CN106295877B - Method for predicting electric energy consumption of smart power grid - Google Patents
Method for predicting electric energy consumption of smart power grid Download PDFInfo
- Publication number
- CN106295877B CN106295877B CN201610648163.XA CN201610648163A CN106295877B CN 106295877 B CN106295877 B CN 106295877B CN 201610648163 A CN201610648163 A CN 201610648163A CN 106295877 B CN106295877 B CN 106295877B
- Authority
- CN
- China
- Prior art keywords
- electric energy
- energy usage
- neural network
- ith
- time point
- 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.)
- Active
Links
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Abstract
The invention discloses a method for predicting the electric energy usage of an intelligent power grid, which comprises the steps of firstly extracting relevant electric energy usage characteristic parameters from electric energy usage data according to different types of electricity users, and preprocessing the data; then, constructing a plurality of neural networks to respectively predict the electric energy usage amount of each electricity service type, wherein the parameters of each neural network are generated by a rapid algorithm; and finally, calculating the predicted total electric energy usage amount according to the electric energy usage amount of each electric service type, wherein the method of the invention fully considers the electric energy usage amount including different types of electric behaviors and the internal space-time correlation between the electric behaviors, solves the problem of the limitation of the traditional prediction method on the prediction precision, and can realize rapid and accurate prediction.
Description
Technical Field
The invention relates to the field of power grids, in particular to a technology for predicting electric energy usage of an intelligent power grid.
Background
Ensuring a balance of supply and demand for the use of electrical energy is an important issue in electrical networks. When the supplied electric energy is less than the required electric energy, the interruption of partial electric power service can be caused, and even large-scale power failure can be caused; if the supplied power is greater than the demanded power, the excess power needs to be additionally transmitted or stored using a suitably sized energy storage device, and the transmission, storage, and maintenance caused by them can significantly increase the cost of power usage. In order to solve the problem, the electric energy consumption of the power grid must be rapidly and accurately predicted, so that the balance of supply and demand of the electric energy is ensured, and the economic benefit and the social benefit of the power network are improved.
The electric energy usage in the power grid mainly includes industrial electricity, agricultural electricity, residential electricity and commercial electricity. Among them, industrial electricity and agricultural electricity have a clear plan in production process, and the use of electric energy has strong regularity. Other electricity consumption, such as residential electricity consumption and commercial electricity consumption, is not obvious in electric energy usage measurement law due to the fact that users are various, the number of users is dynamically changed, the planning performance of electricity consumption is poor and the like, and is difficult to accurately predict.
The smart grid is a modern power network. Compared with the traditional power grid, the intelligent power grid is more reliable, safe and efficient, and the intelligent power grid can obtain more detailed power utilization data by using advanced sensing, measuring, communication and other technologies, so that the power grid can be helped to predict the power utilization more accurately.
At present, the existing electric energy usage prediction method in the power grid mainly has the following problems:
1. the prediction accuracy is not high. The time sequence formed by the change of the electric energy consumption along with the time in the power grid is influenced by multiple factors such as economic development, industrial structure, climate and the like, and the occurrence of a single power utilization behavior has uncertainty, so that the power utilization behaviors have a complex nonlinear relationship; the traditional time series analysis and prediction technology is difficult to accurately reflect the complex nonlinear relation between power utilization behaviors contained in power consumption changes, so that the prediction precision is influenced.
2. The adaptability or real-time performance is poor. Most of the existing electric energy usage amount prediction methods adopt mathematical models to model prediction, the modeling process is complex, and related parameters cannot be adjusted independently when some factors influencing the electric energy usage amount change; some prediction methods have good adaptivity, but have high training/calculation cost and poor real-time performance.
At present, common methods at home and abroad for predicting the electric energy consumption mainly comprise a prediction method based on a time sequence and a prediction method based on regression analysis.
1. The prediction method based on the time series comprises the following steps: by finding some rule of historical data over time, a suitable mathematical model is built to make the prediction. Examples of the prediction method include a prediction method based on an Auto-Regressive model (AR model for short), a prediction method based on a Moving-Average model (MA model), and a prediction method based on an Auto-Regressive Moving-Average model (ARMA model).
2. Prediction method based on regression analysis: and (4) establishing a set of model equations of the electric energy consumption and other related variables by carrying out regression analysis on the historical data to predict. Such as Support Vector Machine (SVM) based prediction methods, K-Nearest Neighbor (K-NN) based prediction methods, etc.
The above methods predict the electric energy usage only through the relation between the historical value and the predicted value of the electric energy usage and the historical change rule, and do not consider the electric energy usage and the internal relation between the electric energy usage and the predicted value (for example, the relation between the electric energy usage of an electric lamp and the electric energy usage of a computer in a company), thereby limiting the prediction accuracy.
Disclosure of Invention
The invention provides a method for predicting the electric energy usage of a smart power grid to solve the technical problems, and the method comprises the steps of firstly extracting relevant electric energy usage characteristic parameters from electric energy usage data according to different types of electricity users, and preprocessing the data; then, constructing a plurality of neural networks to respectively predict the electric energy usage amount of each electricity service type, wherein the parameters of each neural network are generated by a rapid algorithm; and finally, calculating the predicted total electric energy usage amount according to the electric energy usage amount of each electric service type.
The technical scheme adopted by the invention is as follows: a method for predicting electric energy usage of a smart grid comprises the following steps:
s1, determining characteristic parameters of the electric energy usage amount according to the type of the electric service; the electric energy usage characteristic parameter represents the electric energy usage of the application electric service type;
s2, determining the time scale of the characteristic parameter of the electric energy usage, and constructing a multi-time sequence describing the electric energy usage
Wherein the content of the first and second substances,the original electric energy usage of the characteristic parameter of the ith electric energy usage at the tth time point, wherein i is 1,2,3 …, M, M represents the total number of the characteristic parameters of the electric energy usage, and t represents a time point serial number;
s3, obtaining the length of the training time window according to the time scale determined in the step S2; and in the training time windowCarrying out normalization processing;
s4, obtaining the length of the historical time window according to the time scale determined in the step S2; determining an input vector for a time point to be predicted
Wherein, Xi,j(t) represents a required input vector when the ith characteristic parameter of the electric energy consumption is predicted at the tth time pointT represents the historical time window length, said T being the number of unit time scales within a period time length;
s5, initializing M neural networks, and determining the number of hidden layer numbers and hidden layer nodes of each neural network; obtaining the electricity consumption service type electricity consumption amount corresponding to the ith electricity consumption characteristic parameter through predicting the ith' neural network; the predicted value of the ith' neural network corresponds to the value of the electric energy usage of the electric service type corresponding to the characteristic parameter of the ith electric energy usage;
s6, judging whether each neural network parameter group needs to be calculated; if yes, go to step S7, otherwise go to step S9;
s7, training each neural network for K times to obtain K sets of parameter sets corresponding to each neural network;
s8, judging whether the neural network to be trained completes all training, if yes, turning to step S9; otherwise go to step S7;
s9, predicting the neural network, specifically: calculating to obtain a predicted average value of the electric energy usage of the electric energy service type corresponding to each electric energy usage characteristic parameter of the current time point at the next time point of the user;
s10, performing inverse normalization processing on the predicted average value of the electric energy usage of the electricity service type corresponding to each electric energy usage characteristic parameter obtained in the step S9 to obtain a predicted value of the original electric energy usage;
s11, judging whether the next time point needs to be predicted, if yes, turning to the step S2; otherwise, ending.
Further, in the step S3, the multi-time sequence of the amount of electric energy used is specifically defined according to the following equationAnd (3) carrying out normalization treatment:
wherein S isi(t) represents the normalized electric energy usage of the ith electric energy usage characteristic parameter corresponding to the tth time point,and representing the original electric energy usage corresponding to the ith electric energy usage characteristic parameter at the jth moment in the first training time window.
Further, the step S5 of determining the number of hidden layer nodes in each neural network through the sample data test specifically includes:
a1, setting the number L of hidden layer nodes of the ith' neural networki′=1;
A2, inputting D groups of training samples for training, respectively calculating prediction result errors, and calculating the average value of the prediction errors of the D groups of training samples;
a3, when Li′=Li′When the sum is +1, calculating the average value of the prediction errors of the training samples in the group D;
a4, comparing the number L of the current hidden layer nodesi′And the average value of the prediction errors of the D groups of training samples and the number of hidden layer nodes are Li′When the value is-1, whether the reduction ratio of the average value of the prediction errors of the obtained D groups of training samples is smaller than a threshold value, and if so, determining the L at the momenti′The number of hidden layer nodes in the ith' neural network; otherwise, return to step A3.
Further, the step S6 specifically includes the following sub-steps:
s61, defining a parameter evaluation time window which is formed by the current time point and the previous N0-1 time point composition;
s62, defining an average error of power consumption prediction of the power consumption service type corresponding to the ith power consumption characteristic parameter in the parameter evaluation time window;
s63, if each neural network has not been initialized or the number of the predicted time points is less than N0Then go to step S7; otherwise go to step S64;
s64, if the average error of the electric energy usage of the ith type of electric energy usage characteristic parameter predicted by the neural network to the electricity service type in the parameter evaluation time window is larger than a given threshold value, turning to the step S7, otherwise, turning to the step S9.
Further, the step S7 specifically includes the following steps:
s71 input weight matrix for ith' neural networkAnd hidden layer node offset vectorCarrying out assignment;
wherein the content of the first and second substances,representing the input weights of all input nodes in the ith' neural network to the jth hidden layer node;
wherein, bi′,j″Representing the bias value corresponding to the jth hidden node in the ith' neural network;
s72, calculating the output matrix H of the neural network0;
Wherein the content of the first and second substances,representing that the ith' neural network and the jth hidden node correspond to the sampleT is 1,2,3 …, N represents the number of training data;is the input weight of the ith' neural network;
Wherein, the neural network output matrix H is represented0The generalized inverse of (1);Yi′and (t) is the real value of the predicted electric energy usage amount of the ith' neural network at the tth time point, wherein t is 1,2,3 …, and N represent the number of training data.
Further, the step S9 specifically includes the following sub-steps:
s91, calculating an output matrix H of each neural network at the current time point;
s92, calculating the current time point to each electricity of the next time point of the userPredicted value { P) of electric energy usage of electricity service type corresponding to characteristic parameter of energy usagei,k(t)};
Wherein, Pi′,k(t) represents the predicted outcome of the kth set of parameters for the ith' neural network, K being 1,2, …, K;
s93, calculating the predicted average value of the electric energy consumption of the electric energy service type corresponding to each electric energy consumption characteristic parameter of the next time point of the user at the current time point
Wherein the content of the first and second substances,the results of averaging the predicted results of the K sets of parameters representing the ith' neural network.
Further, the predicted value of the original power consumption in step S10 is specifically:
wherein max {. indicates taking the maximum value, and min {. indicates taking the minimum value.
The invention has the beneficial effects that: the invention relates to a method for predicting the electric energy usage of an intelligent power grid, which comprises the steps of firstly extracting relevant electric energy usage characteristic parameters from electric energy usage data according to different types of electricity users, and preprocessing the data; then, a plurality of neural networks are constructed to respectively predict the electric energy usage amount of each electricity service type, and the parameters of each neural network are generated by a rapid algorithm provided by the application; finally, calculating the predicted total electric energy usage amount according to the electric energy usage amount of each electric service type; the method of the invention fully considers the different types of power consumption behaviors contained in the electric energy consumption and the internal space-time correlation between the electric energy consumption and the different types of power consumption behaviors, solves the problem of the limitation of the traditional prediction method on the prediction precision, and can realize rapid and accurate prediction.
Drawings
Fig. 1 is a block diagram of a method for predicting the electric energy usage of a smart grid according to the present invention.
Fig. 2 is a diagram of a neural network structure for predicting the power consumption of the power service type corresponding to the ith characteristic parameter of the power consumption according to the present invention.
Fig. 3 is a flow chart of neural network training for predicting the power consumption of the power consumption service type corresponding to the ith characteristic parameter of power consumption according to the present invention.
Detailed Description
In order to facilitate the understanding of the technical contents of the present invention by those skilled in the art, the present invention will be further explained with reference to the accompanying drawings.
Fig. 1 is a block diagram of a method for predicting the electric energy usage of an intelligent power grid according to the present invention, where the electric energy usage of the power grid is the electric energy used by users in the power grid. The technical scheme of the invention is as follows: a method for predicting electric energy usage of a smart grid comprises the following steps:
s1, determining characteristic parameters of the electric energy usage amount according to the type of the electric service; the power usage characteristic parameter represents power usage for an application electrical service type.
The power usage data may be obtained from a power usage monitoring device, such as a smart meter, in the actual smart grid. These power usage data include: electricity usage, current, voltage, frequency, etc. that the user varies over time. Different electricity consumption service types can show unique characteristics (such as waveforms) on parameters such as electricity consumption, current, voltage, frequency and the like, and through signal analysis and processing, the prior art can extract the electricity consumption of different electricity consumption service types along with time change by analyzing the electricity consumption, the current, the voltage and the frequency of a user along with the time change.
According to different electricity service types, the related electricity usage characteristic parameters describing the electricity usage of the user can be extracted from the electricity usage data. In the application, the electric energy usage amount of a plurality of electricity utilization service types is used as the characteristic parameter of the electric energy usage amount. Specifically, the method comprises the following steps: different power utilization service types are described, and different power utilization quantity characteristic parameters are required to be used. For example, the residential electricity consumers can use 11 electric energy consumption characteristic parameters of lighting electricity consumption, kitchen socket electricity consumption, dish washer electricity consumption, microwave oven electricity consumption, washing machine electricity consumption, oven electricity consumption, refrigerator electricity consumption, bathroom electrical equipment electricity consumption, electric heater electricity consumption, air conditioner electricity consumption and other electricity consumption which change along with time to describe the electric energy consumption. In the application, M electric energy usage characteristic parameters are adopted to describe the electric energy usage of a user, M can take different values according to different electricity utilization service types, and the prediction of the electric energy usage of the user can be realized by predicting the electric energy usage of different electricity utilization service types of the user, namely M characteristic parameters describing the electric energy usage of the user are predicted.
S2, determining the time scale of the characteristic parameter of the electric energy usage, and constructing a multi-time sequence describing the electric energy usage
Wherein the content of the first and second substances,the characteristic parameter of the ith power usage represents the original power usage at the tth time point, i is 1,2,3 …, M represents the total number of the characteristic parameters of the power usage, and t represents the time point serial number.
The characteristic parameter of the amount of energy usage can be selected according to different prediction tasks, such as prediction by the hour or prediction by the day, for example, according to a corresponding time scale U, such as 15 minutes, 1 hour or 1 day. After the time scale U is selected, the electric energy usage of each electric service type in each unit time scale is the electric energy usage of the electric service type corresponding to the characteristic parameter of the electric energy usage, and is recorded as the electric energy usage of the electric service type corresponding to the characteristic parameter of the electric energy usageNamely the original electric energy consumption of the electricity service type corresponding to the characteristic parameter of the ith electric energy consumption at the tth time point. At each time point, the electric energy usage of the user is represented by M electric energy usage characteristic parameters, and the change of all the electric energy usage characteristic parameters along with the time forms a multi-time sequence for describing the electric energy usage, namelyWhere, i is 1,2,3 …, and M, t is a time point number.
S3, obtaining the length of the training time window according to the time scale determined in the step S2; and in the training time windowCarrying out normalization processing;
according to the time scale U selected in step S2, the number N of required training data, i.e., the length of the training time window, is determined. The number of training data N may be 50-100 times the number of unit time scales within a period of time, such as a day, week or month. For U of 15 minutes, N is (60/15) × 24 × 50 ═ 4800; for U of 1 hour, N is 24 × 100 ═ 2400; for 1 day U, N is 30 × 100 — 3000. The data at the current time point and N-1 time points before the current time point form data in a training time window.
In the training time window, the following formula can be used for describing the multi-time sequence of the electric energy usageAnd (6) carrying out normalization.
The multi-time sequence of the electric energy usage is normalized, so that the influence of the electric energy usage characteristic parameter with a large value range on the final prediction result can be avoided when the electric energy usage characteristic parameter with a small value range is covered. Hereinafter, for convenience of description, S in this applicationi(t) watchThe normalized electric energy usage amount of the ith electric energy usage characteristic parameter corresponding to the tth time point is calledThe original electric energy usage amount of the application electric service type corresponding to the ith electric energy usage amount characteristic parameter at the t-th time point,and representing the original electric energy usage corresponding to the ith electric energy usage characteristic parameter at the jth moment in the first training time window.
S4, obtaining the length of the historical time window according to the time scale determined in the step S2; determining an input vector for a time point to be predicted
Wherein, Xi,j(t) represents a required input vector when the ith characteristic parameter of the electric energy consumption is predicted at the tth time pointT represents the historical time window length, said T being the number of time scales per unit time within a period time length.
The historical time window length T is determined from the time scale U selected in step S2. The historical time window length T is within a period time length. Such as a day, a week or a month. The number of unit time scales. For U of 15 minutes, T is (60/15) × 24 ═ 96; t is 24 for U of 1 hour; for a1 day U, T is 30. The data for the current point in time and its previous N-1 points in time constitute the data in the historical time window.
The M electric energy service types described by the M electric energy use quantity characteristic parameters have correlation (corresponding to certain activities of users) at the same time point, and the value of each electric energy use quantity characteristic parameter at the current point has time correlation with the value of the electric energy use quantity characteristic parameter in the historical time window. Therefore, for the ith characteristic parameter of electric energy usage amountAnd predicting a time point value, namely predicting the electric energy usage of the ith electricity service type at the next time point, wherein the values of all the electric energy usage characteristic parameters at the current time point and all the values of the electric energy usage characteristic parameters in the historical time window can be used as input vectors required by prediction. In the application, the predicted value of the ith electric energy use quantity characteristic parameter at the next time point (t +1 time point) is recorded as Pi(t) (i ═ 1,2,3 …, M), the corresponding input vector is:
wherein Z isi(t)={Si(t),Si(t-1),...,Si(t-T+1)},Pi(t) is the electric energy usage S of the electric energy service type corresponding to the ith electric energy usage characteristic parameter at the t +1 th time pointiAnd (t +1) prediction. Y isi(t) is the electric energy usage S of the electricity service type corresponding to the ith electric energy usage characteristic parameter at the t +1 th time pointiThe true value of (t +1), i.e., the target value.
Wherein, Xi,j(t) represents a required input vector when the ith characteristic parameter of the electric energy consumption is predicted at the tth time pointThe jth value of (a). Due to Si(t) has been normalized so that it is not necessary to re-align X at this timei,j(t) normalization is performed. In the subsequent prediction step using the neural network, the application will directly useAs an input vector for the time point to be predicted.
S5, initializing M neural networks, and predicting the ith' neural network to obtain the electricity consumption service type electricity consumption corresponding to the ith electricity consumption characteristic parameter; and determining the number of hidden layer numbers and the number of hidden layer nodes of each neural network.
As shown in fig. 2, a diagram of a neural network structure for predicting the power consumption of the electricity service type corresponding to the ith characteristic parameter of the power consumption,the vector is an M + T-1 dimensional row vector and represents the input weight of the jth hidden layer node in the ith' neural network; bi′,j″Is a value representing the bias corresponding to the jth hidden node in the ith' neural network βi′,j″Representing the external weight value between the node connecting the jth hidden layer in the ith' neural network and the network output;indicating that the jth hidden node corresponds to an input vectorThe output of (i), G (-) is the activation function.
The method and the device use M neural networks to predict the electric energy usage of the M types of electricity utilization service types in parallel. The predicted object of the ith' neural network is the electric energy usage of the electricity service type corresponding to the ith electric energy usage characteristic parameter, and the corresponding input vector (at the t-th time point) isThe predicted value is (at the t-th time point) Pi(t) the target value is (at the t-th time point) Yi(t)。
For each neural network, a single hidden layer network is selected, namely the number of hidden layers is 1. The single hidden layer neural network has the advantages of simple structure, high training speed, difficulty in overfitting and the like.
Number of hidden nodes in each neural networkLi′The method is determined by sample data test, and comprises the following specific steps:
1. firstly, the number L of hidden layer nodes of a target neural network is seti′=1。
2. 50 groups of training samples are input for training, prediction result errors are respectively calculated, and the average value of the prediction errors of the 50 groups of training samples is calculated.
3. Let Li′=Li′+1, the average of the prediction errors of 50 training samples is calculated again when the number of hidden nodes is used.
If L isi′When increasing, the ratio of the reduction of the prediction error average value calculated by using the training sample to the last value is smaller than the threshold, the default value is 0.1 percent, and then the L at the moment is determinedi′The number of hidden layer nodes in the target neural network is shown; otherwise, return to step 3.
The reduction ratio is specifically: make the current hidden layer node number Li′The average value of the prediction errors of the D groups of training samples is Mean1, L i′1 corresponds to an average Mean of 2, the reduction ratio is: i Mean1-Mean 2I/Mean 1.
S6, judging whether each neural network parameter group needs to be calculated; if so, go to step S7, otherwise go to step S9.
Fig. 3 shows a neural network training flowchart for predicting the power consumption of the power consumption service type corresponding to the ith power consumption characteristic parameter, which includes the following specific steps:
first, a parameter evaluation window is defined to help determine whether each neural network parameter set needs to be recalculated. The parameter evaluation time window is composed of the current time point and the previous N 01 time point. In the present application, the parameter estimates the time window length N0Has a default value of 5.
Then, defining the average error of the electricity consumption prediction of the electricity service type corresponding to the ith electricity consumption characteristic parameter in the parameter evaluation time windowComprises the following steps:
wherein, t0As the current time of day, the time of day,is the average of predicted power usage at time point t-1, i.e., for time point S at time point t-1i(t) average prediction value of multiple predictions.
Furthermore, if each neural network is not initialized or the number of time points predicted by the current time point is less than N0And all the neural networks enter a parameter set training step, and the parameter set of each neural network is calculated. Otherwise, whether the parameter group of each neural network needs to be recalculated is sequentially judged.
Finally, if the average error of the electricity consumption service type electricity consumption corresponding to the ith electricity consumption characteristic parameter predicted by the neural network to be judged in the parameter evaluation time windowGreater than a given threshold, depending on the prediction accuracy requirements, e.g. in the present applicationFor training 1.5 times of the Mean value Mean of the sample prediction error, according to different precision requirements, 2 times, 3 times, 1.2 times and the like can be taken, and then the neural network needs to recalculate the parameter group. Otherwise, the neural network does not need to recalculate the parameter set, and when the electricity consumption service type electricity consumption corresponding to the ith electricity consumption characteristic parameter corresponding to the neural network is predicted, the current parameter set of the neural network can be directly used for prediction.
S7, training each neural network for K times to obtain K sets of parameter sets corresponding to each neural network; in this application, it is specifically that the training frequency is set to K50: training according to historical data to obtain a parameter set of the neural network;
the method specifically comprises the following steps:
s71 input weight matrix for ith' neural networkAnd hidden layer node offset vectorCarrying out assignment;
wherein the content of the first and second substances,representing the input weights of all input nodes in the ith' neural network to the jth hidden layer node,
wherein, bi′,jRepresenting a bias value corresponding to a jth hidden node in an ith' neural network;
for the neural network for predicting the electric energy consumption of the ith' electricity service type, parameters of the neural network are requiredAndand carrying out assignment. Wherein the content of the first and second substances,for randomly generated Li′(M + T-1) size matrix,the input weights for all input nodes in the ith' neural network to the jth hidden node are one-dimensional vectors of length M + T-1.For randomly generating Li′Vector of size x 1. Wherein, bi′,j″And indicating the bias value corresponding to the jth hidden node in the ith' neural network. Neural network input weightsAnd hidden layer node bias bi′,j″(j″=1,2,3...,Li′) Randomly generated by the system.
The above random numbers may be generated by a probability density function that is uniformly distributed over [0,1 ].
S72, calculating the output matrix H of the neural network0;
Activation function of neural network we choose Sigmoid function.
For input vectorHaving Li′The output expression of the neural network of each hidden layer node is as follows:
in the formula, Yi′(t) is the true value of the predicted electric energy consumption of the ith' neural network at the t time point, namely Si(t),As input weights to the ith' neural network, bi′,jFor the bias of the ith 'neural network jth' hidden node, βi′,jRepresents the external weight connecting the jth hidden node and the neural network output in the ith' neural network,denotes the ith' neural network jthEach hidden node corresponds to a sampleAnd (4) outputting hidden layer nodes. For a hidden layer node of the additive type,the expression of (a) is:
wherein the content of the first and second substances,representing weight vectorsAnd an input vectorThe inner product of (d). G (. cndot.) is defined herein as R->R is a name G, and the definition domain and the value domain are real mapping; the function G (-) is defined as aboveNamely, it is"·" represents all the arguments of the function; ' R->In R', the former R represents the definition domain as real number, and the latter R represents the value domain as real number->To map the symbols.
Using input data of N (N is the number of training data defined in step 3) continuous time points The output matrix H of the ith' neural network can be calculated0:
The input data of the number of training data defined in step S3 and the number of continuous time points are still used as N in step S72Calculating the output external weight of the ith' neural networkThe calculation formula of (a) is as follows:
wherein the content of the first and second substances,representing neural network output matrix H0The generalized inverse of (1) is, outputting a matrix H for a neural network0The transpose of (a) is performed,is a matrixThe inverse matrix of (d);Yi′(t) predicted electric energy consumption of ith' neural network at tth time pointThe true value, t ═ 1,2,3 …, and N indicates the number of training data.Is a Li′A column vector of x 1.
S8, judging whether the neural network to be trained completes all training, if yes, turning to step S9; otherwise go to step S7. Sequentially judging each neural network to be trained, and if the neural network to be judged does not finish independent training for 50 times, turning to the step S7 to continue training; and when all the neural networks to be trained have finished 50 times of independent training, completing the neural network parameter group calculation required for predicting the electric energy usage amount at the current time point.
A set of parameters of the ith' neural network can be obtained through the step S7Andfor each neural network to be trained, the same input data is adopted in the application50 independent exercises were performed.
And sequentially judging each neural network to be trained. If the neural network does not complete 50 independent trainings, go to step S7 to continue training. When all the neural networks to be trained have completed independent training 50 times, the calculation of the neural network parameter group required for predicting the electric energy usage amount at the current time point is completed, and the process goes to step S9.
S9, predicting the neural network, specifically: calculating to obtain a predicted average value of the electric energy usage of the electric energy service type corresponding to each electric energy usage characteristic parameter of the current time point at the next time point of the user; the method specifically comprises the following steps:
s91, calculating an output matrix H of each neural network at the current time point;
for the ith' godEach set of parameters via the networkAndthe output matrix can be calculated by the following formula:
s92, calculating a predicted value { P) of the electric energy usage of the electric energy service type corresponding to each electric energy usage characteristic parameter of the user at the next time point at the current time pointi,k(t)};
Kth set of parameters for ith' neural networkAndall can obtain a predicted value P of the electricity consumption service type electric energy consumption corresponding to the i-th type electric energy consumption characteristic parameteri,k(t) obtaining P by calculating the prediction result of the kth group of parameters of the ith' neural networki,k(t), the calculation formula is as follows:
s93, calculating the predicted average value of the electric energy consumption of the electric energy service type corresponding to each electric energy consumption characteristic parameter of the next time point of the user at the current time pointCalculating the average value of the predicted results of the i' th neural network with the K-50 group of parameters to obtain the corresponding resultThereby obtaining
S10, performing inverse normalization processing on the predicted average value of the electric energy usage of the electricity service type corresponding to each electric energy usage characteristic parameter obtained in the step S9 to obtain a predicted value of the original electric energy usage;
the predicted average value of the electric energy usage of each electric service type is subjected to inverse normalization to obtain the predicted value of the original electric energy usage of each electric service type
Predicted value of current time point to original electric energy usage amount of each electric service type of user at next time pointThe following formula can be used for calculation:
s11, judging whether the next time point needs to be predicted, if yes, turning to the step S2; otherwise, ending.
It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Various modifications and alterations to this invention will become apparent to those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.
Claims (7)
1. A method for predicting electric energy usage of a smart grid is characterized by comprising the following steps:
s1, determining characteristic parameters of the electric energy usage amount according to the type of the electric service; the electric energy usage characteristic parameter represents the electric energy usage of the application electric service type;
s2, determining the time scale of the characteristic parameter of the electric energy usage, and constructing a multi-time sequence describing the electric energy usage
Wherein the content of the first and second substances,the original electric energy usage of the characteristic parameter of the ith electric energy usage at the tth time point, wherein i is 1,2,3 …, M, M represents the total number of the characteristic parameters of the electric energy usage, and t represents a time point serial number;
s3, obtaining the length of the training time window according to the time scale determined in the step S2; and in the training time windowCarrying out normalization processing;
s4, obtaining the length of the historical time window according to the time scale determined in the step S2; determining an input vector for a time point to be predicted
Wherein the content of the first and second substances,is a T + M-1 dimensional vector, Zi(t)={Si(t),Si(t-1),...,Si(t-T+1)},Si(t) represents the electric energy usage amount of the electricity service type corresponding to the ith electric energy usage amount characteristic parameter at the t-th time point;
s5, initializing M neural networks, and determining the number of hidden layer numbers and hidden layer nodes of each neural network; obtaining the electricity consumption service type electricity consumption amount corresponding to the ith electricity consumption characteristic parameter through predicting the ith' neural network; the predicted value of the ith' neural network corresponds to the value of the electric energy usage of the electric service type corresponding to the characteristic parameter of the ith electric energy usage;
s6, judging whether each neural network parameter group needs to be calculated; if yes, go to step S7, otherwise go to step S9;
s7, training each neural network for K times to obtain K sets of parameter sets corresponding to each neural network;
s8, judging whether the neural network to be trained completes all training, if yes, turning to step S9; otherwise go to step S7;
s9, predicting the neural network, specifically: calculating to obtain a predicted average value of the electric energy usage of the electric energy service type corresponding to each electric energy usage characteristic parameter of the current time point at the next time point of the user;
s10, performing inverse normalization processing on the predicted average value of the electric energy usage of the electricity service type corresponding to each electric energy usage characteristic parameter obtained in the step S9 to obtain a predicted value of the original electric energy usage;
s11, judging whether the next time point needs to be predicted, if yes, turning to the step S2; otherwise, ending.
2. The method for predicting the electric energy usage of the smart grid according to claim 1, wherein the step S3 is a multi-time sequence of the electric energy usage according to the following formulaAnd (3) carrying out normalization treatment:
wherein S isi(t) normalized electric energy consumption at the t-th time point representing the ith electric energy consumption characteristic parameterThe dosage of the composition is controlled by the dosage,and representing the original electric energy usage corresponding to the ith electric energy usage characteristic parameter at the jth moment in the training time window.
3. The method for predicting the electric energy consumption of the smart grid according to claim 1, wherein the step S5 of determining the number of hidden nodes in each neural network through sample data testing specifically comprises:
a1, setting the number L of hidden layer nodes of the ith' neural networki′=1;
A2, inputting D groups of training samples for training, respectively calculating prediction result errors, and calculating the average value of the prediction errors of the D groups of training samples;
a3, when Li′=Li′When the sum is +1, calculating the average value of the prediction errors of the training samples in the group D;
a4, comparing the number L of the current hidden layer nodesi′And the average value of the prediction errors of the D groups of training samples and the number of hidden layer nodes are Li′When the value is-1, whether the reduction ratio of the average value of the prediction errors of the obtained D groups of training samples is smaller than a threshold value, and if so, determining the L at the momenti′The number of hidden layer nodes in the ith' neural network; otherwise, return to step A3.
4. The method for predicting the electric energy usage of the smart grid according to claim 1, wherein the step S6 specifically includes the following sub-steps:
s61, defining a parameter evaluation time window which is formed by the current time point and the previous N0-1 time point composition;
s62, defining an average error of power consumption prediction of the power consumption service type corresponding to the ith power consumption characteristic parameter in the parameter evaluation time window;
s63, if each neural network has not been initialized or the number of the predicted time points is less than N0Then to holdStep S7 is performed; otherwise go to step S64;
s64, if the average error of the electric energy usage of the ith type of electric energy usage characteristic parameter predicted by the neural network to be judged corresponding to the electric service type in the parameter evaluation time window is larger than a given threshold value, turning to the step S7, otherwise, turning to the step S9.
5. The method for predicting the electric energy usage of the smart grid according to claim 1, wherein the step S7 specifically includes the following steps:
s71 input weight matrix for ith' neural networkAnd hidden layer node offset vectorCarrying out assignment;
wherein the content of the first and second substances,representing the input weights of all input nodes in the ith' neural network to the jth hidden layer node;
wherein, bi′,j″Representing the bias value corresponding to the jth hidden node in the ith' neural network;
s72, calculating the output matrix H of the neural network0;
Wherein the content of the first and second substances,representing that the ith' neural network and the jth hidden node correspond to the sampleT is 1,2,3 …, N represents the number of training data;is the input weight of the ith' neural network;
Wherein the content of the first and second substances,representing neural network output matrix H0The generalized inverse of (1);Yi′and (t) is the real value of the predicted electric energy usage amount of the ith' neural network at the tth time point, wherein t is 1,2,3 …, and N represent the number of training data.
6. The method for predicting the electric energy usage of the smart grid according to claim 1, wherein the step S9 specifically includes the following sub-steps:
s91, calculating an output matrix H of each neural network at the current time point;
s92, calculating a predicted value { P) of the electric energy usage of the electric energy service type corresponding to each electric energy usage characteristic parameter of the user at the next time point at the current time pointi,k(t)};
Wherein, Pi′,k(t) represents the predicted outcome of the kth set of parameters for the ith' neural network, K being 1,2, …, K,is composed ofThe elements (A) and (B) in (B),is a Li′A column vector of x 1;
s93, calculating the predicted average value of the electric energy consumption of the electric energy service type corresponding to each electric energy consumption characteristic parameter of the next time point of the user at the current time point
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610648163.XA CN106295877B (en) | 2016-08-09 | 2016-08-09 | Method for predicting electric energy consumption of smart power grid |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610648163.XA CN106295877B (en) | 2016-08-09 | 2016-08-09 | Method for predicting electric energy consumption of smart power grid |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106295877A CN106295877A (en) | 2017-01-04 |
CN106295877B true CN106295877B (en) | 2020-04-14 |
Family
ID=57667273
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610648163.XA Active CN106295877B (en) | 2016-08-09 | 2016-08-09 | Method for predicting electric energy consumption of smart power grid |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106295877B (en) |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109102103A (en) * | 2018-06-26 | 2018-12-28 | 上海鲁班软件股份有限公司 | A kind of multi-class energy consumption prediction technique based on Recognition with Recurrent Neural Network |
CN110837934B (en) * | 2019-11-11 | 2023-04-07 | 四川大学 | Smart grid short-term residential load prediction method based on deep learning |
CN111523807B (en) * | 2020-04-24 | 2023-04-28 | 广西电网有限责任公司崇左供电局 | Electric energy substitution potential analysis method based on time sequence and neural network |
CN111933302B (en) * | 2020-10-09 | 2021-01-05 | 平安科技(深圳)有限公司 | Medicine recommendation method and device, computer equipment and storage medium |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102682349A (en) * | 2012-05-14 | 2012-09-19 | 云南电力试验研究院(集团)有限公司电力研究院 | Electricity consumption intelligent prediction system and method |
CN104573857A (en) * | 2014-12-26 | 2015-04-29 | 国家电网公司 | Power grid load rate prediction method based on intelligent algorithm optimization and combination |
CN105117810A (en) * | 2015-09-24 | 2015-12-02 | 国网福建省电力有限公司泉州供电公司 | Residential electricity consumption mid-term load prediction method under multistep electricity price mechanism |
CN105373865A (en) * | 2015-12-11 | 2016-03-02 | 国网四川省电力公司经济技术研究院 | Industrial structure based electricity consumption demand prediction method |
-
2016
- 2016-08-09 CN CN201610648163.XA patent/CN106295877B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN106295877A (en) | 2017-01-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Wang et al. | A seasonal GM (1, 1) model for forecasting the electricity consumption of the primary economic sectors | |
Khafaga et al. | Forecasting Energy Consumption Using a Novel Hybrid Dipper Throated Optimization and Stochastic Fractal Search Algorithm | |
Li et al. | Building's electricity consumption prediction using optimized artificial neural networks and principal component analysis | |
Javaid et al. | Towards buildings energy management: using seasonal schedules under time of use pricing tariff via deep neuro-fuzzy optimizer | |
CN109145949A (en) | Non-intrusive electrical load monitoring and decomposition method and system based on integrated study | |
Wahid et al. | A prediction approach for demand analysis of energy consumption using k-nearest neighbor in residential buildings | |
CN106295877B (en) | Method for predicting electric energy consumption of smart power grid | |
CN107506868B (en) | Method and device for predicting short-time power load | |
CN111693931A (en) | Intelligent electric energy meter error remote calculation method and device and computer equipment | |
Dong et al. | Applying the ensemble artificial neural network-based hybrid data-driven model to daily total load forecasting | |
Zhang et al. | A novel closed-loop clustering algorithm for hierarchical load forecasting | |
CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
Akpinar et al. | Forecasting natural gas consumption with hybrid neural networks—Artificial bee colony | |
Kong et al. | A remote estimation method of smart meter errors based on neural network filter and generalized damping recursive least square | |
CN112365056A (en) | Electrical load joint prediction method and device, terminal and storage medium | |
Yan et al. | Short-term load forecasting of smart grid based on load spatial-temporal distribution | |
CN115545333A (en) | Method for predicting load curve of multi-load daily-type power distribution network | |
CN114971090A (en) | Electric heating load prediction method, system, equipment and medium | |
US20230324860A1 (en) | Estimating Energy Consumption for a Building Using Dilated Convolutional Neural Networks | |
Tao et al. | Reserve evaluation and energy management of micro-grids in joint electricity markets based on non-intrusive load monitoring | |
CN111337833B (en) | Lithium battery capacity integrated prediction method based on dynamic time-varying weight | |
Souhe et al. | Forecasting of electrical energy consumption of households in a smart grid | |
CN109146553A (en) | Spot Price forecasting system and its method based on multi-density cluster and multicore SVM | |
Al-Hawani et al. | Short-term forecasting of electricity consumption using gaussian processes | |
Xu et al. | Federated learning for interpretable short-term residential load forecasting in edge computing network |
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 |