CN113610328A - Power generation load prediction method - Google Patents
Power generation load prediction method Download PDFInfo
- Publication number
- CN113610328A CN113610328A CN202111087872.2A CN202111087872A CN113610328A CN 113610328 A CN113610328 A CN 113610328A CN 202111087872 A CN202111087872 A CN 202111087872A CN 113610328 A CN113610328 A CN 113610328A
- Authority
- CN
- China
- Prior art keywords
- data
- load prediction
- value
- gate
- model
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 58
- 238000010248 power generation Methods 0.000 title claims abstract description 30
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 16
- 230000010152 pollination Effects 0.000 claims abstract description 10
- 238000005457 optimization Methods 0.000 claims abstract description 9
- 238000007637 random forest analysis Methods 0.000 claims abstract description 9
- 238000005070 sampling Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 12
- 238000003066 decision tree Methods 0.000 claims description 10
- 238000010606 normalization Methods 0.000 claims description 9
- 238000013528 artificial neural network Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000001556 precipitation Methods 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000013506 data mapping Methods 0.000 claims description 4
- 238000009826 distribution Methods 0.000 claims description 3
- 239000012535 impurity Substances 0.000 claims description 3
- 230000014759 maintenance of location Effects 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 2
- 230000009467 reduction Effects 0.000 claims description 2
- 230000000717 retained effect Effects 0.000 claims 1
- 235000013311 vegetables Nutrition 0.000 claims 1
- 230000007774 longterm Effects 0.000 abstract description 4
- 238000007781 pre-processing Methods 0.000 abstract description 3
- 230000007246 mechanism Effects 0.000 abstract description 2
- 230000008859 change Effects 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000005611 electricity Effects 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 241000288105 Grus Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007621 cluster analysis Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 241000894007 species Species 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 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
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Strategic Management (AREA)
- Computational Linguistics (AREA)
- Human Resources & Organizations (AREA)
- Biophysics (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Primary Health Care (AREA)
- Evolutionary Biology (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a power generation load prediction method, which comprises the steps of obtaining load prediction data in a preset time period from a database; normalizing the load prediction data; constructing a random forest model, and calculating the relevant weight between corresponding measuring points of the load prediction data as an initial weight; starting to construct an attention gate, inputting the initial weight into the attention gate if the attention gate is constructed for the first time, and inputting the hyperparameter obtained by the flower pollination optimization algorithm into the attention gate if the attention gate is not constructed for the first time; adding the constructed attention gate weight data into a GRU model for training; setting a threshold value, and calculating the error between the predicted value and the actual value; and obtaining the optimal hyper-parameter according to a flower pollination optimizing algorithm. According to the invention, an attention mechanism is introduced, so that not only can long-term dependence on historical time be processed in prediction, but also sampling based on importance can be carried out, and the prediction precision is improved. Sampling is carried out depending on the importance degree of historical data, excessively complex data preprocessing is not needed, and the operation efficiency is improved.
Description
Technical Field
The invention relates to the field of power generation load prediction, in particular to a power generation load prediction method.
Background
The grid-connected generator set realizes the balance relation of power supply and demand according to the actual conditions of power production and user consumption. An intelligent power generation load prediction algorithm based on data mining is gradually emphasized by researchers. The power generation load prediction is used as an important work in the current-stage smart grid construction, the accuracy of the prediction algorithm can have great influence on economic dispatching, real-time control, operation planning and development planning, and meanwhile, the prediction algorithm is one of important standards for judging whether the management level of a power enterprise reaches the modernization.
In the aspect of power generation load prediction, the classical algorithm comprises various methods such as support vector machine prediction, neural network prediction, autoregressive moving average model (ARIMA) and the like. For example, in the published invention patent "a power load prediction method", patent application publication No. CN 109934375 a, cluster analysis is performed on power load data through a DBSCAN clustering algorithm, different types of data of the power load are automatically divided, and then the data are used as training data to construct an LSTM load prediction model. And combining the real-time similar category data and the original data to be used as the input of an LSTM model for predicting a daily power load curve. The load prediction method considers the deep learning model to be used for the data information learning of the same type of mode, but does not consider the strong and weak relevance of different independent variables and target variables.
In the prior art, the future load is predicted by analyzing the similarity between the current load data and the historical load data and utilizing the load change trend of the most similar day. The load prediction method mainly applies the similarity of the mode data to predict, and if no load data mode exists in history, the prediction method generates false alarm. Only the most similar load data pattern is used as a prediction result, other pattern data satisfying the similarity are not extracted, and the prediction result lacks certain rationality.
Disclosure of Invention
The invention provides a power generation load prediction method, which aims at solving the problem of low accuracy of power plant load prediction by establishing a new power generation load prediction technology around the problems of fully mining the prediction relation between a power generation load and related variables and effectively measuring the prediction accuracy, so that the power plant can be assisted to run and manage better.
A method of generating load prediction, the method comprising:
step 1: acquiring load prediction data in a preset time period from a database;
step 2: normalizing the load prediction data;
and step 3: constructing a random forest model calculated based on the gini coefficient, and calculating the correlation weight between the corresponding measuring points of the load prediction data to serve as an initial weight;
and 4, step 4: starting to construct an attention gate, inputting the initial weight into the attention gate if the attention gate is constructed for the first time, and inputting the hyperparameter obtained by the flower pollination optimization algorithm into the attention gate if the attention gate is not constructed for the first time;
and 5: building a GRU model, and adding the attention gate weight data built in the step 4 into the GRU model for training;
step 6: setting a threshold TthreCalculating the error between the predicted value and the actual value, if the error is FerrorIf the value is larger than the threshold value, entering the step 7, otherwise, outputting an optimal GRU model;
and 7: and obtaining the optimal hyper-parameter according to a flower pollination optimizing algorithm.
The method of the invention also comprises the following steps:
step 11: acquiring weather forecast data of a future preset time period from a database as independent variable data;
step 12: normalizing the independent variable data according to a training normalization model;
step 13: loading the optimal GRU model output in the step 6 of the training stage;
loading an optimal GRU model to obtain optimal network model parameters;
step 14: inputting the preprocessed independent variable data into an optimal GRU model, and operating the model;
acquiring optimal network model parameters, loading data, and calculating the states of a reset door and an update door at the predicted time through formulas (20), (21) and (22);
rt＝σ(wrxt+urht-1) (21)
zt＝σ(wzxt+uzht-1) (22)
finally, obtaining a hidden layer state combination;
step 15: calculating load prediction value F of output independent variable dataforecast；
Calculating the hidden layer h of the current time according to the formula (23) by the input datatLater, the output, w, of the GRU network model may be obtainedyIs the weight;
Fforecast＝σ(wyht) (1)。
according to the technical scheme, the invention has the following advantages:
the power generation load prediction method provided by the invention is used for clustering based on historical load data and predicting load data of a unit in the future preset time. The load of the future preset duration can be predicted according to the requirement.
According to the invention, an attention mechanism is introduced, so that not only can long-term dependence on historical time be processed in prediction, but also sampling based on importance can be carried out, and the prediction precision is improved.
The invention samples according to the importance degree of the historical data, does not need to carry out excessively complicated data preprocessing, and improves the operation efficiency.
According to the method, the random forest node splitting impurity degree change quantity after normalization is selected in the aspect of historical data weight, and then the optimization adjustment is carried out through a flower pollination optimization algorithm, so that the change process can be influenced mutually in real time among reaction parameters in the load prediction calculation process.
The GRU model related by the invention is a variation of the RNN model, and has the advantages of less parameters, high convergence rate and quick iteration. The invention can improve the operation efficiency.
According to the periodicity of short-term power load, the power generation of the conventional power grid is greatly interfered by external factors, and meteorological data precipitation, air pressure, wind speed, air temperature and humidity comprehensively reflect the influences of water, electricity, wind power and photovoltaic, so that the current trend of networking power generation is more carefully reversed. Meanwhile, the electricity utilization condition is closely related to the air temperature, and the key point is that meteorological factors are used as important indexes for load prediction.
Drawings
In order to more clearly illustrate the technical solution of the present invention, the drawings used in the description will be briefly introduced, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained based on these drawings without creative efforts.
FIG. 1 is a flow chart of a method of generating load prediction;
FIG. 2 is a flow chart of the operation of the present invention;
FIG. 3 is a graph of predicted effect;
FIG. 4 is a schematic diagram of flower pollination.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In the power generation load prediction method provided by the present invention, the units and algorithm steps of each example described in the disclosed embodiments can be realized by electronic hardware, computer software, or a combination of both, and in order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functions in the above description. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
In the method for predicting the power generation load provided by the invention, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one logical 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, devices or units, and may also be an electric, mechanical or other form of connection.
The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments in a power generation load prediction method provided by the present invention. In the following description, numerous specific details are provided to provide a thorough understanding of embodiments of the invention. One skilled in the relevant art will recognize, however, that the invention may be practiced without one or more of the specific details, or with other methods, components, devices, steps, and so forth. In other instances, well-known methods, devices, implementations or operations have not been shown or described in detail to avoid obscuring aspects of the invention.
The invention provides a power generation load prediction method based on an FPA-GRU algorithm, which is a brand-new load prediction technical method based on historical load data and meteorological data. The method mainly comprises two processes of establishing a model and operating the model.
The FPA-GRU load prediction model training process comprises the following steps:
FIG. 1 is a flow chart of the modeling of the present invention, and the whole modeling process mainly includes the following steps:
step 1: all historical data of a time period specified by a load forecasting specified relevant parameter are obtained from a database, shutdown data in all the data are identified and eliminated according to start and stop, and the processed data are used as historical data. The load prediction specified related parameter may be load prediction data in a preset time period according to the present invention, and the load prediction data in the preset time period is history data.
The invention takes a first generator set of a certain power plant as an example, and the operation process is as follows: first, the parameters associated with the unit load are selected: and reading 2019-7-1410 from a real-time database of the power plant according to the information at measuring points of precipitation, air pressure, wind speed, air temperature, humidity, historical load of the first unit and the like in the area where the power plant is located: 00 to 2020-11-2410: historical data for the 00 time period.
For a set of data samples with 6 measurement points and m time points, the data at time j can be referred to as an n-dimensional vector, which is expressed as:
u(tm)＝[uj1,uj2,uj3,…,lj6]
wherein u isj1,uj2,uj3,…,lj6And sequentially representing precipitation, air pressure, wind speed, air temperature, humidity and historical load.
The data file is in a matrix format of m x n in total, and the specific form is as follows
Step 2: carrying out normalization processing on the historical data to eliminate the influence of data dimension;
data F for time aaCarrying out data normalization processing to obtain a normalized data matrix Fa-norm. Training data F according to the following formulaaPerforming data mapping to [0,1 ]]Interval(s)
Wherein, Fa-normIs a normalized value of time a, Fa-maxIs the maximum value of the parameter, Fa-minIs the minimum value of the parameter.
And step 3: constructing a random forest model based on gini coefficient calculation, and calculating relevant weights among measuring points of historical data to serve as initial weights;
the gini coefficient refers to randomly choosing a sub-item from a data set, and measuring the probability that it is divided into other parts by errors.
3.1 calculate each measure gini index score for each decision tree where node splitting is not pure.
Wherein m represents the mth measuring point, K represents that K categories are contained in the mth measuring point, and Pmk represents the proportion of the categories K in the node m.
The weight of the features in the node is then calculated:
wherein GIi and GIr respectively represent the Gini indexes of the two new nodes after branching.
And 3.2, accumulating and counting gini index scores of each measuring point to obtain an average score, normalizing, and taking the normalized scores as initial weights for calculating the historical data among the measuring points.
If the feature XjIn the set of nodes M present in the decision tree i, then XjThe reduction in the pureness at the ith tree was:
wherein i represents the ith decision tree, j represents the jth measuring point of the ith tree, and M represents the measuring point set of the decision tree;
assuming that the random forest has n trees, the reduced impurity level of each measurement point in the random forest is:
wherein j is the jth node, and n is the total amount of the decision tree;
and finally, normalizing all the obtained importance scores to obtain weights:
and 4, step 4: starting to construct an attention gate, if the initial weight is input into the attention gate for the first time, otherwise, inputting a flower pollination optimization algorithm to obtain a hyper-parameter input attention gate;
the GRU model helps to capture long term dependencies, but is not sufficient for different degrees of attention to sub-window features over multiple time steps, for which a layer of "attention gate" is artificially built outside the GRU model, similar to the learning method of evolutionary attention. And confirming the mining species of the time relation of the sampling mode based on the importance.
According to the attention gate, the importance of the input data is sampled, and the attention gate weight data:
w＝[w1，w2，w3，...，w6]。
and 5: building a GRU model, and adding the attention gate weight data built in the step 4 into the GRU model for training;
the GRU has the capability of modeling the long-term history information of the time series, and a GRU network is selected to construct a load prediction network. The GRU has a reset gate and an update gate.
And (4) utilizing the attention gate weight in the step (4) to perform importance-based adoption on the input data, sending the data into the GRU network, and learning a nonlinear mapping function of the calculation process of the GRU unit through the following formula:
the GRU neural network mainly uses 2 sigmoid functions and GRUs synthesized by multiplication operation to control information selection, and the memory filling and the current input data are converted to be between 0 and 1 by using the sigmoid functions, so that the memory and the discarding of the data are realized.
In the GRU neural network model, the hidden layer state at the current time is the sum of the hidden layer state at the previous time and the current hidden layer activation state, that is:
the reset gate and the update gate are the combination of the hidden layer states at a time above the input data at the current time:
rt＝σ(wrxt+urht-1) (8)
zt＝σ(wzxt+uzht-1) (9)
wherein: σ is a sigmoid function, r can be expressedtAnd ztLimited between 0 and 1, wr、ur、wz、uzIs the weight of the neural network.
Reset gate rtCan imply the value h of the layer for the last momentt-1The reset is selectively performed, i.e. the reset gate determines the degree of retention, r, in the hidden layer at the previous momenttThe closer to 1, the more data is remained in the previous-time hidden layer, and the previous-time hidden layer data is subjected to reset gate processing and then is input with the current xtThe activation state of the hidden layer at the current moment can be obtained by combining and using the Tanh activation function
nz＝wxt+u(rt⊙ht-1) (11)
In the formula: as an indication of product operation, W, u is the GRU neural network weight.
htMainly by the refresh gate ztCompletion of ztThe closer to 0 the value of (b) is, the greater the degree of retention of the current time information is.
The optimized attention weight continuously improves the performance and cannot fall into local optimization, the attention distribution of the input driving sequence in a plurality of real-time steps of each datum also shows different scales, and the local information in the sampling window is effectively utilized.
Step 6: setting a threshold TthreCalculating a predicted value FforecastWith the actual value FactualIf the error is FerrorIf the value is larger than the threshold value, entering the step 7, otherwise, outputting an optimal GRU model;
Ferror＝Fforecast-Factual (12)
and 7: obtaining the optimal hyper-parameter according to the flower pollination optimization algorithm is shown in the figure 3 and the figure 4, and returning to the step 4;
7.1 initializing parameters, the number n of flower populations and the conversion probability p;
7.2 calculating the fitness value of each solution, and solving the current optimal solution and the optimal value;
and 7.3 if the condition that the conversion probability p is greater than the rand is satisfied, updating the solution according to the following formula and performing border crossing processing.
Wherein:
the solutions of the t +1 th generation and the t th generation respectively; theta represents a step control quantity, generally 1,
point-to-point multiplication is shown, Levy is a Lai-dimensional flight search path, and the value range of beta is [1, 3 ]]Calculation of Levy, generally taken as 3/2The formula:
wherein: the value range of λ is [0, 2], generally 3/2 is taken, Γ (λ) is a standard gamma function, S is a search path, and the calculation formula of S is:
where μ, v are random numbers obeying a normal distribution
7.4 if the p > rand condition is not satisfied, performing optimal solution update according to the following formula, and performing border crossing treatment:
wherein: e is [0,1 ]]Subject to uniformly distributed random numbers,
are the same plant.
And 7.5, calculating the fitness value corresponding to the new solution obtained by 7.3 or 7.4, if the fitness of the new solution is excellent, replacing the current solution and the current fitness value with the fitness value corresponding to the new solution and the new solution respectively, and otherwise, keeping the current solution and the current fitness.
7.6 if the fitness value corresponding to the new solution is better than the global optimal value, updating the global optimal solution and the global optimal value.
7.7 judging the end condition, if yes, outputting the optimal solution, otherwise, returning to 7.3.
Fig. 2 is a flow chart showing the operation of the model of the present invention, and the whole process mainly includes the following steps:
step 1: and acquiring weather forecast data of 24 hours in the future from the database as independent variable data.
Acquiring meteorological data F of 24 hours in the future of a predicted momentb: rainfall, air pressure, wind speed, air temperature and humidity are used as independent variable data;
wherein 5 columns are respectively precipitation, air pressure, wind speed, air temperature and humidity
Step 2: normalizing the independent variable data according to a training normalization model;
for data FbCarrying out data normalization processing to obtain a normalized data matrix Fb-norm. Training data F according to the following formulabPerforming data mapping to [0,1 ]]An interval.
Wherein, Fb-normAs normalized value of time, Fb-maxAs maximum of corresponding measuring points of the normalized model, Fb-minThe minimum value of the corresponding measuring point of the normalized model is obtained.
And step 3: loading the optimal GRU model output in the step 6 of the training stage;
and loading the optimal GRU model to obtain the optimal network model parameters.
And 4, step 4: inputting the preprocessed independent variable data into an optimal GRU model, and operating the model;
obtaining optimal network model parameters, loading data, calculating and predicting the state of the reset door and the update door at the moment through formulas (20), (21) and (22)
rt＝σ(wrxt+urht-1) (21)
xt＝σ(wzxt+uzht-1) (22)
Finally, obtaining a hidden layer state combination;
and 5: calculating load prediction value F of output independent variable dataforecast；
Calculating the hidden layer h of the current time according to the formula (23) by the input datatLater, the output, w, of the GRU network model may be obtainedyIs a weight value.
Fforecast＝σ(wyht) (23)
As an embodiment of the present invention, the #1 generator set online prediction process:
and acquiring data 24 hours after 10:00 of 24 days of 11 months and 24 days of 2020, selecting time, precipitation, air pressure, wind speed, air temperature and humidity as characteristics, and performing data preprocessing and loading a model. Inputting data into model, running model, and load prediction data Fforecast＝[u1，u2，…un]。
Compared with the traditional method, the method has higher operation efficiency, and the maximum prediction error of the unit load is less than 7%. The method can guide the start and stop optimization of the important auxiliary machines of the power plant generator set, and is beneficial to improving the economic operation level of the set. The short-term load prediction is an important means for running and controlling the generator set, and the method has important significance for operation decisions such as scheduling of generating capacity, reliability analysis, maintenance of generator set equipment and the like.
The FPA-GRU algorithm based power generation load prediction method provided by the present invention is the units and algorithm steps of the examples described in connection with the embodiments disclosed herein, and can be implemented in electronic hardware, computer software, or a combination of both, and in the above description the components and steps of the examples have been generally described in terms of functionality in order to clearly illustrate the interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
The FPA-GRU algorithm-based power generation load prediction method provided by the present invention may write program code for performing the operations of the present disclosure in any combination of one or more programming languages, including an object-oriented programming language such as Java, C + +, etc., and a conventional procedural programming language such as "C" or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., through the internet using an internet service provider).
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. 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 invention. Thus, the present invention 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 method for predicting a power generation load, the method comprising:
step 1: acquiring load prediction data in a preset time period from a database;
step 2: normalizing the load prediction data;
and step 3: constructing a random forest model calculated based on the gini coefficient, and calculating the correlation weight between the corresponding measuring points of the load prediction data to serve as an initial weight;
and 4, step 4: starting to construct an attention gate, inputting the initial weight into the attention gate if the attention gate is constructed for the first time, and inputting the hyperparameter obtained by the flower pollination optimization algorithm into the attention gate if the attention gate is not constructed for the first time;
and 5: building a GRU model, and adding the attention gate weight data built in the step 4 into the GRU model for training;
step 6: setting a threshold TthreCalculating the error between the predicted value and the actual value, if the error is FerrorIf the value is larger than the threshold value, entering the step 7, otherwise, outputting an optimal GRU model;
and 7: and obtaining the optimal hyper-parameter according to a flower pollination optimizing algorithm.
2. The power generation load prediction method according to claim 1,
in the step 1, shutdown data in the load prediction data in the preset time period is identified and rejected according to start and shutdown, and the rejected load prediction data is used as historical data in the steps 2 to 7.
3. The power generation load prediction method according to claim 1,
in the step 2, the step of the method is carried out,
load prediction data F for time aaCarrying out data normalization processing to obtain a normalized data matrix Fa-normTraining data F according to the following formulaaPerforming data mapping to [0,1 ]]Interval(s)
Wherein, Fa-normIs a normalized value of time a, Fa-maxIs the maximum value of the parameter, Fa-minIs the most significant of the parameterA small value.
4. The power generation load prediction method according to claim 1,
in the step 3, the step of the method is that,
calculating each measuring point gini index score of each decision tree;
wherein m represents the mth measuring point, K represents that the m measuring points contain K categories, and Pmk represents the proportion of the category K in the node m;
calculating the weight of the features in the nodes:
wherein GIi and GIr respectively represent Gini indexes of two new nodes after branching;
accumulating and counting gini index scores of each measuring point to obtain average scores, normalizing, and taking the normalized scores as initial weights for calculating historical data among the measuring points;
if the feature XjIn the set of nodes M present in the decision tree i, then XjThe reduction in the pureness at the ith tree was:
wherein i represents the ith decision tree, j represents the jth measuring point of the ith tree, and M represents the measuring point set of the decision tree; assuming that the random forest has n trees, the reduced impurity level of each measurement point in the random forest is:
wherein j is the jth node, n is the total number of decision trees
And finally, normalizing all the obtained importance scores to obtain weights:
5. the power generation load prediction method according to claim 1,
in the step 4, the process of the method,
according to the attention gate, the importance sampling is carried out on the load prediction data, and the attention gate weight data:
w＝[w1，w2，w3，...，w6]。
6. the power generation load prediction method according to claim 1,
in the step 5, the process is carried out,
sending the load prediction data into the GRU neural network by using the attention gate weight in the step 4;
the GRU neural network is provided with GRU synthesized by 2 sigmoid functions and multiplication operations to control information selection, and the memory filling and the current input data are converted to be between 0 and 1 by the aid of the sigmoid functions to realize data memory and discarding;
in the GRU neural network model, the hidden layer state at the current time is the sum of the hidden layer state at the previous time and the current hidden layer activation state, that is:
the reset gate and the update gate are the combination of the hidden layer states at a time above the input data at the current time:
rt＝σ(wrxt+urht-1) (8)
Zt＝σ(wzxt+uzht-1) (9)
wherein: σ is sigmoid function, rtAnd ZtLimited between 0 and 1, Wr、ur、WZ、uZIs the weight of the neural network;
reset gate rtFor the value h of the hidden layer at the last momentt-1Resetting is performed, resetting the gate configuration to the extent, r, that was retained in the hidden layer at the previous timetThe closer to 1, the more data is remained in the previous-time hidden layer, and the previous-time hidden layer data is subjected to reset gate processing and then is input with the current xtCombining and obtaining the activation state of the hidden layer at the current moment through a Tanh activation function
nz＝wxt+u(rt⊙ht-1) (11)
In the formula: an indicator product operation, W, u is GRU neural network weight
htMainly by the refresh gate ztCompletion of ztThe closer to 0 the value of (b) is, the greater the degree of retention of the current time information is.
7. The power generation load prediction method according to claim 1,
in the step 7, the process is carried out,
7.1 initializing parameters, the number n of flower populations and the conversion probability p;
7.2 calculating the fitness value of each solution, and solving the current optimal solution and the optimal value;
7.3 if the condition that the conversion probability p is more than rand is satisfied, updating the solution according to the following formula and carrying out border crossing treatment;
wherein
The solutions of the t +1 th generation and the t th generation respectively; theta represents a step control quantity, generally 1,
point-to-point multiplication is shown, Levy is a vegetable dimension flight search path, and the value range of beta is [1, 3 ]]Generally, 3/2, Levy calculation formula is taken
Wherein: the value range of λ is [0, 2], generally 3/2 is taken, Γ (λ) is a standard gamma function, S is a search path, and the calculation formula of S is:
where μ, V are random numbers obeying a normal distribution
7.4 if the p > rand condition is not satisfied, performing optimal solution update according to the following formula, and performing border crossing treatment:
wherein: e is [0,1 ]]Subject to uniformly distributed random numbers,
are the same plant;
7.5 calculating a fitness value corresponding to the new solution obtained by 7.3 or 7.4, if the fitness of the new solution is excellent, respectively replacing the current solution and the current fitness value with the fitness value corresponding to the new solution and the new solution, and otherwise, keeping the current solution and the current fitness;
7.6 if the fitness value corresponding to the new solution is better than the global optimum value, updating the global optimum solution and the global optimum value;
7.7 judging the end condition, if yes, outputting the optimal solution, otherwise, returning to 7.3.
8. The power generation load prediction method according to claim 1, characterized by further comprising:
step 11: acquiring weather forecast data of a future preset time period from a database as independent variable data;
step 12: normalizing the independent variable data according to a training normalization model;
step 13: loading the optimal GRU model output in the step 6 of the training stage;
loading an optimal GRU model to obtain optimal network model parameters;
step 14: inputting the preprocessed independent variable data into an optimal GRU model, and operating the model;
acquiring optimal network model parameters, loading data, and calculating the states of a reset door and an update door at the predicted time through formulas (20), (21) and (22);
rt＝σ(wrxt+urht-1) (21)
zt＝σ(wzxt+uzht-1) (22)
finally, obtaining a hidden layer state combination;
step 15: calculating load prediction value F of output independent variable dataforecast；
Calculating the hidden layer h of the current time according to the formula (23) by the input datatLater, the output, w, of the GRU network model may be obtainedyIs the weight;
Fforecast＝σ(wyht) (23)。
9. the power generation load prediction method according to claim 8, characterized by further comprising:
step 11 further comprises:
acquiring meteorological data F of future preset time periodb: rainfall, air pressure, wind speed, air temperature and humidity are used as independent variable data;
wherein 5 columns are respectively precipitation, air pressure, wind speed, air temperature and humidity
10. The power generation load prediction method according to claim 8, characterized by further comprising:
in step 12, data F is processedbCarrying out data normalization processing to obtain a normalized data matrix Fb-norm；
Training data F according to the following formulabPerforming data mapping to [0,1 ]]An interval;
wherein, Fb-normAs normalized value of time, Fb-maxAs maximum of corresponding measuring points of the normalized model, Fb-minThe minimum value of the corresponding measuring point of the normalized model is obtained.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110983783 | 2021-08-25 | ||
| CN202110983783X | 2021-08-25 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113610328A true CN113610328A (en) | 2021-11-05 |
Family
ID=78343141
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111087872.2A Pending CN113610328A (en) | 2021-08-25 | 2021-09-16 | Power generation load prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113610328A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114509704A (en) * | 2022-02-15 | 2022-05-17 | 湖南小快智造电子科技有限公司 | Intelligent monitor for safety power utilization |
| CN115660226A (en) * | 2022-12-13 | 2023-01-31 | 国网冀北电力有限公司 | Power load prediction model construction method and construction device based on digital twins |
| CN117553821A (en) * | 2024-01-12 | 2024-02-13 | 中闽(平潭)风电有限公司 | Robot inspection global path planning method for wind farm booster station |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110728391A (en) * | 2018-07-17 | 2020-01-24 | 广西大学 | Depth regression forest short-term load prediction method based on expandable information |
| US20200161867A1 (en) * | 2018-11-15 | 2020-05-21 | Hefei University Of Technology | Method, system and storage medium for load dispatch optimization for residential microgrid |
| CN111738512A (en) * | 2020-06-22 | 2020-10-02 | 昆明理工大学 | Short-term power load prediction method based on CNN-IPSO-GRU hybrid model |
| CN112581172A (en) * | 2020-12-18 | 2021-03-30 | 四川中电启明星信息技术有限公司 | Multi-model fusion electricity sales quantity prediction method based on empirical mode decomposition |
| CN112819192A (en) * | 2020-11-09 | 2021-05-18 | 江苏科技大学 | Method for predicting short-term power load of RF _ GRU network based on swarm algorithm optimization |
| CN112906292A (en) * | 2021-01-26 | 2021-06-04 | 西安热工研究院有限公司 | Method, system, equipment and storage medium for plant-level heat and power load online optimal distribution of cogeneration unit |
-
2021
- 2021-09-16 CN CN202111087872.2A patent/CN113610328A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110728391A (en) * | 2018-07-17 | 2020-01-24 | 广西大学 | Depth regression forest short-term load prediction method based on expandable information |
| US20200161867A1 (en) * | 2018-11-15 | 2020-05-21 | Hefei University Of Technology | Method, system and storage medium for load dispatch optimization for residential microgrid |
| CN111738512A (en) * | 2020-06-22 | 2020-10-02 | 昆明理工大学 | Short-term power load prediction method based on CNN-IPSO-GRU hybrid model |
| CN112819192A (en) * | 2020-11-09 | 2021-05-18 | 江苏科技大学 | Method for predicting short-term power load of RF _ GRU network based on swarm algorithm optimization |
| CN112581172A (en) * | 2020-12-18 | 2021-03-30 | 四川中电启明星信息技术有限公司 | Multi-model fusion electricity sales quantity prediction method based on empirical mode decomposition |
| CN112906292A (en) * | 2021-01-26 | 2021-06-04 | 西安热工研究院有限公司 | Method, system, equipment and storage medium for plant-level heat and power load online optimal distribution of cogeneration unit |
Non-Patent Citations (4)
| Title |
|---|
| 庞昊 等: "基于多神经网络融合的短期负荷预测方法", 《电力自动化设备》, vol. 40, no. 06, pages 37 - 42 * |
| 李华 等: "基于花朵授粉算法的组合式风速预测", 《科学技术与工程》, vol. 20, no. 04, pages 1436 - 1441 * |
| 牛庆 等: "基于花朵授粉算法和BP神经网络的短期负荷预测", 《电网与清洁能源》, vol. 36, no. 10, pages 28 - 32 * |
| 黄青平 等: "基于模糊聚类与随机森林的短期负荷预测", 《电测与仪表》, vol. 54, no. 23, pages 41 - 46 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114509704A (en) * | 2022-02-15 | 2022-05-17 | 湖南小快智造电子科技有限公司 | Intelligent monitor for safety power utilization |
| CN115660226A (en) * | 2022-12-13 | 2023-01-31 | 国网冀北电力有限公司 | Power load prediction model construction method and construction device based on digital twins |
| CN117553821A (en) * | 2024-01-12 | 2024-02-13 | 中闽(平潭)风电有限公司 | Robot inspection global path planning method for wind farm booster station |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| CN113610328A (en) | Power generation load prediction method | |
| CN110751318B (en) | Ultra-short-term power load prediction method based on IPSO-LSTM | |
| CN111027772B (en) | Multi-factor short-term load prediction method based on PCA-DBILSTM | |
| CN111915092B (en) | Ultra-short-term wind power prediction method based on long-short-term memory neural network | |
| CN110443417A (en) | Multi-model integrated load prediction method based on wavelet transformation | |
| CN114757427B (en) | Autoregressive-corrected LSTM intelligent wind power plant ultra-short-term power prediction method | |
| Liu et al. | Heating load forecasting for combined heat and power plants via strand-based LSTM | |
| CN111241755A (en) | Power load prediction method | |
| CN112329990A (en) | User power load prediction method based on LSTM-BP neural network | |
| CN114792156A (en) | Photovoltaic output power prediction method and system based on curve characteristic index clustering | |
| CN113822418A (en) | Wind power plant power prediction method, system, device and storage medium | |
| Kim et al. | Tutorial on time series prediction using 1D-CNN and BiLSTM: A case example of peak electricity demand and system marginal price prediction | |
| CN114757104B (en) | Method for constructing hydraulic real-time regulation and control model of series gate group water transfer project | |
| CN112836876B (en) | Power distribution network line load prediction method based on deep learning | |
| CN111697560B (en) | Method and system for predicting load of power system based on LSTM | |
| CN116187835A (en) | Data-driven-based method and system for estimating theoretical line loss interval of transformer area | |
| CN116205334A (en) | Intelligent prediction method and device for power grid load | |
| CN117390550A (en) | Low-carbon park carbon emission dynamic prediction method and system considering emission training set | |
| CN117134315A (en) | Distribution transformer load prediction method and device based on BERT algorithm | |
| CN116402194A (en) | Multi-time scale load prediction method based on hybrid neural network | |
| Kumar et al. | A Comparative Analysis of Time Series and Machine Learning Models for Wind Speed Prediction | |
| CN115619449A (en) | Deep learning model-based fertilizer price prediction method, device and storage medium | |
| CN115759343A (en) | E-LSTM-based user electric quantity prediction method and device | |
| CN115545319A (en) | Power grid short-term load prediction method based on meteorological similar day set |
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 |