CN105809264A - Electrical load predicting method and device - Google Patents
Electrical load predicting method and device Download PDFInfo
- Publication number
- CN105809264A CN105809264A CN201410838273.3A CN201410838273A CN105809264A CN 105809264 A CN105809264 A CN 105809264A CN 201410838273 A CN201410838273 A CN 201410838273A CN 105809264 A CN105809264 A CN 105809264A
- Authority
- CN
- China
- Prior art keywords
- electrical load
- historical data
- influence
- data
- network 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.)
- Granted
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides an electrical load predicting method and device. The method comprises the steps of training a first deep belief network model for predicting electrical load according to the historic data of the electrical load using the historic data of the electrical load to obtain a first prediction result for predicting future electrical load; training a second deep belief network model for predicting the electrical load according to the historic data of the electrical load and the historic data of a predefined leading influencing factor using the historic data of the electrical load and the historic data of the predefined leading influencing factor to obtain a second prediction result for predicting future electrical load; and predicting the future electrical load according to the first prediction result and the second prediction result. The method and the device improve the precision of electrical load prediction, particularly under the condition that the historic data is partially missing or huge.
Description
Technical field
The present invention relates to intelligent predicting field, particularly relate to a kind of electrical load Forecasting Methodology and device.
Background technology
Power prediction is a vital task in the operation of current Utilities Electric Co. and planning.Only reasonable prediction future electrical energy load, just can carry out rational power planning.But, limited in historical information or when history information data amount is very big, to accurately predict what electrical load was difficult to.
Summary of the invention
In view of this, one of problem that one embodiment of the present of invention solves is the precision improving electrical load prediction.
According to one embodiment of present invention, provide a kind of electrical load Forecasting Methodology, including: predict the first degree of depth belief network model of electrical load for the historical data according to electrical load with the training of the historical data of electrical load, thus producing for predicting that the first of future electrical energy load predicts the outcome；The second degree of depth belief network model of the historical data prediction electrical load of the historical data according to electrical load and predefined leading factor of influence is trained, thus generation is for predicting that the second of future electrical energy load predicts the outcome by the historical data of the historical data of electrical load and predefined leading factor of influence；Predict the outcome according to first and predict the outcome with second, it was predicted that future electrical energy load.
Alternatively, train by the historical data of electrical load and predict that the step of the first degree of depth belief network model of electrical load farther includes for the historical data according to electrical load: when the historical data excalation of electrical load, the historical data of electrical load described in first degree of depth belief network model part according to the electrical load data prediction disappearance before the part of disappearance completion.
Alternatively, the method also includes: be used for predefining according to the predefined part dominating factor of influence prediction disappearance before the part of disappearance completion the 3rd degree of depth belief network model of the historical data of leading factor of influence when the historical data lack part of predefined leading factor of influence with the historical data training of predefined leading factor of influence, thus completion predefines the historical data of leading factor of influence when the historical data excalation of predefined leading factor of influence.
Alternatively, first, second, each in 3rd degree of depth belief network model includes multilayer logic node, wherein the bottom logical node of multilayer logic node receives input data, the logical node of other layer except bottom logical node and the logical node of next layer are related, described relation coefficient represents, thus first, second, each in 3rd degree of depth belief network model is expressed as coefficient matrix, first, second, the training process of each in 3rd degree of depth belief network model includes: obtain the coefficient in coefficient matrix from bottom to top for multilayer logic node from input data；The coefficient in coefficient matrix is verified from top to bottom by the backward multilayer logic node that fits to.
Alternatively, the coefficient value of first, second and third degree of depth belief network model adopts greedy nothing supervision feature learning to obtain.
Alternatively, the historical data of electrical load is the historical data of the electrical load after normalization, and the historical data of predefined leading factor of influence is the historical data of the predefined leading factor of influence after normalization.
Alternatively, predefined leading factor of influence includes at least one in weather data, temperature data, humanistic environment data.
One embodiment of the present of invention is provided, provide a kind of electrical load prediction unit, including: the first training unit, it is configured to predict for the historical data according to electrical load the first degree of depth belief network model of electrical load with the historical data training of electrical load, thus producing for predicting that the first of future electrical energy load predicts the outcome；Second training unit, it is configured to the historical data with electrical load and the predefined historical data dominating factor of influence trains the historical data of the historical data according to electrical load and predefined leading factor of influence to predict the second degree of depth belief network model of electrical load, thus generation is for predicting that the second of future electrical energy load predicts the outcome；Predicting unit, being configured to predicts the outcome according to first predicts the outcome with second, it was predicted that future electrical energy load.
Alternatively, first training unit is configured to when the historical data excalation of electrical load, by first degree of depth belief network model part according to the electrical load data prediction disappearance before the part of disappearance the historical data of completion electrical load.
Alternatively, electrical load prediction unit also includes: the 3rd training unit, it is configured to train the 3rd degree of depth belief network model of the historical data for predefining leading factor of influence according to the predefined part dominating factor of influence prediction disappearance before the part of disappearance completion when the historical data excalation of predefined leading factor of influence by the historical data of predefined leading factor of influence, thus completion predefines the historical data of leading factor of influence when the historical data excalation of predefined leading factor of influence.
Alternatively, first, second, each in 3rd degree of depth belief network model includes multilayer logic node, wherein the bottom logical node of multilayer logic node receives input data, the logical node of other layer except bottom logical node and the logical node of next layer are related, described relation coefficient represents, thus first, second, each in 3rd degree of depth belief network model is expressed as coefficient matrix, first, second, the training process of each in 3rd degree of depth belief network model includes: obtain the coefficient in coefficient matrix from bottom to top for multilayer logic node from input data；The coefficient in coefficient matrix is verified from top to bottom by the backward multilayer logic node that fits to.
Alternatively, the coefficient value of first, second and third degree of depth belief network model adopts greedy nothing supervision feature learning to obtain.
Alternatively, the historical data of electrical load is the historical data of the electrical load after normalization, and the historical data of predefined leading factor of influence is the historical data of the predefined leading factor of influence after normalization.
Alternatively, predefined leading factor of influence includes at least one in weather data, temperature data, humanistic environment data.
The first degree of depth belief network model predicting electrical load for the historical data according to electrical load is not use only due to the embodiment of the present invention, additionally use the second degree of depth belief network model of the historical data prediction electrical load of the historical data according to electrical load and predefined leading factor of influence, make to predict the outcome and not only reflect the variation tendency of electrical load itself, also reflect the variation tendency on the impact of electrical load such as relevant leading factor of influence (such as environment, weather), more objective so that predicting the outcome, improve the precision of prediction.
Owing to, in one embodiment of the present of invention, the first degree of depth belief network model is the model predicting electrical load for the historical data according to electrical load, and it reflects the trend of electrical load Self-variation.The trend of electrical load Self-variation affects the synergistic result of all factors of influence of electrical load often.These factors of influence there are many accidentalia.Therefore, predicting with the model including very big accidentalia, making to predict the outcome often also has occasionality.Some important factors of influence are often buried in the factor of influence of a large amount of occasionality, make the inaccuracy that predicts the outcome.And the second degree of depth belief network model compensate for this defect well, it investigates some predefined leading factor of influence impacts on predicting the outcome emphatically, eliminate the impact of occasionality, in enabling some factors being likely to play decisive role for predicting the outcome to be reflected in and predict the outcome.In conjunction with two models, it is possible to both reduced the impact of the middle occasionality that predicts the outcome, have recognised again the middle disappearance that predicts the outcome objectively should have the impact of certain occasionality, improves precision of prediction.
Additionally, due to embodiment of the present invention employing is degree of depth belief network model, degree of depth belief network model is a kind of multilayered nonlinear network structure, a kind of simple function expression and the relation of some parameter expressions complexity can be used, enable in particular to processing accuracy when improving towards huge historical data and efficiency.Therefore, degree of depth belief network model is adopted to improve precision and the efficiency of prediction electrical load.
Additionally, the coefficient value of the degree of depth belief network model owing to adopting in the embodiment of the present invention adopts greedy nothing supervision feature learning to obtain, the greedy feature without supervision feature learning is that it can feature that clearly expression data is hidden behind and complicated pattern, even if therefore historical data excalation, it still can pass through to find data relation behind, thus finding, from the historical data of incomplete electrical load, the relation that data hide behind, it is achieved predict accurately.
Accompanying drawing explanation
Other feature of the present invention, feature, advantage and benefit be will become apparent from by the detailed description below in conjunction with accompanying drawing.
Fig. 1 illustrates the flow chart of electrical load Forecasting Methodology according to an embodiment of the invention.
Fig. 2 illustrates the flow chart of electrical load Forecasting Methodology in accordance with another embodiment of the present invention.
Fig. 3 illustrates the structure of the degree of depth belief network model of a standard.
Fig. 4 illustrates the example of the first and second degree of depth belief network models adopted according to one embodiment of the invention.
Fig. 5 A-5B illustrates the example of two the 3rd degree of depth belief network models adopted according to a further embodiment of the invention.
Fig. 6 illustrates the block diagram of electrical load prediction unit according to an embodiment of the invention.
Fig. 7 illustrates the block diagram of electrical load prediction unit in accordance with another embodiment of the present invention.
Fig. 8 illustrates the structure chart of the pre-measurement equipment of electrical load according to an embodiment of the invention.
Detailed description of the invention
Below, will be described in detail with reference to accompanying drawings each embodiment of the present invention.
Fig. 1 illustrates the flow chart of electrical load Forecasting Methodology according to an embodiment of the invention 1.This electrical load Forecasting Methodology is usable in Utilities Electric Co.'s electrical load according to existing historical data following certain time (such as 1 year, one month) a certain area (such as East China, Haidian District, a certain community) of prediction or a certain unit, i.e. power consumption, thus predicting the outcome according to this carry out power scheduling and planning.
In step S1, train the first degree of depth belief network model predicting electrical load for the historical data according to electrical load by the historical data of electrical load, thus producing for predicting that the first of future electrical energy load predicts the outcome.
Electrical load data are exactly the data of power consumption.The historical data of electrical load is exactly the power consumption of time per unit (year, month, day) in history.Such as, assume to be currently in December, 2014, when needing the power consumption of prediction 1-12 this community every month month in 2015, the power consumption of this community 2014 1-12 every month month, the power consumption of the 1-12 every month month in 2013, the power consumption of the 1-12 every month month in 2012 ... contribute to the historical data of the electrical load of the electrical load of prediction 1-12 this community every month month in 2015.
Degree of depth belief network model is the one of known degree of depth study architecture.As it is shown on figure 3, it includes visible layer 41 and hidden layer 42.Visible layer 41 is in the bottom (the layer v in Fig. 3 of degree of depth belief network model.It (is three layers h in figure 3 that hidden layer is probably multilamellar1、h2、h3)。
Degree of depth belief network model be characterized in that can from the feature of extracting data data, thus finding data relation behind.Therefore, each layer of Fig. 3 may be considered the feature in layer extracting data from top to bottom, it has been found that the process of the relation that data are hidden behind.It is characterized by the vector of certain dimension.Generally, it is seen that the node in layer 41 receives multiple input data.Each node of the first hidden layer h1 goes out each input data feature in a certain respect from each input extracting data of visible layer 41 respectively.So, the first hidden layer h1Each node just there occurs with each node (input data) of visible layer 41 relation (see restriction Boltzmann machine energy model).In like manner, the second hidden layer h2Each node respectively from the first hidden layer h1Each input data in extract feature in a certain respect further.So, the second hidden layer h2Each node just with the first hidden layer h1Each node there occurs relation.In like manner, the 3rd hidden layer h3Each node just with the second hidden layer h2Each node there occurs relation.The relation of the specific node of last layer in degree of depth belief network model Yu the specific node of next layer is regarded as the coefficient of degree of depth belief network model.Coefficient is generated in model training process by cognitive weight and generation weight.Degree of depth belief network model is considered as a coefficient matrix including a lot of coefficient.
Build that degree of depth belief network model is actual includes two processes: previous process is successively to extract feature from input data, obtains the relation between each node of last layer and each node of next layer, namely obtain the process of each coefficient in coefficient matrix；Later process is to be verified the process of each coefficient coefficient matrix from top to bottom from degree of depth belief network model by backward matching.Degree of depth belief network model depth belief network model depth belief network model is in an embodiment of the present invention, such as, when needing to predict the power consumption of certain community in December, 2014 4-6 every day day, assume the power consumption L1 of every day day of storage You Gai community in December, 2014 1-3 in history, L2, L3, just can by the power consumption L1 of this community in December, 2014 1-3 every day day, L2, L3 inputs the first degree of depth belief network model N1 respectively as each input data, produce first of the power consumption for predicting this community in December, 2014 4-6 every day day to predict the outcome Y1, as shown in Figure 4.Certainly, it is more than example.In reality, in order to improve the precision predicted the outcome, it will usually use the power consumption of longer time as historical data.
Degree of depth belief network model can have different establishment algorithms.Except above-mentioned restriction Boltzmann machine energy model, it is also possible to adopt the such as method such as sparse coding, convolutional neural networks.
It addition, when the historical data lack part of electrical load, it is also possible to by first degree of depth belief network model part according to the electrical load data prediction disappearance before the part of disappearance completion.
For example, it is assumed that the power consumption of only storage You Gai community in December, 2014 1-2 every day day, the power consumption disappearance of 3 days.At this point it is possible to the power consumption of Xian Zhijianggai community in December, 2014 1-2 every day day inputs the first degree of depth belief network model.First degree of depth belief network model electricity demand forecasting Chu Gai community in December, 2014 power consumption of 3 days according to this community in December, 2014 1-2 every day day, then with the completion that predicts the outcome.Again the power consumption of in December, 2014 1-3 every days day after completion is inputted the first degree of depth belief network model, the first degree of depth belief network model output predict the outcome Y1 for predict the power consumption of this community in December, 2014 4-6 every day day first.So, the problem solving the precision how improving prediction electrical load when historical data excalation further.
In step S3, the second degree of depth belief network model of the historical data prediction electrical load of the historical data according to electrical load and predefined leading factor of influence is trained, thus generation is for predicting that the second of future electrical energy load predicts the outcome by the historical data of the historical data of electrical load and predefined leading factor of influence.
Leading factor of influence refers to the principal element affecting electrical load.The leading factor of influence of electrical load is likely to have a lot, tends not to exhaustive in practice, generally adopts some the leading factors of influence pre-defined, namely predefine and dominate factor of influence.Such as, it includes at least one in weather data, temperature data, humanistic environment data etc..Weather data refers to the data relevant with the weather conditions such as fine, cloudy, cloudy.Such as, when weather is cloudy, often people likes staying in, therefore generally may Multifunctional electric.Temperature data refers to the data relevant with temperature.Such as, in cold winter and extremely hot summer, owing to turning on the aircondition, often Multifunctional electric.Humanistic environment data are the data that reflection refers to may have influence on the social-human factors of electrical load except the natural environment such as weather, temperature, such as coal system issues the incentive measure that worker's moon power consumption is little recently, this community is again coal system worker community, and electricity consumption can be lacked after being likely in this community.
Owing to predefined leading factor of influence includes weather data, temperature data, humanistic environment data etc., the historical data of predefined leading factor of influence potentially includes the historical data of weather data, temperature data, humanistic environment data etc..The still power consumption to predict certain community in December, 2014 4-6 day, as shown in Figure 4, except the power consumption with in December, 2014 1-3 day, also weather data and temperature data with in December, 2014 1-3 day are predicted.Weather data respectively W1, W2, W3 of in December, 2014 1-3 day.Temperature data respectively T1, T2, T3 of in December, 2014 1-3 day.So, the data of each node of visible layer being input to the second degree of depth belief network model N4 are not one-dimensional datas, but three-dimensional vector, it is (L1, W1, T1), (L2, W2, T2), (L3, W3, T3) respectively.The output of the second degree of depth belief network model N4 is Y4, namely second predicts the outcome.It is different that the Y1 that predicts the outcome from first lays particular emphasis on the variation tendency investigating electrical load data itself, and second Y4 that predicts the outcome lays particular emphasis on investigation predefined factor of influence of dominating the electrical load of the impact of electrical load data is predicted the outcome.
Degree of depth belief network model not only can process one-dimensional input, namely at input historical data a1、a2……anWhen (n is natural number, represents the number of input historical data), according to the historical data of input prediction Future Data an+1、an+2..., also can process multidimensional input, namely accept historical data a1、a2……anWhile, also accept constraints b1、b2……bn,c1、c2……cnDeng input, by feature extraction layer by layer, study is to a1、a2……anAnd b1、b2……bn,c1、c2……cnDeng relation, output consider constraints b1、b2……bn,c1、c2……cnDeng to a1、a2……anWhen affecting prediction Future Data an+1、an+2…….This feature of degree of depth belief network model is used by the present inventor in the present invention, not only create the first of following electrical load of the historical data trend according only to electrical load with the first degree of depth belief network model to predict the outcome, also create with the second degree of depth belief network model and investigated various predefined leading factor of influence (such as weather data, temperature data etc.) the second of the impact of electrical load is predicted the outcome, comprehensive first and second predict the outcome, just can produce both to have considered the variation tendency of electrical load itself, it not again isolated investigate electrical load itself and investigate one of surrounding environment influence yet and more objectively predict the outcome.
In step s 4, predict the outcome according to first and second predict the outcome, it was predicted that future electrical energy load.
In one embodiment, it is possible to Y1 and the second that predict the outcome the first simply Y4 that predicts the outcome averages, future electrical energy load Y=(Y1+Y4)/2 of prediction is obtained.
In another embodiment, it is possible to be weighted by average method and obtain the future electrical energy load of prediction, i.e. Y=α Y1+ β Y4, wherein alpha+beta=1, α and β can be set by empirical value, it is also possible to according to the satisfaction finally predicted the outcome, constantly regulating α and β.
It is of course also possible to predicted the outcome Y4 according to first Y1 and the second that predict the outcome by other ways known, obtain the future electrical energy load Y of prediction.
Fig. 2 illustrates the flow chart of electrical load Forecasting Methodology in accordance with another embodiment of the present invention.Compared to Figure 1, this embodiment increases step S2: train by the historical data of predefined leading factor of influence and be used for dominate the part of factor of influence prediction disappearance the 3rd degree of depth belief network model of completion according to predefining before the part of disappearance when the historical data lack part of predefined leading factor of influence, thus completion predefines the historical data of leading factor of influence when the predefined historical data lack part dominating factor of influence.
The power consumption of Reng Yiyongmou community in December, 2014 1-3 day, weather data, temperature data predict that the power consumption of this community in December, 2014 4-6 day is example.Assume the weather data of only storage You Gai community in December, 2014 1-2 every day day, the weather data disappearance of 3 days.Now, as shown in Figure 5A, it is possible to weather data input the 3rd degree of depth belief network model N2 of Xian Zhijianggai community in December, 2014 1-2 every day day.3rd degree of depth belief network model N2 dopes this community in December, 2014 weather data of 3 days according to the weather data of this community in December, 2014 1-2 every day day, and the coefficient matrix of the 3rd degree of depth belief network model N2 then obtained with training is inverse calculates completion missing data.The weather data of in December, 2014 1-3 every days day after completion for the second degree of depth belief network model N4 when the prediction of step S3.Y2 is one about future weather and predicts the outcome, but do not use in embodiments of the present invention, because the embodiment of the present invention uses the 3rd degree of depth belief network model N2 to be primarily to the completion carrying out weather data when the historical data of lack part weather, do not use the future weather of the 3rd degree of depth belief network model N2 prediction.
In like manner, it is assumed that the only temperature data of storage You Gai community in December, 2014 1-2 every day day, the temperature data disappearance of 3 days.Now, as shown in Figure 5 B, it is possible to temperature data input the 3rd degree of depth belief network model N3 of Xian Zhijianggai community in December, 2014 1-2 every day day.3rd degree of depth belief network model N3 dopes this community in December, 2014 temperature data of 3 days according to the temperature data of this community in December, 2014 1-2 every day day, and the coefficient matrix of the 3rd degree of depth belief network model N3 then obtained with training is inverse calculates completion missing data.The temperature data of in December, 2014 1-3 every days day after completion for the second degree of depth belief network model N4 when the prediction of step S3.Y3 is one about future temperature and predicts the outcome, but do not use in embodiments of the present invention, because the embodiment of the present invention uses the 3rd degree of depth belief network model N2 to be primarily to the completion carrying out temperature data when the historical data of lack part temperature, do not use the future temperature of the 3rd degree of depth belief network model N2 prediction.
So, the problem just solving the precision how improving prediction electrical load when the historical data excalation historical data excalation of temperature (the historical data excalation of such as weather) of predefined leading factor of influence further.
Preferably, in above process, the coefficient value of first, second and third degree of depth belief network model adopts greedy nothing supervision feature learning to obtain.It is to say, the feature extraction of the different layers of first, second and third degree of depth belief network model is based on greedy without supervision feature learning.This makes the feature extraction from next layer to last layer very natural, it is not necessary to manual intervention.Feature extraction at the different layers of some degree of depth belief network model needs artificially extraction result to be corrected.The pattern of the historical data can clearly expressed due to greediness nothing supervision feature learning and the feature hidden of the historical data of data correlation factor and complexity, when the historical data of the historical data of lack part or part data correlation factor, it is easy to by finding that the relation hidden between data corrects it.Therefore, even if remain to the essence of reducing loaded pattern when historical data lacks.
Preferably, the first degree of depth belief network model is being trained by the historical data of electrical load, or train the second degree of depth belief network model by the historical data of the historical data of electrical load and predefined leading factor of influence, or during with historical data training the 3rd degree of depth belief network model of predefined leading factor of influence, all adopts the predefined historical data dominating factor of influence after the historical data of the electrical load after normalization or normalization.Normalization refers to and the electrical load data in not commensurate, different range or predefined leading factor of influence is unified into the electrical load data in same unit, same scope or predefined leading factor of influence.Here scope can be the multiple scopes such as the scope of the scope of region, statistical time range.A normalized example as unit is, the use degree Celsius that the temperature data of storage has is expressed, and what have expresses by degrees Fahrenheit, it is possible to unified being adjusted to uses a degree Celsius expression.Normalization as the scope of region, in December, 2014 power consumption of 1 day of storage is the power consumption of whole community, in December, 2014 power consumption of 2 days of storage is again the power consumption of half community (building in such as You Liangge building, whole community), in December, 2014 power consumption of 2 days can be multiplied by 2 power consumptions being reduced into whole community so that it is there is same comparison basis.A normalized example as the scope of statistical time range is, in December, 2014 power consumption of 1 day of storage is to add up the whole community power consumption of 24 hours, in December, 2014 power consumption of 2 days of storage is the whole community power consumption of 12 hours, in December, 2014 power consumption of 2 days can be multiplied by certain experiences coefficient (owing to daytime and electricity consumption in evening are unbalanced, it is not necessarily 2) estimate the whole community power consumption of 24 hours so that it is there is same comparison basis.Through such normalization, also improve precision of prediction.
As shown in Figure 6, an alternative embodiment of the invention provides a kind of electrical load prediction unit 2, including first training unit the 21, second training unit 23 predicting unit 24.First training unit 21 is configured to predict for the historical data according to electrical load the first degree of depth belief network model of electrical load with the historical data training of electrical load, thus producing for predicting that the first of future electrical energy load predicts the outcome.Second training unit 23 is configured to the historical data with electrical load and the predefined historical data dominating factor of influence trains the historical data of the historical data according to electrical load and predefined leading factor of influence to predict the second degree of depth belief network model of electrical load, thus generation is for predicting that the second of future electrical energy load predicts the outcome.Predicting unit 24 is configured to predict the outcome according to first predict the outcome with second, it was predicted that future electrical energy load.Each unit in Fig. 6 can utilize the mode of software, hardware (such as integrated circuit, FPGA etc.) or software and hardware combining to realize.Predefined leading factor of influence
Alternatively, first training unit 21 is configured to when the historical data excalation of electrical load, by first degree of depth belief network model part according to the electrical load data prediction disappearance before the part of disappearance the historical data of completion electrical load.
Alternatively, as shown in Figure 7, electrical load prediction unit 2 can also include: the 3rd training unit 22, it is configured to train the 3rd degree of depth belief network model of the historical data for predefining leading factor of influence according to the predefined part dominating factor of influence prediction disappearance before the part of disappearance completion when the historical data excalation of predefined leading factor of influence by the historical data of predefined leading factor of influence, thus completion predefines the historical data of leading factor of influence when the historical data excalation of predefined leading factor of influence.
Alternatively, the coefficient value of first, second and third degree of depth belief network model adopts greedy nothing supervision feature learning to obtain.
Alternatively, the historical data of electrical load is the historical data of the electrical load after normalization, and the historical data of predefined leading factor of influence is the historical data of the predefined leading factor of influence after normalization.
Alternatively, predefined leading factor of influence also includes at least one in weather data, temperature data, humanistic environment data.
With reference now to Fig. 8, it illustrates the schematic diagram according to the pre-measurement equipment 3 of the electrical load of one embodiment of the invention.As shown in Figure 8, the pre-measurement equipment 3 of electrical load can include memorizer 31 and processor 32.Memorizer 31 can store executable instruction.Processor 32 can according to the stored executable instruction of memorizer 31, it is achieved the operation performed by the unit of aforementioned means 2.
Additionally, the embodiment of the present invention also provides for a kind of machine readable media, on it, storage has executable instruction, when described executable instruction is performed so that machine performs the operation that processor 32 realizes.
It will be appreciated by those skilled in the art that each embodiment disclosed above, it is possible to make various deformation and change when not necessarily departing from invention essence.Therefore, protection scope of the present invention should be defined by the appended claims.
Claims (12)
1. an electrical load Forecasting Methodology (1), including:
The first degree of depth belief network model predicting electrical load for the historical data according to electrical load is trained, thus producing for predicting that the first of future electrical energy load predicts the outcome (S1) by the historical data of electrical load；
The second degree of depth belief network model of the historical data prediction electrical load of the historical data according to electrical load and predefined leading factor of influence is trained, thus producing for predicting that the second of future electrical energy load predicts the outcome (S3) by the historical data of the historical data of electrical load and predefined leading factor of influence；With
Predict the outcome according to first and predict the outcome with second, it was predicted that future electrical energy load (S4).
2. electrical load Forecasting Methodology (1) according to claim 1, wherein train by the historical data of electrical load and predict that the step of first degree of depth belief network model (S1) of electrical load farther includes for the historical data according to electrical load: when the historical data excalation of electrical load, the first degree of depth belief network model historical data of electrical load described in the part lacked according to the electrical load data prediction before lack part completion.
3. electrical load Forecasting Methodology (1) according to claim 1, also include: be used for predefining according to the predefined part dominating factor of influence prediction disappearance before the part of disappearance completion the 3rd degree of depth belief network model of the historical data of leading factor of influence when the historical data lack part of predefined leading factor of influence with the historical data training of predefined leading factor of influence, thus completion predefines the historical data (S2) of leading factor of influence when the historical data excalation of predefined leading factor of influence.
4. the electrical load Forecasting Methodology (1) according to claim 1 or 3, wherein the coefficient value of first, second and third degree of depth belief network model adopts greedy nothing supervision feature learning to obtain.
5. the electrical load Forecasting Methodology (1) according to claim 1 or 2, wherein the historical data of electrical load is the historical data of the electrical load after normalization, and the historical data of predefined leading factor of influence is the historical data of the predefined leading factor of influence after normalization.
6. electrical load Forecasting Methodology (1) according to claim 1, wherein predefined leading factor of influence includes at least one in weather data, temperature data, humanistic environment data.
7. an electrical load prediction unit (2), including:
First training unit (21), it is configured to predict for the historical data according to electrical load the first degree of depth belief network model of electrical load with the historical data training of electrical load, thus producing for predicting that the first of future electrical energy load predicts the outcome；
Second training unit (23), it is configured to the historical data with electrical load and the predefined historical data dominating factor of influence trains the historical data of the historical data according to electrical load and predefined leading factor of influence to predict the second degree of depth belief network model of electrical load, thus generation is for predicting that the second of future electrical energy load predicts the outcome；
Predicting unit (24), being configured to predicts the outcome according to first predicts the outcome with second, it was predicted that future electrical energy load.
8. electrical load prediction unit (2) according to claim 7, wherein the first training unit (21) is configured to when the historical data excalation of electrical load, by first degree of depth belief network model part according to the electrical load data prediction disappearance before the part of disappearance the historical data of completion electrical load.
9. electrical load prediction unit (2) according to claim 7, also include: the 3rd training unit (22), it is configured to train the 3rd degree of depth belief network model of the historical data for predefining leading factor of influence according to the predefined part dominating factor of influence prediction disappearance before the part of disappearance completion when the historical data excalation of predefined leading factor of influence by the historical data of predefined leading factor of influence, thus completion predefines the historical data of leading factor of influence when the historical data excalation of predefined leading factor of influence.
10. the electrical load prediction unit (2) according to claim 7 or 9, wherein the coefficient value of first, second and third degree of depth belief network model adopts greedy nothing supervision feature learning to obtain.
11. the electrical load prediction unit (2) according to claim 7 or 9, wherein the historical data of electrical load is the historical data of the electrical load after normalization, and the historical data of predefined leading factor of influence is the historical data of the predefined leading factor of influence after normalization.
12. electrical load prediction unit (2) according to claim 7, wherein predefined leading factor of influence includes at least one in weather data, temperature data, humanistic environment data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410838273.3A CN105809264B (en) | 2014-12-29 | 2014-12-29 | Power load prediction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410838273.3A CN105809264B (en) | 2014-12-29 | 2014-12-29 | Power load prediction method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105809264A true CN105809264A (en) | 2016-07-27 |
| CN105809264B CN105809264B (en) | 2022-08-02 |
Family
ID=56979890
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410838273.3A Active CN105809264B (en) | 2014-12-29 | 2014-12-29 | Power load prediction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105809264B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529747A (en) * | 2017-01-04 | 2017-03-22 | 成都四方伟业软件股份有限公司 | Power load predicting method and system based on large data |
| KR101917729B1 (en) * | 2017-03-06 | 2018-11-12 | 온동네피엠씨 주식회사 | Average power consumption control system |
| CN109903165A (en) * | 2018-12-14 | 2019-06-18 | 阿里巴巴集团控股有限公司 | A kind of model merging method and device |
| CN110212520A (en) * | 2019-05-24 | 2019-09-06 | 国网天津市电力公司 | A kind of power predicating method based on convolutional neural networks |
| CN112381266A (en) * | 2020-10-22 | 2021-02-19 | 国网湖北省电力有限公司武汉供电公司 | System and method for predicting future power supply amount based on historical power supply and weather data |
| CN117353300A (en) * | 2023-12-04 | 2024-01-05 | 拓锐科技有限公司 | Rural power consumption demand analysis method based on big data |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2004094437A (en) * | 2002-08-30 | 2004-03-25 | Fuji Electric Holdings Co Ltd | Data prediction method and data prediction system |
| CN103514491A (en) * | 2013-10-18 | 2014-01-15 | 国网四川省电力公司自贡供电公司 | Power load forecasting method |
-
2014
- 2014-12-29 CN CN201410838273.3A patent/CN105809264B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2004094437A (en) * | 2002-08-30 | 2004-03-25 | Fuji Electric Holdings Co Ltd | Data prediction method and data prediction system |
| CN103514491A (en) * | 2013-10-18 | 2014-01-15 | 国网四川省电力公司自贡供电公司 | Power load forecasting method |
Non-Patent Citations (3)
| Title |
|---|
| 张国忠等: "人工神经网络在上海电力负荷预测中的应用", 《华东电力》 * |
| 罗贤举: "组合预测方法及其在电力系统负荷预测中的应用研究", 《万方学位论文数据库》 * |
| 肖同录等: "基于深度信念网络的短期电力负荷预测", 《电子科技》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529747A (en) * | 2017-01-04 | 2017-03-22 | 成都四方伟业软件股份有限公司 | Power load predicting method and system based on large data |
| KR101917729B1 (en) * | 2017-03-06 | 2018-11-12 | 온동네피엠씨 주식회사 | Average power consumption control system |
| CN109903165A (en) * | 2018-12-14 | 2019-06-18 | 阿里巴巴集团控股有限公司 | A kind of model merging method and device |
| WO2020119299A1 (en) * | 2018-12-14 | 2020-06-18 | 阿里巴巴集团控股有限公司 | Model merging method and device |
| TWI718690B (en) * | 2018-12-14 | 2021-02-11 | 開曼群島商創新先進技術有限公司 | Model merging method and device |
| CN110212520A (en) * | 2019-05-24 | 2019-09-06 | 国网天津市电力公司 | A kind of power predicating method based on convolutional neural networks |
| CN112381266A (en) * | 2020-10-22 | 2021-02-19 | 国网湖北省电力有限公司武汉供电公司 | System and method for predicting future power supply amount based on historical power supply and weather data |
| CN112381266B (en) * | 2020-10-22 | 2024-01-09 | 国网湖北省电力有限公司武汉供电公司 | System and method for predicting future power supply quantity based on historical power supply and weather data |
| CN117353300A (en) * | 2023-12-04 | 2024-01-05 | 拓锐科技有限公司 | Rural power consumption demand analysis method based on big data |
| CN117353300B (en) * | 2023-12-04 | 2024-02-23 | 拓锐科技有限公司 | Rural power consumption demand analysis method based on big data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105809264B (en) | 2022-08-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105809264A (en) | Electrical load predicting method and device | |
| Kamal et al. | Role of energy efficiency policies on energy consumption and CO2 emissions for building stock in Qatar | |
| Vinothkumar et al. | Distributed generation planning: A new approach based on goal programming | |
| Liu et al. | A two-stage method of quantitative flood risk analysis for reservoir real-time operation using ensemble-based hydrologic forecasts | |
| CN105678403A (en) | Region saturation load prediction method based on model family decomposition and integration technology | |
| JP6498976B2 (en) | Estimation apparatus, estimation method, and computer program | |
| WO2019056887A1 (en) | Method for performing probabilistic modeling of large-scale renewable-energy data | |
| CN106170708A (en) | Individual electricity equipment working state estimation unit and method thereof | |
| Ahmadi et al. | A lexicographic optimization and augmented ϵ-constraint technique for short-term environmental/economic combined heat and power scheduling | |
| CN105678414A (en) | Data processing method of predicting resource consumption | |
| JP2019087030A (en) | Prediction model generation device, prediction model generation method and prediction model generation program | |
| CN103268279A (en) | Compound poisson process-based software reliability prediction method | |
| CN113591368A (en) | Comprehensive energy system multi-energy load prediction method and system | |
| Mortazavi et al. | Adaptive gradient-assisted robust design optimization under interval uncertainty | |
| Pao et al. | Competition and stability analyses among emissions, energy, and economy: Application for Mexico | |
| Markoska et al. | Towards smart buildings performance testing as a service | |
| Kalu et al. | Development of matlab-based software for peak load estimation and forecasting: a case study of faculty of engineering, Imo State University Owerri, Imo state, Nigeria | |
| Wilson et al. | Use of meteorological data for improved estimation of risk in capacity adequacy studies | |
| CN113919162A (en) | Voltage sag risk early warning method based on simulation and multi-source measured data fusion | |
| CN103093094A (en) | Software failure time forecasting method based on kernel partial least squares regression algorithm | |
| CN102306353A (en) | Method and system for estimating credibility of simulation system | |
| CN103106139B (en) | Based on the software failure time forecasting methods that relevance vector regression is estimated | |
| Macdonald et al. | Constraint analysis techniques for active networks | |
| Jalori | Leveraging smart meter data through advanced analytics: Applications to building energy efficiency | |
| Liu et al. | An Enhanced Input Uncertainty Representation Method for Response Surface Models in Automotive Weight Reduction Applications |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |