h (t-1) and h (t) are hidden layer intermediate outputs at the time t-1 and t, respectively; sigma is a Sigmoid nonlinear activation function;
is a non-linear activation function; an inner product operation; the forgetting gate determines a reserved part and a forgetting part of the state memory unit at the time t-1; the input gate and the forgetting gate jointly determine the updating of the state memory unit at the time t; the output gate determines the hidden layer intermediate output at time t.
It should be noted that, in the load prediction model in this embodiment, each LSTM layer introduces a local long jump connection and a local short jump connection, which can provide knowledge learning and memory capabilities for different periods of data, and each LSTM layer introduces an attention mechanism, which can perform importance determination and give different weights to different periods of data that affect the prediction result. In the embodiment, a load prediction model of a long-time memory network based on an attention mechanism and three types of jump connection is built, an LSTM basic unit is introduced, excellent learning capacity of the LSTM basic unit for data in a certain period is reserved, and meanwhile, three types of jump connection modes are introduced, so that knowledge can be reserved or learned to new knowledge when the network is deepened, the memory capacity and learning capacity of the network are improved, and data in a longer period can be learned.
The LSTM basic unit group in the embodiment comprises an attention mechanism layer, and the constructed long-term and short-term memory network comprises a plurality of LSTM basic unit groups, so that the attention mechanism is added at a plurality of positions in the whole load prediction process, and the load prediction model can realize effective discrimination and weighting processing of important features and non-important features by using the attention mechanism.
The embodiment provides a load prediction method based on a ubiquitous power internet of things, which is used for predicting loads of target users according to various data of multiple users in the range of the ubiquitous power internet of things, and reducing uncertainty of load prediction by fully exploiting potential features among various data in the internet of things through high-dimensional features; influence of other users on the target user is fully mined through the inter-user associated data, a feature extraction model is built to extract associated features between the first data set and the target user electrical data, and load prediction accuracy of the target user can be improved; the influence of the production elements on the load is fully mined through the production element data in the user, and the production element data are classified, so that the knowledge extraction of the production element data can be improved, and the efficiency of load prediction is improved; through the load prediction model based on the long-term and short-term memory network, the knowledge learning and memory capabilities of data in different periods can be improved, and the load prediction can be accurately performed on the target user.
Referring to fig. 5, the present invention further provides an embodiment of a load prediction apparatus based on the internet of things of ubiquitous power, including:
the data set acquisition module 11 is configured to acquire data sets of a target user and other users in a ubiquitous power internet of things range, and preprocess the data sets; wherein the data set comprises electrical data, inter-user association data, production element data and load related data within a user;
the feature extraction module 12 is configured to construct a feature extraction model based on a deep residual error network, where the feature extraction model extracts correlation features between the first data set and the electrical data of the target user; wherein the first data set comprises electrical data of other users, the inter-user correlation data and the load-related data;
and the load prediction module 13 is configured to construct a load prediction model based on a long-term and short-term memory network, where the load prediction model performs load prediction according to a second data set to obtain load data of the target user, the second data set includes the first data set, the associated features, remaining production element data and a remaining production element category, and the remaining production element data and the remaining production element category are determined according to the production element data.
Optionally, the method further comprises:
and the production element data processing module is used for calculating the similarity among all the production elements according to the production element data, removing the production elements with the similarity exceeding a preset threshold value, taking the data of other production elements as residual production element data, and clustering the residual production element data to obtain the residual production element category.
Optionally, the calculating, by the production element data processing module, the similarity between all production elements according to the production element data includes:
using formulas
Calculating the similarity between the two production elements;
where I (X, Y) is the similarity between production elements X, Y, X, Y are production elements, X is the data of production element X, Y is the data of production element Y, p (X) is the probability of X occurring in all events including production element X, p (Y) is the probability of Y occurring in all events including production element Y, and p (X, Y) is the probability of X and Y occurring simultaneously in all events.
The embodiment provides a load prediction device based on a ubiquitous power internet of things, which is used for predicting the load of a target user according to various data of a plurality of users in the range of the ubiquitous power internet of things, and fully exploiting potential features among various data in the internet of things by using high-dimensional features, so that the uncertainty of load prediction can be reduced; influence of other users on the target user is fully mined through the inter-user associated data, a feature extraction model is built to extract associated features between the first data set and the target user electrical data, and load prediction accuracy of the target user can be improved; the influence of the production elements on the load is fully mined through the production element data in the user, and the production element data are classified, so that the knowledge extraction of the production element data can be improved, and the efficiency of load prediction is improved; through the load prediction model based on the long-term and short-term memory network, the knowledge learning and memory capabilities of data in different periods can be improved, and the load prediction can be accurately performed on the target user.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above described systems, systems and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the embodiments provided in the present application, it should be understood that the disclosed system, system and method may be implemented in other ways. For example, the above-described system embodiments are merely illustrative, and for example, the division of the units is only one logical functional division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, systems or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A load prediction method based on a ubiquitous power Internet of things is characterized by comprising the following steps:
respectively acquiring data sets of a target user and other users in the ubiquitous power Internet of things range, and preprocessing the data sets; wherein the data set comprises electrical data, inter-user association data, production element data and load related data within a user;
constructing a feature extraction model based on a depth residual error network, wherein the feature extraction model extracts correlation features between a first data set and electrical data of a target user; wherein the first data set comprises electrical data of other users, the inter-user correlation data and the load-related data;
and constructing a load prediction model based on a long-term and short-term memory network, wherein the load prediction model carries out load prediction according to a second data set to obtain the load data of the target user, the second data set comprises the first data set, the associated characteristics, the residual production element data and the residual production element type, and the residual production element data and the residual production element type are determined according to the production element data.
2. The ubiquitous power internet of things-based load prediction method according to claim 1, wherein before constructing the long-term and short-term memory network-based load prediction model, the method further comprises:
and calculating the similarity among all the production elements according to the production element data, removing the production elements with the similarity exceeding a preset threshold value, taking the data of other production elements as residual production element data, and clustering the residual production element data to obtain the residual production element category.
3. The ubiquitous power internet of things-based load prediction method according to claim 2, wherein calculating the similarity between all production elements according to the production element data comprises:
using formulas
Calculating the similarity between the two production elements;
where I (X, Y) is the similarity between production elements X, Y, X, Y are production elements, X is the data of production element X, Y is the data of production element Y, p (X) is the probability of X occurring in all events including production element X, p (Y) is the probability of Y occurring in all events including production element Y, and p (X, Y) is the probability of X and Y occurring simultaneously in all events.
4. The ubiquitous power internet of things-based load prediction method according to claim 1, wherein the inter-user correlation data comprises:
network topology association relation, user service association degree and electricity utilization behavior similarity;
the user service association degree represents the association degree of service traffic among users, and the electricity utilization behavior similarity is obtained by clustering the electrical data among the users.
5. The ubiquitous power internet of things-based load prediction method according to claim 1, wherein the extracting, by the feature extraction model, the associated features between the first data set and the electrical data of the target user comprises:
arranging data in the first data set into a multi-channel matrix form, extracting shallow layer correlation features through a convolutional layer and an activation function layer, extracting deep layer correlation features through a plurality of residual blocks, and performing weighted summation on the shallow layer correlation features and the deep layer correlation features through a first global long jump connection to obtain the correlation features; the first global long-hop connection spans several residual blocks and one convolutional layer.
6. The ubiquitous power internet of things-based load prediction method according to claim 1, wherein the load prediction model performing load prediction according to a second data set to obtain the load data of the target user comprises:
arranging the data in the second data set into a multi-channel matrix form, carrying out nonlinear conversion through a convolution layer and an activation function layer to obtain a nonlinear conversion result, carrying out characteristic learning and transfer through a plurality of long-term and short-term memory network layers, and outputting the load data of the target user through a batch normalization layer, the activation function layer and a full connection layer.
7. The ubiquitous power internet of things-based load prediction method according to claim 6, wherein the learning and transferring of the features through the plurality of long-term and short-term memory network layers comprises:
the long and short term memory network basic unit group keeps learned knowledge or new learning knowledge through local short hop connection;
the long-short term memory network layer reserves the learned knowledge or the new learned knowledge through local long-jump connection;
superposing the nonlinear conversion result and the learning result of the long-short term memory network layer through a second global long-jump connection;
wherein the local short-hop connection spans a set of long-short term memory network primitives, the local long-hop connection spans a long-short term memory network layer, the second global long-hop connection spans multiple long-short term memory network layers, and the long-short term memory network layer comprises multiple sets of long-short term memory network primitives.
8. A load prediction device based on ubiquitous power Internet of things is characterized by comprising:
the data set acquisition module is used for respectively acquiring data sets of a target user and other users in the ubiquitous power Internet of things range and preprocessing the data sets; wherein the data set comprises electrical data, inter-user association data, production element data and load related data within a user;
the characteristic extraction module is used for constructing a characteristic extraction model based on a depth residual error network, and the characteristic extraction model is used for extracting the correlation characteristics between the first data set and the electrical data of the target user; wherein the first data set comprises electrical data of other users, the inter-user correlation data and the load-related data;
and the load prediction module is used for constructing a load prediction model based on a long-term and short-term memory network, the load prediction model carries out load prediction according to a second data set to obtain the load data of the target user, the second data set comprises the first data set, the association characteristics, the residual production element data and the residual production element type, and the residual production element data and the residual production element type are determined according to the production element data.
9. The ubiquitous power internet of things-based load prediction device according to claim 8, further comprising:
and the production element data processing module is used for calculating the similarity among all the production elements according to the production element data, removing the production elements with the similarity exceeding a preset threshold value, taking the data of other production elements as residual production element data, and clustering the residual production element data to obtain the residual production element category.
10. The ubiquitous power internet of things-based load prediction device according to claim 8, wherein the production element data processing module calculating the similarity between all production elements from the production element data comprises:
using formulas
Calculating the similarity between the two production elements;
where I (X, Y) is the similarity between production elements X, Y, X, Y are production elements, X is the data of production element X, Y is the data of production element Y, p (X) is the probability of X occurring in all events including production element X, p (Y) is the probability of Y occurring in all events including production element Y, and p (X, Y) is the probability of X and Y occurring simultaneously in all events.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111138775.1A CN113780679B (en) | 2021-09-27 | 2021-09-27 | Load prediction method and device based on ubiquitous power Internet of things |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111138775.1A CN113780679B (en) | 2021-09-27 | 2021-09-27 | Load prediction method and device based on ubiquitous power Internet of things |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113780679A true CN113780679A (en) | 2021-12-10 |
| CN113780679B CN113780679B (en) | 2024-04-16 |
Family
ID=78853955
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111138775.1A Active CN113780679B (en) | 2021-09-27 | 2021-09-27 | Load prediction method and device based on ubiquitous power Internet of things |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113780679B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117134470A (en) * | 2023-10-26 | 2023-11-28 | 珠海市嘉德电能科技有限公司 | Sleep control method and related device of BMS battery management system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109299401A (en) * | 2018-07-12 | 2019-02-01 | 中国海洋大学 | Metropolitan area space-time stream Predicting Technique based on deep learning model LSTM-ResNet |
| CN110889545A (en) * | 2019-11-20 | 2020-03-17 | 国网重庆市电力公司电力科学研究院 | Power load prediction method and device and readable storage medium |
| CN111104599A (en) * | 2019-12-23 | 2020-05-05 | 北京百度网讯科技有限公司 | Method and apparatus for outputting information |
| CN112200031A (en) * | 2020-09-27 | 2021-01-08 | 上海眼控科技股份有限公司 | Network model training method and equipment for generating image corresponding word description |
| CN112364975A (en) * | 2020-10-14 | 2021-02-12 | 山东大学 | Terminal operation state prediction method and system based on graph neural network |
-
2021
- 2021-09-27 CN CN202111138775.1A patent/CN113780679B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109299401A (en) * | 2018-07-12 | 2019-02-01 | 中国海洋大学 | Metropolitan area space-time stream Predicting Technique based on deep learning model LSTM-ResNet |
| CN110889545A (en) * | 2019-11-20 | 2020-03-17 | 国网重庆市电力公司电力科学研究院 | Power load prediction method and device and readable storage medium |
| CN111104599A (en) * | 2019-12-23 | 2020-05-05 | 北京百度网讯科技有限公司 | Method and apparatus for outputting information |
| CN112200031A (en) * | 2020-09-27 | 2021-01-08 | 上海眼控科技股份有限公司 | Network model training method and equipment for generating image corresponding word description |
| CN112364975A (en) * | 2020-10-14 | 2021-02-12 | 山东大学 | Terminal operation state prediction method and system based on graph neural network |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117134470A (en) * | 2023-10-26 | 2023-11-28 | 珠海市嘉德电能科技有限公司 | Sleep control method and related device of BMS battery management system |
| CN117134470B (en) * | 2023-10-26 | 2024-02-13 | 珠海市嘉德电能科技有限公司 | Sleep control method and related device of BMS battery management system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113780679B (en) | 2024-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| Fan et al. | Development of prediction models for next-day building energy consumption and peak power demand using data mining techniques | |
| Hadavandi et al. | Integration of genetic fuzzy systems and artificial neural networks for stock price forecasting | |
| Hong et al. | SVR with hybrid chaotic genetic algorithms for tourism demand forecasting | |
| Froelich et al. | Fuzzy cognitive maps in the modeling of granular time series | |
| Zeng et al. | A learning framework based on weighted knowledge transfer for holiday load forecasting | |
| Han et al. | Short-term forecasting of individual residential load based on deep learning and K-means clustering | |
| Hong et al. | Locational marginal price forecasting using deep learning network optimized by mapping-based genetic algorithm | |
| Zhang et al. | Short-term load forecasting using recurrent neural networks with input attention mechanism and hidden connection mechanism | |
| CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
| Ajitha et al. | Design and development of Residential Sector Load Prediction model during COVID-19 Pandemic using LSTM based RNN | |
| CN116108984A (en) | Urban flow prediction method based on flow-POI causal relationship reasoning | |
| Zhuang et al. | Simulation of urban land expansion in China at 30 m resolution through 2050 under shared socioeconomic pathways | |
| Cao et al. | PSO-Stacking improved ensemble model for campus building energy consumption forecasting based on priority feature selection | |
| Royuela et al. | HouSI: Heuristic for delimitation of housing submarkets and price homogeneous areas | |
| Gong | Deep belief network-based multifeature fusion music classification algorithm and simulation | |
| CN113780679B (en) | Load prediction method and device based on ubiquitous power Internet of things | |
| CN114358449A (en) | Electric vehicle charging load space-time distribution prediction method based on graph neural network | |
| Nightingale et al. | Effect of clustering in federated learning on non-iid electricity consumption prediction | |
| Abdelaziz et al. | Convolutional Neural Network With Genetic Algorithm for Predicting Energy Consumption in Public Buildings | |
| CN116523001A (en) | Method, device and computer equipment for constructing weak line identification model of power grid | |
| Yong et al. | Short-term building load forecasting based on similar day selection and LSTM network | |
| CN113537607B (en) | Power failure prediction method | |
| Prashanthi et al. | A comparative study of the performance of machine learning based load forecasting methods | |
| Kim et al. | Identifying the impact of decision variables for nonlinear classification tasks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |