CN116885705A - Regional load prediction method and device - Google Patents

Regional load prediction method and device Download PDF

Info

Publication number
CN116885705A
CN116885705A CN202310840563.0A CN202310840563A CN116885705A CN 116885705 A CN116885705 A CN 116885705A CN 202310840563 A CN202310840563 A CN 202310840563A CN 116885705 A CN116885705 A CN 116885705A
Authority
CN
China
Prior art keywords
load
prediction
sequence
user
environment
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.)
Pending
Application number
CN202310840563.0A
Other languages
Chinese (zh)
Inventor
曾顺奇
冀浩然
张镇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangzhou Power Supply Bureau of Guangdong Power Grid Co Ltd
Original Assignee
Guangzhou Power Supply Bureau of Guangdong Power Grid Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guangzhou Power Supply Bureau of Guangdong Power Grid Co Ltd filed Critical Guangzhou Power Supply Bureau of Guangdong Power Grid Co Ltd
Priority to CN202310840563.0A priority Critical patent/CN116885705A/en
Publication of CN116885705A publication Critical patent/CN116885705A/en
Pending legal-status Critical Current

Links

Classifications

    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
    • 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
    • G06F18/211—Selection of the most significant subset of features
    • G06F18/2113—Selection of the most significant subset of features by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation
    • 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
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • 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
    • G06F18/2337—Non-hierarchical techniques using fuzzy logic, i.e. fuzzy 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
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • 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
    • G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • 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
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Business, Economics & Management (AREA)
  • Biophysics (AREA)
  • Economics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Strategic Management (AREA)
  • Marketing (AREA)
  • Physiology (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Operations Research (AREA)
  • Genetics & Genomics (AREA)
  • Game Theory and Decision Science (AREA)
  • Fuzzy Systems (AREA)
  • Automation & Control Theory (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Quality & Reliability (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Water Supply & Treatment (AREA)
  • Public Health (AREA)

Abstract

The application discloses a regional load prediction method and a regional load prediction device, wherein the method can determine the environmental characteristic type related to load and all user types of a region, each user type corresponds to a plurality of users in the region, and each user corresponding to each user type forms all users of the region; collecting an environment value corresponding to the environment characteristic type; acquiring a load sequence corresponding to each user type and a long-term and short-term memory neural network BiLSTM prediction network; the environment value corresponding to the environment characteristic type is used as a labeling label of each load sequence and labeled on the load sequence; inputting the marked load sequence into a corresponding BiLSTM prediction network to obtain a load prediction sequence output by the BiLSTM prediction network; and superposing the load prediction sequences corresponding to the user types according to the number of users corresponding to each user type to form a prediction result of the region. Therefore, the accuracy of regional load prediction can be further improved.

Description

Regional load prediction method and device
Technical Field
The application relates to the technical field of load prediction, in particular to a regional load prediction method and a regional load prediction device.
Background
Load prediction is an important component of power management. The accurate prediction of the load is key to maintaining the safe and stable operation of the power grid system. The transformer area refers to a power supply range or area of a transformer, and the reliability of the transformer area load prediction result obtained by the prior art is low due to the fact that the overall load change regularity of the transformer area level is weak. Thus, how to accurately predict the load of a region becomes an important point of attention.
Disclosure of Invention
In view of the above, the present application provides a method and apparatus for predicting regional load, which are used to solve the disadvantage that it is difficult to accurately predict the load of a station in the prior art.
In order to achieve the above object, the following solutions have been proposed:
a regional load prediction method comprising:
determining the environment characteristic type related to the load and all user types of an area, wherein each user type corresponds to a plurality of users in the area, and each user corresponding to each user type forms all users of the area;
collecting an environment value corresponding to the environment characteristic type;
acquiring a load sequence corresponding to each user type and a long-term and short-term memory neural network BiLSTM prediction network;
The environment value corresponding to the environment characteristic type is used as a labeling label of each load sequence and labeled on the load sequence;
inputting the marked load sequence into a corresponding BiLSTM prediction network to obtain a load prediction sequence output by the BiLSTM prediction network;
and superposing load prediction sequences corresponding to the user types according to the number of users corresponding to the user types to form a prediction result of the area.
Optionally, the determining the type of the environmental characteristic related to the load includes:
determining a plurality of extrinsic environment types;
acquiring an environment value sequence of each external environment type, and combining the environment value sequences to form a variable set;
screening and dimension reduction are carried out on the variable set by adopting a principal component analysis method, and a characteristic value corresponding to each external environment type is generated;
and extracting principal components in the plurality of extrinsic environment types as environment characteristic types related to loads according to the characteristic values.
Optionally, the extracting the principal component in the multiple extrinsic environment types as the environment feature type related to the load according to each feature value includes:
Calculating variance contribution values corresponding to the environment values in sequence;
sequencing all variance contribution values from large to small to obtain sequencing results;
initializing an addend;
selecting a variance contribution value with the ranking as a target contribution value from the ranking result, deleting the target contribution value from the ranking result, and recording an external environment type corresponding to the target contribution value;
judging whether the sum of the target contribution value and the addend is larger than a preset contribution threshold value, if not, taking the sum of the target contribution value and the addend as a new addend, and returning to execute the step of selecting the variance contribution value with the first rank from the ranking result as the target contribution value until the sum of the target contribution value and the addend exceeds the contribution threshold value;
and forming the environment characteristic type related to the load by taking all the external environment types obtained by the final record as main components.
Optionally, determining all user types of the area includes:
determining a historical load curve for each user in the area;
extracting the electricity utilization behavior characteristics corresponding to the users from the historical load curves corresponding to the users;
Clustering each electricity behavior feature by adopting a fuzzy C-means clustering genetic algorithm GA-FCM to obtain a clustering result comprising a plurality of clustering sets, wherein each clustering set comprises electricity behavior features corresponding to a plurality of users, and all the electricity behavior features of each clustering set form electricity behavior features corresponding to all users in the area;
and determining the user types corresponding to each cluster set, wherein the user types corresponding to each cluster set form all the user types of the region.
Optionally, the extracting the electricity behavior feature corresponding to the user from the historical load curve corresponding to each user includes:
selecting a plurality of fluctuation points from each historical load curve in sequence, wherein the number of the fluctuation points of each historical load curve is the same;
dividing the historical load curve according to the fluctuation points of the historical load curve to form a plurality of sections of sub-curves;
and extracting curve characteristics from each section of the sub-curve, wherein each curve characteristic corresponding to the user and each fluctuation point form the electricity utilization behavior characteristic corresponding to the user.
Optionally, obtaining a long-term and short-term memory neural network BiLSTM prediction network corresponding to each user type includes:
Acquiring an initialized BiLSTM model corresponding to the user type;
acquiring a plurality of training load sequences corresponding to the user types, wherein each training load sequence is marked with a training environment value corresponding to an environment characteristic type, and the training load sequences are spliced with a previous cycle historical load sequence and a later cycle historical load sequence;
and training the initialization BiLSTM model by utilizing the training load sequences and combining a random weight average algorithm SWA algorithm until the initialization BiLSTM model converges, and taking the finally obtained initialization BiLSTM model as a long-short-period memory neural network BiLSTM prediction network corresponding to the user type.
Optionally, the training the initializing BiLSTM model by using each training load sequence and combining a random weight average algorithm SWA algorithm includes:
setting SWA weight parameters;
initializing iteration times;
sequentially inputting each training load sequence into the initialization BiLSTM model to obtain a predicted sequence output by the initialization BiLSTM model based on a previous cycle history load sequence in the training load sequence;
adding 1 to the iteration times;
Comparing the predicted sequence with a later period historical load sequence in the training load sequence to obtain a comparison result;
according to the comparison result and the SWA weight parameter, carrying out random gradient descent on the parameters of the initialized BiLSTM model;
judging whether the iteration times are equal to a preset iteration threshold value or not;
if the number of the SWA weight parameters is equal to the number of the SWA weight parameters, a weighted average value of the adjusted parameters is obtained, and the weighted average value is used as a new SWA weight parameter, and the step of executing the initialization iteration times is returned;
and if not, returning to execute the step of inputting each training load sequence into the initialization BiLSTM model in sequence to obtain a predicted sequence output by the initialization BiLSTM model based on a previous cycle history load sequence in the training load sequences.
Optionally, the inputting the marked load sequence into a corresponding BiLSTM prediction network to obtain a load prediction sequence output by the BiLSTM prediction network includes:
inputting the marked load sequence to a corresponding BiLSTM prediction network, and carrying out load prediction by utilizing the BiLSTM prediction network based on the load sequence and an environment value corresponding to the environment characteristic type marked on the load sequence to form and output a load prediction sequence.
Optionally, the stacking the load prediction sequences corresponding to the user types according to the number of users corresponding to each user type to form a prediction result of the area includes:
referring to the load prediction sequence corresponding to each user type and the number of corresponding users, obtaining a load prediction result corresponding to the user type;
and summarizing the load prediction results corresponding to the user types to form the prediction results of the areas.
An area load prediction apparatus comprising:
the system comprises a determining module, a load-related environment feature type determining module and a load-related environment feature type determining module, wherein the determining module is used for determining the environment feature type related to the load and all user types of an area, each user type corresponds to a plurality of users in the area, and all users in the area are formed by the corresponding users;
the acquisition module is used for acquiring the environment value corresponding to the environment characteristic type;
the acquisition module is used for acquiring a load sequence corresponding to each user type and a long-short-term memory neural network BiLSTM prediction network;
the labeling module is used for labeling the environment values corresponding to the environment characteristic types as labeling labels of the load sequences;
The prediction module is used for inputting the marked load sequence into a corresponding BiLSTM prediction network to obtain a load prediction sequence output by the BiLSTM prediction network;
and the superposition module is used for superposing the load prediction sequences corresponding to the user types according to the number of users corresponding to each user type to form a prediction result of the region.
According to the technical scheme, the regional load prediction method provided by the application can determine the environment characteristic type related to the load and all user types of the region, each user type corresponds to a plurality of users in the region, and each user corresponding to each user type forms all users of the region; collecting an environment value corresponding to the environment characteristic type; acquiring a load sequence corresponding to each user type and a long-term and short-term memory neural network BiLSTM prediction network; the environment value corresponding to the environment characteristic type is used as a labeling label of each load sequence and labeled on the load sequence; inputting the marked load sequence into a corresponding BiLSTM prediction network to obtain a load prediction sequence output by the BiLSTM prediction network; therefore, the application can divide all users in the area into a plurality of types, thereby better distinguishing the load change rule of each user type, completing load prediction according to the user type in a targeted way, and being capable of comprehensively referring to the load sequence and the external environment characteristics to carry out load prediction, and further the reliability and the accuracy of the load prediction of the application; superposing load prediction sequences corresponding to the user types to form a prediction result of the area; therefore, the load prediction sequences of all user types can be summarized according to the number of the user persons of all user types, so that the load prediction result of the whole area is obtained. The area also belongs to one of the areas, and therefore, the accuracy of area load prediction can be further improved.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present application, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a regional load prediction method disclosed in an embodiment of the present application;
FIG. 2 is a schematic diagram of a BiLSTM prediction network according to an embodiment of the present application;
FIG. 3 is a schematic diagram of an LSTM neuron according to an embodiment of the present application;
FIG. 4 is a block diagram of a regional load prediction apparatus according to an embodiment of the present application;
fig. 5 is a block diagram of a hardware structure of a regional load prediction apparatus according to an embodiment of the present application.
Detailed Description
The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
The embodiment of the invention provides a regional load prediction method, which can be applied to various computer terminals or intelligent terminals, wherein an execution subject of the regional load prediction method can be a processor or a server of the computer terminal or the intelligent terminal, and a flow chart of the regional load prediction method is shown in fig. 1, and specifically comprises the following steps:
and S1, determining all user types of the environment characteristic types and the areas related to the load.
Specifically, each user type corresponds to a plurality of users in the area, and each user corresponding to each user type forms all users in the area.
The area may be larger or smaller, and may correspond to a range corresponding to one area or a range corresponding to a plurality of areas.
More than two user types can exist in one area, and the load change rule of the same user type is basically consistent.
There may be one or more load-related environmental characteristic types, which may be environmental factors that affect the use of the appliance, such as month, highest air temperature, lowest air temperature, rainfall, or humidity.
And S2, collecting an environment value corresponding to the environment characteristic type.
Specifically, an environmental value corresponding to each environmental feature type may be acquired, for example, a value of the highest air temperature, a value of the lowest air temperature, and the like may be acquired.
And S3, acquiring a load sequence corresponding to each user type and a long-short-term memory neural network BiLSTM prediction network.
Specifically, different user types correspond to different BiLSTM prediction networks, and each user type corresponds to each BiLSTM prediction network one by one.
The BiLSTM prediction network is obtained through the training of the real load sequence of the user type. The BiLSTM prediction network may predict a load prediction sequence corresponding to any user of the user type based on the load sequences corresponding to that user type.
The BiLSTM prediction network may be formed by interconnecting two long and short-term memory networks LSTM, one LSTM constituting a forward propagation hidden layer of the BiLSTM prediction network and the other LSTM constituting a reverse propagation hidden layer of the BiLSTM prediction network, as shown in FIG. 2.
X in FIG. 2 t ' is the load value corresponding to the t moment in the input load sequence, y t For the load predicted value corresponding to the t time in the load predicted sequence, h' t The hidden layer state corresponding to the time t is obtained.
The relationships in the BiLSTM predictive network are:
The state of the hidden layer is propagated forward corresponding to the time t; />The state of the hidden layer is back propagation corresponding to the time t.
The LSTM network may be mainly composed of a forget gate, an input gate, an output gate, and a memory unit, as shown in fig. 3. The state of the memory unit at the moment t can be composed of a forgetting part state and a reserved part state; wherein the state of the forgetting part is input from the time t, the state of the memory unit at the time t-1 and the intermediate output h at the time t t The common decision, the reserved part state is determined by the output of the forgotten part state after the sigma function and the tanh function are respectively transformed. h is a t The output gate is obtained by transforming the state of the forgotten part and the state of the memory unit t, wherein the sigma function can be a sigmoid function. the tanh function is a hyperbolic tangent function.
The variable calculation formulas in fig. 3 can be as follows:
f t ＝σ(W f ×[h t＝1 ,x t ]+b f )
i t ＝σ(W i ×[h t-1 ,x t ]+b i )
g t ＝tanh(W g ×[h t-1 ,x t ]+b g )
o t ＝σ(W o ×[h t-1 ,x t ]+b o )
c t ＝c t-1 ×f t +g t ×i t
h t ＝o t ×tanh(c t )
x t the input at time t can be; w (W) f A weight matrix that can be a forget gate; x is x t The input corresponding to the time t can be obtained; b f Can be a genetic codeForget the bias term of the door; w (W) i W and W g A weight matrix which can be an input gate; b i B g May be a bias term for the input gate; b o A bias term that may be an output gate; c t-1 The state of the memory cell t-1 at the moment; h is a t Can be the middle output of the time t; w (W) o May be a weight matrix of output gates.
The load sequence corresponding to each user type may be obtained in various manners, for example, the load sequence of the last period of any user in each user type may be obtained as the load sequence corresponding to the user type, or the load sequence of the last period of any plurality of users in each user type may be obtained, and an average sequence among the plurality of load sequences may be used as the load sequence corresponding to the user type.
The load sequence may be preprocessed to compensate for outliers and missing values in the load sequence.
In the preprocessing process, the load value of each moment in each period can be compared with the load average value of the same moment in the previous period, the previous two periods, the next period and the same period of the previous period, and when the difference value between the load value and the load average value is smaller than a preset difference value range, the load value is determined to be free of errors; and when the difference value range is exceeded, updating the load value to the load average value.
The load average can be calculated using the following formula:
wherein L is m,n May be a load average; l (L) m,n-1 The load value can be the same time as the load value of the previous time period; l (L) m,n-2 The load value can be the same time in the first two time periods; l (L) m,n+1 The load value can be the same time in the latter period; l (L) m,n+2 The load value can be the load value at the same time in the latter two time periods; l (L) m-1,n The load value can be the same time and the same period of the previous period.
The load sequence may include load values at various times in the previous cycle, and the load prediction sequence may include load prediction values at various times in the current cycle.
And S4, taking the environment value corresponding to the environment characteristic type as a labeling label of each load sequence, and labeling the environment value on the load sequence.
Specifically, each load sequence may be labeled with a respective environmental characteristic type and its corresponding environmental value.
And S5, inputting the marked load sequence into a corresponding BiLSTM prediction network to obtain a load prediction sequence output by the BiLSTM prediction network.
Specifically, the marked load sequence may be input into a BiLSTM prediction network corresponding to the load sequence, and the load prediction network corresponding to the load sequence is utilized to predict the load of the current period based on the load change condition of the load sequence and the environment value corresponding to each environment characteristic type, so as to obtain the load prediction sequence corresponding to the user type.
And S6, superposing load prediction sequences corresponding to the user types according to the number of users corresponding to the user types to form a prediction result of the area.
Specifically, each load prediction sequence may be weighted according to the number of users corresponding to each user type, to form a prediction result of the region.
According to the technical scheme, the regional load prediction method provided by the embodiment of the application can determine the environment characteristic type related to the load and all user types of the region, wherein each user type corresponds to a plurality of users in the region, and each user corresponding to each user type forms all users of the region; collecting an environment value corresponding to the environment characteristic type; acquiring a load sequence corresponding to each user type and a long-term and short-term memory neural network BiLSTM prediction network; the environment value corresponding to the environment characteristic type is used as a labeling label of each load sequence and labeled on the load sequence; inputting the marked load sequence into a corresponding BiLSTM prediction network to obtain a load prediction sequence output by the BiLSTM prediction network; therefore, the application can divide all users in the area into a plurality of types, thereby better distinguishing the load change rule of each user type, completing load prediction according to the user type in a targeted way, and being capable of comprehensively referring to the load sequence and the external environment characteristics to carry out load prediction, and further the reliability and the accuracy of the load prediction of the application; superposing load prediction sequences corresponding to the user types to form a prediction result of the area; therefore, the load prediction sequences of all user types can be summarized according to the number of the user persons of all user types, so that the load prediction result of the whole area is obtained. The area also belongs to one of the areas, and therefore, the accuracy of area load prediction can be further improved.
In some embodiments of the present application, the process of determining the type of environmental characteristic related to the load in step S1 is described in detail as follows:
s10, determining a plurality of external environment types.
Specifically, a plurality of extrinsic environment types that may be associated with the load, such as a highest air temperature, a lowest air temperature, an average air temperature on the same day, an average air temperature on the first 1 day, humidity, precipitation, weather conditions, date type, month type, and the like, may be determined.
The date type may be day of the week.
Weather conditions may be of the sunny, cloudy, and gust weather type.
S11, acquiring an environment value sequence of each external environment type, and combining the environment value sequences to form a variable set.
Specifically, the environmental values for a plurality of periods in the previous cycle for each extrinsic environmental type may be obtained, forming a sequence of environmental values.
And forming a variable set by taking each environment value sequence as one row in the set.
And S12, screening and dimension reduction are carried out on the variable set by adopting a principal component analysis method, and feature values corresponding to each external environment type are generated.
Specifically, a covariance matrix corresponding to the variable set may be calculated, and a eigenvalue and an orthogonalization unit eigenvector corresponding to each external environment type may be calculated according to the covariance matrix corresponding to the variable set.
The composition of the covariance matrix may be as follows:
S＝∑(s ij ) n×n
p may be the number of periods contained in the previous period; n may be the total number of extrinsic environment types; the value range of i and j is [1, n ]]；The average value among the environment values of the ith external environment type can be obtained; />The average value among the environment values of the j-th external environment type can be the average value; x is x ki The environment value corresponding to the kth row and the ith column in the variable set can be obtained; x is x kj The context value corresponding to the kth row and jth column in the variable set may be used.
S13, extracting principal components in the multiple external environment types as environment characteristic types related to loads according to the characteristic values.
Specifically, the feature values may be ranked, and the external environment types corresponding to the first N feature values are selected as environment feature types related to the load.
As can be seen from the above technical solutions, the present embodiment provides an optional manner of selecting an environmental feature type related to a load, by which a main component in an external environmental type can be extracted as an environmental feature type, so as to reduce difficulty in data processing and ensure accuracy of load prediction.
In some embodiments of the present application, a process of extracting principal components in the plurality of extrinsic environment types as load-related environment feature types according to each feature value in step S13 is described in detail, and the steps are as follows:
S130, calculating variance contribution values corresponding to the characteristic values in sequence.
Specifically, the variance contribution value corresponding to each environmental value may be calculated by means of variance contribution value calculation.
The variance contribution value calculation method may be as follows:
λ i may be the i-th eigenvalue; η (eta) i The variance contribution value corresponding to the ith eigenvalue can be used; lambda (lambda) k May be the kth eigenvalue.
S131, sorting all variance contribution values from large to small to obtain sorting results.
Specifically, the values are larger than the first, and the values are smaller than the second, and the variance contribution values are ranked to obtain a ranking result.
S132, initializing an addend.
Specifically, the addend may be initialized to obtain an addend having a value of 0.
S133, selecting a variance contribution value with the first ranking from the ranking results as a target contribution value, deleting the target contribution value from the ranking results, and recording the external environment type corresponding to the target contribution value.
Specifically, the variance contribution value with the largest value can be selected from the sorting result as the target contribution value, the target contribution value is deleted from the sorting result, and meanwhile, the external environment type corresponding to the target contribution value is saved and recorded.
S134, judging whether the sum of the target contribution value and the addend is larger than a preset contribution threshold value, if not, taking the sum of the target contribution value and the addend as a new addend, and returning to the step S133 until the sum of the target contribution value and the addend exceeds the contribution threshold value.
Specifically, the sum of the target contribution value and the addend may be calculated to obtain an addition result, and when the addition result is not greater than the contribution threshold, the addition result may be regarded as a new addend, and step S133 is performed back until the sum of the target contribution value and the addend exceeds the contribution threshold.
The contribution threshold may be set according to actual requirements, for example, the contribution threshold may be set to 85%.
S135, forming environment characteristic types related to the load by taking all the external environment types obtained by final recording as main components.
Specifically, each extrinsic environment type stored in the record may be regarded as an environment feature type.
From the above technical solution, it can be seen that this embodiment provides an optional manner of selecting an environmental feature type from the external environmental types, and by using the foregoing manner, an external environmental type that has a greater contribution to the variable set may be better selected as the environmental feature type.
In some embodiments of the present application, the process of determining all user types of the area in step S1 is described in detail as follows:
s14, determining a historical load curve of each user in the area.
Specifically, a historical load curve for all periods within one cycle for each user in the area may be extracted.
S15, extracting the power utilization behavior characteristics corresponding to the users from the historical load curves corresponding to the users.
Specifically, point-of-use behavior features that can reflect the user's primary power usage features may be extracted from each historical load curve.
S16, clustering the electricity utilization behavior features by adopting a fuzzy C-means clustering genetic algorithm GA-FCM to obtain a clustering result comprising a plurality of clustering sets, wherein each clustering set comprises electricity utilization behavior features corresponding to a plurality of users, and all the electricity utilization behavior features of each clustering set form electricity utilization behavior features corresponding to all users in the area.
Specifically, a GA-FCM algorithm can be adopted to complete clustering of the power utilization behavior characteristics, and clustering results are obtained.
The clustering result can comprise a plurality of clustering sets, and each clustering set comprises a plurality of similar electricity utilization behavior characteristics.
Parameters such as the number of clustering centers, the cross probability, the iteration threshold and the like can be set, binary coding is carried out on the clustering centers, then the fitness of each power consumption behavior characteristic is calculated, and the parameters such as the power consumption behavior characteristic, the number of the clustering centers, the cross probability, the iteration threshold and the like in each clustering set are updated through genetic operations such as selection, cross or mutation and the like according to the fitness value, so that clustering is completed.
The clustering formula is as follows:
wherein J can be a clustering function, N can be the number of electricity behavior characteristics contained in a clustering set, and c can be the number of clustering centers; m may be fuzzy weighted fingerA number; x is x i The ith electricity behavior feature; c j Is the j-th cluster center; u (u) ij For sample x i Belonging to the membership degree of the cluster set j; f is the fitness value; x is x j The electricity utilization behavior characteristic is j-th; c i Is the ith cluster center; c k Is the kth cluster center.
S17, determining the user types corresponding to each cluster set, wherein the user types corresponding to each cluster set form all the user types of the area.
Specifically, for each electricity behavior feature in each cluster set, a corresponding user type can be determined, each user type is used as the user type corresponding to the cluster set, and the type division of the users in the area is completed.
After the types of the users are divided, the cluster sets and the corresponding user types can be stored, and the cluster sets and the corresponding user types are directly called during load prediction.
As can be seen from the above technical solution, the present embodiment provides an optional manner of determining all user types in an area, and by using the foregoing manner, clustering of all users in an area can be completed, thereby completing determination of user types.
In some embodiments of the present application, the process of extracting the power consumption behavior feature corresponding to the user from the historical load curve corresponding to each user in step S15 is described in detail, and the steps are as follows:
s150, sequentially selecting a plurality of fluctuation points from each historical load curve, wherein the number of the fluctuation points of each historical load curve is the same.
Specifically, the number of fluctuation points may be determined according to actual conditions, and in general, the number of fluctuation points may be 2 in consideration of a large load fluctuation of a user when going home and going to sleep, or a large load fluctuation of a user in summer and winter in one year.
The first n coordinate points with the largest fluctuation can be selected from each historical load curve as fluctuation points.
S151, dividing the historical load curve according to the fluctuation points of the historical load curve to form a plurality of sections of sub-curves.
Specifically, the historical load curve may be divided by using the fluctuation point as a division point, so as to obtain a plurality of sub-curves.
And S152, extracting curve characteristics from each section of the sub-curve, wherein each curve characteristic corresponding to the user and each fluctuation point form an electricity utilization behavior characteristic corresponding to the user.
Specifically, the load mean value and the load variance corresponding to each section of the sub-curve can be extracted from each section of the sub-curve as curve characteristics, and the curve characteristics and the fluctuation points corresponding to the same user form the power utilization behavior characteristics corresponding to the user.
According to the technical scheme, the embodiment provides an optional mode for determining the electricity utilization behavior characteristics of each user, and redundant information in the load curve can be further removed through the mode, so that simplification of the electricity utilization behavior characteristics is completed.
In some embodiments of the present application, a process of acquiring the long-short-period memory neural network BiLSTM prediction network corresponding to each user type in step S3 is described in detail, and the steps are as follows:
S30, acquiring an initialized BiLSTM model corresponding to the user type.
Specifically, an initialized untrained BiLSTM model may be obtained as an initialized BiLSTM model corresponding to the user type.
S31, acquiring a plurality of training load sequences corresponding to the user types, wherein each training load sequence is marked with a training environment value corresponding to the environment characteristic type, and the training load sequences are spliced with a previous cycle historical load sequence and a later cycle historical load sequence.
Specifically, a training load sequence marked with training environment values corresponding to the environment feature types may be obtained, and the training load sequence may be formed by splicing two continuous-period historical load sequences.
S32, training the initialization BiLSTM model by utilizing the training load sequences and combining a random weight average algorithm SWA algorithm until the initialization BiLSTM model converges, and taking the finally obtained initialization BiLSTM model as a long-short-period memory neural network BiLSTM prediction network corresponding to the user type.
Specifically, a SWA algorithm may be applied, and each training load sequence is used to train the initialized BiLSTM model until the initialized BiLSTM model converges, where the initialized BiLSTM model obtained by training is a BiLSTM prediction network corresponding to the user type.
From the above technical solution, it can be seen that this embodiment provides an optional way of training the BiLSTM prediction network corresponding to each user type, by which the BiLSTM prediction network can be obtained by training in combination with the SWA algorithm and the supervised mode, so that training of the BiLSTM prediction network can be better completed.
In some embodiments of the present application, the training process of the initializing BiLSTM model by using each training load sequence in step S32 and combining with the SWA algorithm, which is a random weight average algorithm, is described in detail as follows:
s320, setting SWA weight parameters.
Specifically, SWA weight parameters may be set according to actual requirements.
S321, initializing iteration times.
Specifically, the number of iterations may be set to 0.
S322, sequentially inputting each training load sequence into the initialization BiLSTM model to obtain a predicted sequence output by the initialization BiLSTM model based on a previous cycle history load sequence in the training load sequence.
Specifically, the training load sequence may be input into the initialization BiLSTM model, to obtain a predicted sequence output by the initialization BiLSTM model based on a previous cycle of historical load sequence in the input training load sequence.
S323, adding 1 to the iteration number.
Specifically, the number of iterations +1 may be counted.
S324, comparing the predicted sequence with a later period historical load sequence in the training load sequence to obtain a comparison result.
Specifically, the loss value may be calculated based on the predicted sequence and a later period history load sequence of the inputted training load sequences.
S325, according to the comparison result and the SWA weight parameter, carrying out random gradient descent on the parameters of the initialized BiLSTM model.
Specifically, according to the loss value and the SWA weight parameter, the parameter initializing the BiLSTM model is subjected to random gradient descent, and parameter adjustment is completed.
S326, judging whether the iteration times are equal to a preset iteration threshold, if so, executing the step S327, and if not, executing the step S322.
Specifically, whether the iteration number is equal to the preset iteration threshold may be compared, and when the iteration number is equal to the preset iteration threshold, the step S327 is executed again, and if the iteration number is not equal to the preset iteration threshold, the step S322 is executed again.
S327, a weighted average value of the adjusted parameters is obtained, and the weighted average value is used as a new SWA weight parameter, and the step S321 is executed.
Specifically, when the number of iterations is equal to the preset iteration threshold, the SWA weight parameter may be updated, and the step S321 may be executed back.
From the above technical solution, it can be seen that this embodiment provides an optional way of training the initialization BiLSTM model by using each training load sequence and combining with the SWA algorithm of the random weight average algorithm, and by using the above way, the training convergence speed of the initialization BiLSTM model can be further improved, and the training efficiency is improved.
In some embodiments of the present application, a process of inputting the marked load sequence into the corresponding BiLSTM prediction network to obtain the load prediction sequence output by the BiLSTM prediction network in step S5 is described in detail, and the steps are as follows:
s50, inputting the marked load sequence into a corresponding BiLSTM prediction network, and performing load prediction by utilizing the BiLSTM prediction network based on the load sequence and an environment value corresponding to an environment characteristic type marked on the load sequence to form and output a load prediction sequence.
Specifically, the BiLSTM prediction network can complete load prediction based on the load sequence, each environment characteristic type marked in the load sequence, the environment value thereof and the load change rule of the corresponding user type, and output a load prediction sequence.
According to the technical scheme, the embodiment provides an optional mode for acquiring the load prediction sequence, so that the load prediction can be better completed and the reliability of the load prediction can be improved.
In some embodiments of the present application, in step S6, according to the number of users corresponding to each user type, the load prediction sequences corresponding to each user type are overlapped, and the process of forming the prediction result of the area is described in detail, where the steps are as follows:
s60, referring to the load prediction sequences corresponding to the user types and the number of corresponding user numbers, and obtaining a load prediction result corresponding to the user types.
Specifically, the number of users corresponding to the user type may be multiplied by the load prediction sequence to obtain a load prediction result obtained by the user type.
And S61, summarizing load prediction results corresponding to the user types to form prediction results of the areas.
Specifically, the load prediction results corresponding to the user types may be superimposed to obtain a prediction result corresponding to the whole area.
From the above technical solution, it can be seen that this embodiment provides an alternative way to summarize the load prediction sequence, and by this way, load prediction of the whole area can be better completed.
Next, the area load predicting apparatus provided by the present application will be described in detail with reference to fig. 4, and the area load predicting apparatus provided hereinafter may be compared with the area load predicting method provided above.
As can be seen from fig. 4, the area load prediction apparatus may include:
the determining module 1 is used for determining the environment characteristic type related to the load and all user types of the area, wherein each user type corresponds to a plurality of users in the area, and each user corresponding to each user type forms all users of the area;
the acquisition module 2 is used for acquiring the environment value corresponding to the environment characteristic type;
the acquisition module 3 is used for acquiring a load sequence corresponding to each user type and a long-short-term memory neural network BiLSTM prediction network;
the labeling module 4 is used for labeling the environment values corresponding to the environment characteristic types as labeling labels of the load sequences;
the prediction module 5 is used for inputting the marked load sequence into a corresponding BiLSTM prediction network to obtain a load prediction sequence output by the BiLSTM prediction network;
and the superposition module 6 is used for superposing the load prediction sequences corresponding to the user types according to the number of users corresponding to each user type to form a prediction result of the region.
Further, the determining module may include:
an extrinsic environment type determining unit configured to determine a plurality of extrinsic environment types;
the environment value acquisition sequence unit is used for acquiring an environment value sequence of each external environment type and combining the environment value sequences to form a variable set;
the variable set dimension reduction unit is used for screening and dimension reduction of the variable set by adopting a principal component analysis method to generate a characteristic value corresponding to each external environment type;
and the environment characteristic type extraction unit is used for extracting principal components in the plurality of external environment types as environment characteristic types related to loads according to the characteristic values.
Further, the environmental feature type extraction unit may include:
the first environment feature type extraction subunit is used for sequentially calculating variance contribution values corresponding to each feature value;
the second environment characteristic type extraction subunit is used for sequencing all variance contribution values from large to small to obtain a sequencing result;
a third environmental feature type extraction subunit for initializing an addend;
a fourth environmental feature type extraction subunit, configured to select, from the ranking result, a variance contribution value that is ranked as a target contribution value, delete the target contribution value from the ranking result, and record an external environmental type corresponding to the target contribution value;
A fifth environmental feature type extracting subunit, configured to determine whether a sum of the target contribution value and the addend is greater than a preset contribution threshold, and if not, return to executing a step of selecting, from the ranking result, a variance contribution value that is ranked as a target contribution value, with the sum of the target contribution value and the addend as a new addend, until the sum of the target contribution value and the addend exceeds the contribution threshold;
and a sixth environmental characteristic type extraction subunit, configured to form an environmental characteristic type related to the load by using all the external environmental types obtained by the final recording as main components.
Further, the determining module may further include:
a history load curve determining unit configured to determine a history load curve of each user in the area;
the electricity consumption behavior extraction feature unit is used for extracting electricity consumption behavior features corresponding to the users from the historical load curves corresponding to the users;
the clustering result acquisition unit is used for clustering the electricity utilization behavior characteristics by adopting a fuzzy C-means clustering genetic algorithm GA-FCM to obtain clustering results comprising a plurality of clustering sets, wherein each clustering set comprises electricity utilization behavior characteristics corresponding to a plurality of users, and all the electricity utilization behavior characteristics of each clustering set form electricity utilization behavior characteristics corresponding to all users in the area;
And the user type determining unit is used for determining the user type corresponding to each cluster set, and the user types corresponding to each cluster set form all the user types of the area.
Further, the electricity behavior extraction feature unit may include:
the first electric behavior extraction characteristic subunit is used for sequentially selecting a plurality of fluctuation points from each historical load curve, wherein the number of the fluctuation points of each historical load curve is the same;
the second power consumption behavior extraction characteristic subunit is used for dividing the historical load curve according to the fluctuation points of the historical load curve to form a plurality of sections of sub-curves;
and the third electricity behavior extraction feature subunit is used for extracting curve features from each section of the sub-curve, and each curve feature corresponding to the user and each fluctuation point form electricity behavior features corresponding to the user.
Further, the acquisition module may include:
an initialized BiLSTM model obtaining unit, configured to obtain an initialized BiLSTM model corresponding to the user type;
the training load sequence acquisition unit is used for acquiring a plurality of training load sequences corresponding to the user types, each training load sequence is marked with a training environment value corresponding to the environment characteristic type, and the training load sequences are spliced with a previous cycle historical load sequence and a later cycle historical load sequence;
The BiLSTM prediction network obtaining unit is used for training the initialization BiLSTM model by utilizing each training load sequence and combining a random weight average algorithm SWA algorithm until the initialization BiLSTM model converges, and taking the finally obtained initialization BiLSTM model as a long-short-period memory neural network BiLSTM prediction network corresponding to the user type.
Further, the BiLSTM predictive network acquisition unit may include:
the first BiLSTM prediction network acquisition subunit is used for setting SWA weight parameters;
the second BiLSTM prediction network acquires a subunit, which is used for initializing iteration times;
the third BiLSTM prediction network acquisition subunit is used for sequentially inputting each training load sequence into the initialization BiLSTM model to obtain a prediction sequence output by the initialization BiLSTM model based on a previous cycle history load sequence in the training load sequence;
a fourth BiLSTM prediction network acquisition subunit, configured to add 1 to the iteration number;
a fifth BiLSTM prediction network acquisition subunit, configured to compare the prediction sequence with a later period historical load sequence in the training load sequence to obtain a comparison result;
A sixth BiLSTM prediction network obtaining subunit, configured to perform random gradient descent on the parameters of the initialized BiLSTM model according to the comparison result and the SWA weight parameter;
and the seventh BiLSTM prediction network acquisition subunit is used for judging whether the iteration times are equal to a preset iteration threshold value, if so, calling the eighth BiLSTM prediction network acquisition subunit, and if not, calling the third BiLSTM prediction network acquisition subunit and the subsequent units thereof.
And the eighth BiLSTM prediction network acquisition subunit is used for calculating a weighted average value of the adjusted parameters, taking the weighted average value as a new SWA weight parameter, and calling the second BiLSTM prediction network acquisition subunit and a subsequent unit thereof.
Further, the prediction module may include:
and the load prediction sequence output unit is used for inputting the marked load sequence to a corresponding BiLSTM prediction network, and carrying out load prediction by utilizing the BiLSTM prediction network based on the load sequence and an environment value corresponding to the environment characteristic type marked on the load sequence to form and output a load prediction sequence.
Further, the superimposing module may include:
the user number reference unit is used for referring to the load prediction sequence corresponding to each user type and the number of corresponding users to obtain a load prediction result corresponding to the user type;
And the load prediction result summarizing unit is used for summarizing the load prediction results corresponding to the user types to form the prediction results of the areas.
The regional load prediction device provided by the embodiment of the application can be applied to regional load prediction equipment, such as PC terminals, cloud platforms, servers, server clusters and the like. Alternatively, fig. 5 shows a block diagram of a hardware structure of the area load prediction apparatus, and referring to fig. 5, the hardware structure of the area load prediction apparatus may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4;
in the embodiment of the application, the number of the processor 1, the communication interface 2, the memory 3 and the communication bus 4 is at least one, and the processor 1, the communication interface 2 and the memory 3 complete the communication with each other through the communication bus 4;
processor 1 may be a central processing unit CPU, or a specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement embodiments of the present application, etc.;
the memory 3 may comprise a high-speed RAM memory, and may further comprise a non-volatile memory (non-volatile memory) or the like, such as at least one magnetic disk memory;
Wherein the memory stores a program, the processor is operable to invoke the program stored in the memory, the program operable to:
determining the environment characteristic type related to the load and all user types of an area, wherein each user type corresponds to a plurality of users in the area, and each user corresponding to each user type forms all users of the area;
collecting an environment value corresponding to the environment characteristic type;
acquiring a load sequence corresponding to each user type and a long-term and short-term memory neural network BiLSTM prediction network;
the environment value corresponding to the environment characteristic type is used as a labeling label of each load sequence and labeled on the load sequence;
inputting the marked load sequence into a corresponding BiLSTM prediction network to obtain a load prediction sequence output by the BiLSTM prediction network;
and superposing load prediction sequences corresponding to the user types according to the number of users corresponding to the user types to form a prediction result of the area.
Alternatively, the refinement function and the extension function of the program may be described with reference to the above.
The embodiment of the present application also provides a readable storage medium storing a program adapted to be executed by a processor, the program being configured to:
Determining the environment characteristic type related to the load and all user types of an area, wherein each user type corresponds to a plurality of users in the area, and each user corresponding to each user type forms all users of the area;
collecting an environment value corresponding to the environment characteristic type;
acquiring a load sequence corresponding to each user type and a long-term and short-term memory neural network BiLSTM prediction network;
the environment value corresponding to the environment characteristic type is used as a labeling label of each load sequence and labeled on the load sequence;
inputting the marked load sequence into a corresponding BiLSTM prediction network to obtain a load prediction sequence output by the BiLSTM prediction network;
and superposing load prediction sequences corresponding to the user types according to the number of users corresponding to the user types to form a prediction result of the area.
Alternatively, the refinement function and the extension function of the program may be described with reference to the above.
Finally, it is further noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Various embodiments of the present application may be combined with each other. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. A regional load prediction method, comprising:
determining the environment characteristic type related to the load and all user types of an area, wherein each user type corresponds to a plurality of users in the area, and each user corresponding to each user type forms all users of the area;
collecting an environment value corresponding to the environment characteristic type;
Acquiring a load sequence corresponding to each user type and a long-term and short-term memory neural network BiLSTM prediction network;
the environment value corresponding to the environment characteristic type is used as a labeling label of each load sequence and labeled on the load sequence;
inputting the marked load sequence into a corresponding BiLSTM prediction network to obtain a load prediction sequence output by the BiLSTM prediction network;
and superposing load prediction sequences corresponding to the user types according to the number of users corresponding to the user types to form a prediction result of the area.
2. The regional load prediction method of claim 1, wherein determining the type of environmental characteristic associated with the load comprises:
determining a plurality of extrinsic environment types;
acquiring an environment value sequence of each external environment type, and combining the environment value sequences to form a variable set;
screening and dimension reduction are carried out on the variable set by adopting a principal component analysis method, and a characteristic value corresponding to each external environment type is generated;
and extracting principal components in the plurality of extrinsic environment types as environment characteristic types related to loads according to the characteristic values.
3. The regional load prediction method according to claim 2, wherein the extracting the principal component in the plurality of extrinsic environment types as the load-related environment feature type based on each of the feature values includes:
calculating variance contribution values corresponding to the characteristic values in sequence;
sequencing all variance contribution values from large to small to obtain sequencing results;
initializing an addend;
selecting a variance contribution value with the ranking as a target contribution value from the ranking result, deleting the target contribution value from the ranking result, and recording an external environment type corresponding to the target contribution value;
judging whether the sum of the target contribution value and the addend is larger than a preset contribution threshold value, if not, taking the sum of the target contribution value and the addend as a new addend, and returning to execute the step of selecting the variance contribution value with the first rank from the ranking result as the target contribution value until the sum of the target contribution value and the addend exceeds the contribution threshold value;
and forming the environment characteristic type related to the load by taking all the external environment types obtained by the final record as main components.
4. The area load prediction method according to claim 1, wherein determining all user types of an area comprises:
determining a historical load curve for each user in the area;
extracting the electricity utilization behavior characteristics corresponding to the users from the historical load curves corresponding to the users;
clustering each electricity behavior feature by adopting a fuzzy C-means clustering genetic algorithm GA-FCM to obtain a clustering result comprising a plurality of clustering sets, wherein each clustering set comprises electricity behavior features corresponding to a plurality of users, and all the electricity behavior features of each clustering set form electricity behavior features corresponding to all users in the area;
and determining the user types corresponding to each cluster set, wherein the user types corresponding to each cluster set form all the user types of the region.
5. The regional load prediction method according to claim 4, wherein the extracting the electricity behavior feature corresponding to the user from the historical load curve corresponding to each user includes:
selecting a plurality of fluctuation points from each historical load curve in sequence, wherein the number of the fluctuation points of each historical load curve is the same;
Dividing the historical load curve according to the fluctuation points of the historical load curve to form a plurality of sections of sub-curves;
and extracting curve characteristics from each section of the sub-curve, wherein each curve characteristic corresponding to the user and each fluctuation point form the electricity utilization behavior characteristic corresponding to the user.
6. The regional load prediction method according to claim 1, wherein obtaining long-term memory neural network BiLSTM prediction networks corresponding to each user type comprises:
acquiring an initialized BiLSTM model corresponding to the user type;
acquiring a plurality of training load sequences corresponding to the user types, wherein each training load sequence is marked with a training environment value corresponding to an environment characteristic type, and the training load sequences are spliced with a previous cycle historical load sequence and a later cycle historical load sequence;
and training the initialization BiLSTM model by utilizing the training load sequences and combining a random weight average algorithm SWA algorithm until the initialization BiLSTM model converges, and taking the finally obtained initialization BiLSTM model as a long-short-period memory neural network BiLSTM prediction network corresponding to the user type.
7. The method of claim 6, wherein training the initialization BiLSTM model using the respective training load sequences in combination with a random weight averaging algorithm SWA algorithm, comprises:
setting SWA weight parameters;
initializing iteration times;
sequentially inputting each training load sequence into the initialization BiLSTM model to obtain a predicted sequence output by the initialization BiLSTM model based on a previous cycle history load sequence in the training load sequence;
adding 1 to the iteration times;
comparing the predicted sequence with a later period historical load sequence in the training load sequence to obtain a comparison result;
according to the comparison result and the SWA weight parameter, carrying out random gradient descent on the parameters of the initialized BiLSTM model;
judging whether the iteration times are equal to a preset iteration threshold value or not;
if the number of the SWA weight parameters is equal to the number of the SWA weight parameters, a weighted average value of the adjusted parameters is obtained, and the weighted average value is used as a new SWA weight parameter, and the step of executing the initialization iteration times is returned;
and if not, returning to execute the step of inputting each training load sequence into the initialization BiLSTM model in sequence to obtain a predicted sequence output by the initialization BiLSTM model based on a previous cycle history load sequence in the training load sequences.
8. The regional load prediction method according to claim 1, wherein the inputting the marked load sequence into the corresponding BiLSTM prediction network to obtain the load prediction sequence output by the BiLSTM prediction network includes:
inputting the marked load sequence to a corresponding BiLSTM prediction network, and carrying out load prediction by utilizing the BiLSTM prediction network based on the load sequence and an environment value corresponding to the environment characteristic type marked on the load sequence to form and output a load prediction sequence.
9. The method for predicting regional load according to claim 1, wherein the superimposing the load prediction sequences corresponding to the user types according to the number of users corresponding to each user type to form the prediction result of the region includes:
referring to the load prediction sequence corresponding to each user type and the number of corresponding users, obtaining a load prediction result corresponding to the user type;
and summarizing the load prediction results corresponding to the user types to form the prediction results of the areas.
10. A regional load prediction apparatus, comprising:
The system comprises a determining module, a load-related environment feature type determining module and a load-related environment feature type determining module, wherein the determining module is used for determining the environment feature type related to the load and all user types of an area, each user type corresponds to a plurality of users in the area, and all users in the area are formed by the corresponding users;
the acquisition module is used for acquiring the environment value corresponding to the environment characteristic type;
the acquisition module is used for acquiring a load sequence corresponding to each user type and a long-short-term memory neural network BiLSTM prediction network;
the labeling module is used for labeling the environment values corresponding to the environment characteristic types as labeling labels of the load sequences;
the prediction module is used for inputting the marked load sequence into a corresponding BiLSTM prediction network to obtain a load prediction sequence output by the BiLSTM prediction network;
and the superposition module is used for superposing the load prediction sequences corresponding to the user types according to the number of users corresponding to each user type to form a prediction result of the region.
CN202310840563.0A 2023-07-10 2023-07-10 Regional load prediction method and device Pending CN116885705A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310840563.0A CN116885705A (en) 2023-07-10 2023-07-10 Regional load prediction method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310840563.0A CN116885705A (en) 2023-07-10 2023-07-10 Regional load prediction method and device

Publications (1)

Publication Number Publication Date
CN116885705A true CN116885705A (en) 2023-10-13

Family

ID=88263820

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310840563.0A Pending CN116885705A (en) 2023-07-10 2023-07-10 Regional load prediction method and device

Country Status (1)

Country Link
CN (1) CN116885705A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117787652A (en) * 2024-01-11 2024-03-29 湖北华中电力科技开发有限责任公司 Park power demand prediction method, related device and storage medium

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117787652A (en) * 2024-01-11 2024-03-29 湖北华中电力科技开发有限责任公司 Park power demand prediction method, related device and storage medium

Similar Documents

Publication Publication Date Title
CN109902801B (en) Flood collective forecasting method based on variational reasoning Bayesian neural network
CN111144468B (en) Method and device for labeling power consumer information, electronic equipment and storage medium
CN111027772B (en) Multi-factor short-term load prediction method based on PCA-DBILSTM
CN112734135B (en) Power load prediction method, intelligent terminal and computer readable storage medium
CN114792156B (en) Photovoltaic output power prediction method and system based on curve characteristic index clustering
CN109492748B (en) Method for establishing medium-and-long-term load prediction model of power system based on convolutional neural network
CN111985719B (en) Power load prediction method based on improved long-term and short-term memory network
CN110781947A (en) Power load prediction model training and power load prediction method and device
CN115115125B (en) Photovoltaic power interval probability prediction method based on deep learning fusion model
CN113449919B (en) Power consumption prediction method and system based on feature and trend perception
CN113240153A (en) Photovoltaic power generation data prediction method and device, computing equipment and storage medium
CN110007613A (en) Warming prediction method and system for heat storage type electric heater and storage medium
CN116885705A (en) Regional load prediction method and device
CN112819256A (en) Convolution time sequence room price prediction method based on attention mechanism
CN112990587A (en) Method, system, equipment and medium for accurately predicting power consumption of transformer area
CN113868938A (en) Short-term load probability density prediction method, device and system based on quantile regression
CN118014137A (en) Method and system for predicting power conversion requirements of power conversion station based on data driving
CN115829115A (en) PCA-LSTM-MTL-based photovoltaic power station-containing area load prediction method
CN113344279B (en) Resident load prediction method based on LSTM-SAM model and pooling
CN113962874A (en) Bus load model training method, device, equipment and storage medium
CN113591391A (en) Power load control device, control method, terminal, medium and application
CN112446550B (en) Short-term building load probability density prediction method
Sun et al. Information entropy based neural network model for short-term load forecasting
CN115013858B (en) Heat supply secondary network household regulation and control method based on lag time identification
CN116011657B (en) Optimization method, device and system for power distribution network load prediction model based on miniature PMU

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