CN110852514A - Energy prediction method for energy-available sensor node based on BP neural network - Google Patents

Energy prediction method for energy-available sensor node based on BP neural network Download PDF

Info

Publication number
CN110852514A
CN110852514A CN201911116683.6A CN201911116683A CN110852514A CN 110852514 A CN110852514 A CN 110852514A CN 201911116683 A CN201911116683 A CN 201911116683A CN 110852514 A CN110852514 A CN 110852514A
Authority
CN
China
Prior art keywords
neural network
solar energy
output
training
data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201911116683.6A
Other languages
Chinese (zh)
Inventor
陶洋
李正阳
周远林
杨柳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chongqing University of Post and Telecommunications
Original Assignee
Chongqing University of Post and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Chongqing University of Post and Telecommunications filed Critical Chongqing University of Post and Telecommunications
Priority to CN201911116683.6A priority Critical patent/CN110852514A/en
Publication of CN110852514A publication Critical patent/CN110852514A/en
Pending legal-status Critical Current

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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Abstract

The invention provides an energy prediction method of an energy-available sensor node based on a BP neural network, and belongs to the field of wireless sensor network communication. The method comprises the following steps: step 1, selecting historical solar energy collection power data of a target area to perform classified sampling to obtain two groups of training sample sets; step 2, respectively preprocessing the two groups of data; step 3, respectively training two BP neural networks by using the two groups of training samples obtained in the step 2; and 4, inputting historical solar energy collection power data to obtain a predicted value. The method and the device predict the solar energy collecting power at the next moment based on the historical solar energy collecting power, and avoid the defect of reduced precision caused by neglecting some factors by a model when meteorological factors are selected as input vectors. According to different weather conditions, the invention trains two solar energy collection power prediction models, and ensures higher prediction precision in sunny days or rainy days.

Description

Energy prediction method for energy-available sensor node based on BP neural network
Technical Field
The invention belongs to the field of wireless sensor network communication, and relates to a method for predicting energy of an energy-available sensor node based on a BP neural network.
Background
An Energy Harvesting Wireless Sensor Network (EH-WSN) is a new and rapidly developing research field. The sensor nodes collect energy from the surrounding environment, the service life of the wireless sensor network is prolonged, and the defect that the nodes cannot work due to energy consumption is overcome. However, new technology introduction also brings new challenges. The environmental energy has the characteristics of time randomness, periodicity and uneven regional distribution, so that the energy collecting capacity of the sensor nodes has asymmetry and instability. Thus, predicting the collected energy ahead of time may help the node to take a better strategy. For example, in the EH-WSN, if the collected energy is predicted to be less in the future period, the node reduces the energy consumption of the node; and on the contrary, more data acquisition, processing and transmission tasks can be assumed.
Solar energy is characterized by high power intensity and periodicity and is therefore more predictable than other forms of environmental energy. The solar energy collection technology has made a long-standing progress nowadays, and the energy collection can be carried out through solar cell panel to the sensor node, uses the lithium cell as energy memory, is controlled the energy consumption by microcontroller according to sensor node state at last.
The solar Energy prediction methods commonly used at present are EWMA, WCMA, Pro-Energy and the like, which utilize the Energy collected in the past time to inductively analyze and predict according to the characteristic of the periodic change of solar Energy. The traditional prediction method cannot guarantee accurate prediction under the condition of frequent change of weather, and has limitations.
The BP neural network can train, learn and store the mapping relation according to the input and the output of historical data without knowing the mathematical relation between the input and the output. Conventional solar collection power modeling typically selects meteorological factors such as time, wind speed, air temperature and relative humidity as input parameters. Such models always omit factors to simplify the calculations, which results in a reduction of model accuracy, which may not accurately predict future solar power due to a lack of certain meteorological factors. The direct prediction model uses historical solar energy collection power as an input variable of a neural network, and effectively predicts the solar energy collection power at the next moment.
Disclosure of Invention
In order to overcome the defects of a traditional solar energy collection prediction method of a wireless sensor network, the invention provides an energy prediction method of an energy-available sensor node based on a BP neural network, and solves the problem of low future energy collection prediction precision of the sensor node in the energy acquisition wireless sensor network. The invention trains different BP neural networks aiming at different weathers, and the solar energy collection power prediction in sunny days and rainy days has higher precision. In addition, the invention dynamically changes the network weight, the threshold value and the learning rate in the training process, and improves the defect that the BP neural network is easy to fall into a local extremum.
In order to achieve the above object, the present invention provides the following technical solution, a method for predicting energy of an energy-available sensor node based on a BP neural network, comprising the steps of:
s1) selecting historical solar energy collection power data of a target area to be classified and sampled to obtain two groups of training sample sets;
s2) carrying out data preprocessing on the two groups of training sample sets obtained in the step S1 to obtain two groups of training samples;
s3) respectively training two BP neural networks by using the two groups of training samples obtained in the step S2;
s4) performing data preprocessing on the prediction input vector;
s5) selecting a proper prediction model to predict an output vector, and performing inverse normalization to obtain the predicted solar energy collection power.
Step S1) selects the solar energy collection power of the first 24 hours and the solar energy collection power of the 25 th hour as an input vector and an output vector, respectively. And classifying historical solar energy collecting power according to the daily weather condition, if the weather of a certain day is a sunny day, adding the solar energy collecting data of 24 hours of the day into a sunny day training sample set, and otherwise, adding the solar energy collecting power of 24 hours of the day into a rainy day training sample set.
Step S2) specifically includes the following steps:
step S21) the sample training sets in sunny days and rainy days are processed separately, and abnormal data in the sample sets are eliminated. The anomaly data includes negative data and data that the solar power collected is greater than the theoretical maximum collected. Adopting zero padding to repair negative value data, adopting a linear interpolation method to repair missing data, and adopting the following formula:
Figure BDA0002274259880000021
where t is the time at which interpolation is required, t0Is a time value before t, t1Is a time value after t, y0And y1Are respectively t0And t1Solar collection power, y being linearly interpolated solar collection power. t is t0And t1And respectively taking the previous time and the next time of t.
Step S22) predicts the solar energy collection power at the next time as an output vector and the solar energy collection power in the first 24 hours as an input vector from the time series. By T1,，T2，...，T24And T25Representing input and output vectors, respectively, prediction T25The expression is as follows:
T25＝f(T1,T2,...,T24)
step S23), normalizing the input vector and the output vector obtained in step S22 to obtain a normalized input vector and a normalized output vector, wherein the normalization formula is as follows:
Figure BDA0002274259880000031
i＝1,2,...,n
k＝1,2,...,m
where n is the number of input layer nodes, m is the number of output layer nodes, xiFor normalizing the input vector of the pre-neural network to the i-th component, ykFor normalizing the pre-processing neural network output vector k component, xi,maxAnd xi,minThe maximum value and the minimum value, y, of the ith component of all the original input vectors before normalization processingk,maxAnd yk,minRespectively the maximum and minimum of the k-th component of all the raw output vectors before normalization processing,to normalize the processed input vector for the ith component,
Figure BDA0002274259880000036
is the k component of the normalized output vector. Normalizing all input and output data to [0.1,0.9 ]]Within range, data is guaranteed to be in the same order of magnitude.
The step S3) specifically includes the following steps:
step S31), a prediction model is established, and the network parameters of the BP neural network are set as follows: the number of nodes of the input layer is 24, the number of nodes of the output layer is 1, the number of nodes of the hidden layer is set to be a single layer, the number of nodes of the hidden layer is set to be 13, the transfer function of the hidden layer is selected from logsig, and the transfer function of the output layer is selected from purelin. Levenberg-Marquardt gold (LM) is selected as the iteration method.
Step S32), respectively training two BP neural networks by using the two groups of training samples obtained in the step S2 according to the prediction model parameters obtained in the step S31. The hidden layer output calculation formula is:
Figure BDA0002274259880000033
j＝1,2,...,l
where n is the number of nodes in the input layer, l is the number of nodes in the hidden layer, HjFor the jth node output of the hidden layer, XiIs the ith input value, W, of the BP neural networkijIs the connection weight of the ith node of the input layer and the jth node of the hidden layer, ajFor the hidden layer threshold, f is the neuron excitation function.
The output layer outputs a calculation formula:
Figure BDA0002274259880000034
k＝1,2,...,m
where m is the number of nodes in the output layer, WjkIs the connection weight, O, between the jth node of the hidden layer and the kth node of the output layerkFor the predicted output of the BP neural network, bkIs the output layer threshold.
According to the foregoing method for predicting energy of an capacitatable sensor node based on a BP neural network, the training model of the BP neural network in step S3 has an elastic adaptive characteristic, and the weight, the threshold, and the learning rate of the network are adjusted according to an error in the training process. The learning rate adjustment formula is:
k＝1,2,...,m
η denotes the learning rate used in the training process, η' denotes the updated value of the learning rate after the training, c is a constant greater than 1, d is a constant greater than 0 and less than 1, ekError obtained after the last training process, ek' is the error obtained after the training process is finished.
The error calculation formula after each training process is as follows:
ek＝Yk-Ok
k＝1,2,...,m
in the formula YkIs the desired output. The weight value updating formula is as follows:
i＝1,2,...,n；j＝1,2,...,l
Wjk′＝Wjk+ηHjek
j＝1,2,...,l；k＝1,2,...,m
in the formula Wij＇、Wjk"is the updated weight. The threshold update formula is:
Figure BDA0002274259880000043
j＝1,2,...,l
bk′＝bk+ek
k＝1,2,...,m
in the formula aj＇、bk"is the updated threshold.
Step S4), collecting solar energy collection power 24 hours before the prediction time as an input vector of the BP neural network, eliminating abnormal data, and carrying out normalization processing to obtain the input vector after the normalization processing.
Step S5), the output vector of the BP neural network is subjected to inverse normalization processing to obtain the predicted solar energy collection power at the moment, and the inverse normalization processing formula is as follows:
Figure BDA0002274259880000044
k＝1,2,...,m
Figure BDA0002274259880000045
predicting and outputting the k component, y before the inverse normalization processing for the BP neural networkkFor solar energy collection power after reverse normalization, yk,max、yk,minRespectively the maximum value and the minimum value of the k component in the original output vector before normalization processing. m is the number of output layer nodes of the BP neural network, and m is set to be 1 in the invention.
The invention has the beneficial effects that: the method and the device predict the solar energy collecting power at the next moment based on the historical solar energy collecting power, and avoid the problem that the future collecting power cannot be accurately predicted due to the lack of certain meteorological factors. According to different weather conditions, the invention trains two solar energy collection power prediction models, and ensures higher prediction precision in sunny days or rainy days. The weight, the threshold value and the learning rate of the neural network are adaptively changed according to the error elasticity in the training process, so that the algorithm is prevented from falling into a local minimum value.
Drawings
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings, in which:
FIG. 1 architecture diagram of a node of an energy-conserving sensor
FIG. 2BP neural network topology structure diagram
Detailed Description
The following describes a prediction example of the present invention in detail with reference to the accompanying drawings.
The invention provides an energy prediction method of an energy-available sensor node based on a BP neural network, wherein the energy prediction method of the energy-available sensor node has a structure shown in figure 1 and comprises the following steps:
s1) selecting historical solar energy collection power data of a target area, wherein the data sampling interval is 1 hour, and classifying according to the weather condition of the data to obtain two groups of training sample sets;
s2) eliminating abnormal data from the two groups of training sample sets obtained in the step S1, setting the solar energy collecting power at a certain moment as an output vector, setting the solar energy collecting power 24 hours before the moment as an input vector, and carrying out normalization processing on the input and output vectors to obtain two groups of training samples;
s3) respectively training two BP neural networks by using the two groups of training samples obtained in the step S2 to obtain a trained prediction model;
s4) collecting solar energy collection power 24 hours before the prediction moment, and carrying out data preprocessing to obtain a prediction input vector;
s5) selecting a proper prediction model according to the weather condition of the prediction time, inputting the prediction input vector generated in the step S4 into the prediction model to obtain an output vector of the model, and performing inverse normalization processing on the prediction output vector to obtain the predicted solar energy collection power at the time.
Step S1) processing the training sample set to obtain training samples, which is described in detail as follows:
in the present invention, the influence factor of the solar collected power predicted at the next time is the solar collected power of the first 24 hours. The data sampling interval was 1 hour, and the solar energy collection power for the first 24 hours and the solar energy collection power for the 25 th hour were the input vector and the output vector, respectively. And classifying historical solar energy collecting power according to the daily weather condition, if the weather of a certain day is a sunny day, adding the solar energy collecting data of 24 hours of the day into a sunny day training sample set, and otherwise, adding the solar energy collecting power of 24 hours of the day into a rainy day training sample set. And taking the weather condition of the output vector as a judgment basis, and adding the input vector and the output vector into a corresponding sample set.
Step S2) carrying out data preprocessing on the training sample obtained in the step S1;
further, step S2) includes the following steps:
step S21) the sample training sets in sunny days and rainy days are processed separately, and abnormal data in the sample sets are eliminated. The abnormal data comprises negative value data and data that the solar energy collecting power is larger than the theoretical collecting maximum value, and the theoretical collecting maximum value is set to be 400W/m2. Adopting zero padding to repair negative value data, adopting a linear interpolation method to repair missing data, and adopting the following formula:
where t is the time at which interpolation is required, t0Is a time value before t, t1Is a time value after t, y0And y1Are respectively t0And t1Solar collection power, y being linearly interpolated solar collection power. t is t0And t1Respectively taking the previous moment and the later moment of t, if the solar energy collecting power at the moment is also abnormal data to be eliminated, continuing to take the previous moment or the later moment of the moment until the solar energy collecting power at the moment is a normal value.
Step S22) predicts the solar energy collection power at the next time as an output vector and the solar energy collection power in the first 24 hours as an input vector from the time series. By T1,，T2，...，T24And T25Representing input and output vectors, respectively, prediction T25The expression is as follows:
T25＝f(T1,T2,...,T24)
step S23), normalizing the input vector and the output vector obtained in step S22 to obtain a normalized input vector and a normalized output vector, wherein the normalization formula is as follows:
Figure BDA0002274259880000062
i＝1,2,...,n
Figure BDA0002274259880000063
k＝1,2,...,m
where n is the number of input layer nodes, m is the number of output layer nodes, xiFor normalizing the input vector of the pre-neural network to the i-th component, ykFor normalizing the pre-processing neural network output vector k component, xi,maxAnd xi,minThe maximum value and the minimum value, y, of the ith component of all the original input vectors before normalization processingk,maxAnd yk,minRespectively the maximum and minimum of the k-th component of all the raw output vectors before normalization processing,
Figure BDA0002274259880000064
to normalize the processed input vector for the ith component,
Figure BDA0002274259880000065
is the k component of the normalized output vector. Since during 24 hours of the day, the night solar collection power is zero, the day solar collection power is greater than zero and larger values are likely to occur, normalizing all input and output data to [0.1,0.9 ]]Within range, data is guaranteed to be in the same order of magnitude.
Step S3) train the BP neural network using the training samples output at S2:
further, step S3) includes the following steps:
step S31), a prediction model is established, the BP neural network topological graph is shown in fig. 2, and the network parameters of the BP neural network are set as follows: the number of nodes of the input layer is 24, the number of nodes of the output layer is 1, the hidden layer is set to be a single layer, and the number range of the nodes of the hidden layer is determined by the following formula:
Figure BDA0002274259880000071
n is the number of nodes of an input layer, m is the number of nodes of an output layer, α is a constant between 1 and 10, the number of nodes l of an implicit layer is calculated by a formula, the number l of the nodes of the implicit layer is 6 to 15, the implicit layer is sequentially taken from 6 to 15, performance indexes (a coefficient of solution R2, an average percentage error MPE and a root mean square error RMSE) are calculated, l in the best index is taken as the number of nodes of the model implicit layer, the value l of the method is 13, a transfer function of the implicit layer selects 'logsig', a transfer function of the output layer selects 'purelin', an iteration method selects Levenberg-Marquardt algorithm (LM), and the algorithm can help a BP neural network to improve convergence speed and precision and avoid falling into a local minimum value.
Step S32), respectively training two BP neural networks by using the two groups of training samples obtained in the step S2 according to the prediction model parameters obtained in the step S31. The hidden layer output calculation formula is:
Figure BDA0002274259880000072
j＝1,2,...,l
where n is the number of nodes in the input layer, l is the number of nodes in the hidden layer, HjFor the jth node output of the hidden layer, XiIs the ith input value, W, of the BP neural networkijIs the connection weight of the ith node of the input layer and the jth node of the hidden layer, ajFor the hidden layer threshold, f is the neuron excitation function.
The output layer outputs a calculation formula:
Figure BDA0002274259880000073
k＝1,2,...,m
where m is the number of nodes in the output layer, WjkIs the connection weight, O, between the jth node of the hidden layer and the kth node of the output layerkFor the predicted output of the BP neural network, bkIs the output layer threshold.
According to the foregoing method for predicting energy of an capacitatable sensor node based on a BP neural network, the training model of the BP neural network in step S3 has an elastic adaptive characteristic, and the weight, the threshold, and the learning rate of the network are adjusted according to an error in the training process. The learning rate adjustment formula is:
Figure BDA0002274259880000081
k＝1,2,...,m
η denotes the learning rate used in the training process, η' denotes the updated value of the learning rate after the training, c is a constant greater than 1, d is a constant greater than 0 and less than 1, ekError obtained after the last training process, ek' is the error obtained after the training process is finished.
The error calculation formula after each training process is as follows:
ek＝Yk-Ok
k＝1,2,...,m
in the formula YkIs the desired output. The weight value updating formula is as follows:
i＝1,2,...,n；j＝1,2,...,l
Wjk′＝Wjk+ηHjek
j＝1,2,...,l；k＝1,2,...,m
in the formula Wij＇、Wjk"is the updated weight. The threshold update formula is:
Figure BDA0002274259880000083
j＝1,2,...,l
bk′＝bk+ek
k＝1,2,...,m
in the formula aj＇、bk"is the updated threshold.
Step S4), collecting solar energy collection power 24 hours before the prediction time as an input vector of the BP neural network, eliminating abnormal data, and carrying out normalization processing, wherein the input vector is normalized according to the maximum value and the minimum value of the input vector of the training sample to obtain the input vector after the normalization processing.
Step S5), the output vector output by the BP neural network is subjected to inverse normalization processing to obtain the predicted solar energy collecting power at the moment, and the output vector is subjected to inverse normalization according to the maximum value and the minimum value of the output vector of the training sample. The anti-normalization processing formula is as follows:
Figure BDA0002274259880000084
k＝1,2,...,m
Figure BDA0002274259880000091
predicting and outputting the k component, y before the inverse normalization processing for the BP neural networkkFor solar energy collection power after reverse normalization, yk,max、yk,minRespectively the maximum value and the minimum value of the k component in the original output vector before normalization processing. m is the number of output layer nodes of the BP neural network, and m is set to be 1 in the invention.

Claims (7)

1. A node energy prediction method of an energy-available sensor based on a BP neural network is characterized by comprising the following steps: according to the method, the historical solar energy collection power is selected as the input parameter of the BP neural network, and the defect that the accuracy is reduced due to the fact that some factors are ignored by a model when meteorological factors are selected as the input vector is avoided. The method trains two BP neural networks according to weather conditions, and ensures that power prediction has higher precision in both sunny days and rainy days. The model predicts the solar energy collecting power of the next hour by using the solar energy collecting power of the first 24 hours, and reduces the prediction error of the model by taking measures such as data preprocessing, setting appropriate network parameters and transmission functions, training optimal connection weight and the like.
The method specifically comprises the following steps:
s1) selecting historical solar energy collection power data of a target area to be classified and sampled to obtain two groups of training sample sets;
s2) carrying out data preprocessing on the two groups of training sample sets obtained in the step S1 to obtain two groups of training samples;
s3) respectively training two BP neural networks by using the two groups of training samples obtained in the step S2;
s4) performing data preprocessing on the prediction input vector;
s5) selecting a proper prediction model to predict an output vector, and performing inverse normalization to obtain the predicted solar energy collection power.
2. The energy prediction method of the available energy sensor nodes based on the BP neural network as claimed in claim 1, wherein the step S1 is to classify the historical solar energy collection power according to the daily weather condition, if the weather of a certain day is a sunny day, the solar energy collection data of 24 hours of the day is added into a sunny day training sample set, otherwise, the solar energy collection power of 24 hours of the day is added into a cloudy and rainy day training sample set.
3. The energy prediction method of the capacitatable sensor node based on the BP neural network as claimed in claim 1, wherein in step S2, the processing of the input vector of the BP neural network comprises the following steps:
step S21) the sample training sets in sunny days and rainy days are processed separately, and abnormal data in the sample sets are eliminated. The anomaly data includes negative data and data that the solar power collected is greater than the theoretical maximum collected. Adopting zero padding to repair negative value data, adopting a linear interpolation method to repair missing data, and adopting the following formula:
Figure FDA0002274259870000011
where t is the time at which interpolation is required, t0Is a time value before t, t1After t isA time value, y0And y1Are respectively t0And t1Solar energy collection power, y is linearly interpolated solar energy collection power, t and t0、t1The time difference from t is as short as possible.
Step S22) predicts the solar energy collection power at the next time as an output vector and the solar energy collection power in the first 24 hours as an input vector from the time series. By T1,，T2，...，T24And T25Representing input and output vectors, respectively, prediction T25The expression is as follows:
T25＝f(T1,T2,...,T24)
step S23), normalizing the input vector and the output vector obtained in step S22 to obtain a normalized input vector and a normalized output vector, wherein the normalization formula is as follows:
Figure FDA0002274259870000021
Figure FDA0002274259870000022
where n is the number of input layer nodes, m is the number of output layer nodes, xiFor normalizing the input vector of the pre-neural network to the i-th component, ykFor normalizing the pre-processing neural network output vector k component, xi,maxAnd xi,minThe maximum value and the minimum value, y, of the ith component of all the original input vectors before normalization processingk,maxAnd yk,minRespectively the maximum and minimum of the k-th component of all the raw output vectors before normalization processing,
Figure FDA0002274259870000023
to normalize the processed input vector for the ith component,
Figure FDA0002274259870000024
is the k component of the normalized output vector. Normalizing all input and output data to [0.1,0.9 ]]Within range, data is guaranteed to be in the same order of magnitude.
4. The energy prediction method for the capacitatable sensor nodes based on the BP neural network as claimed in claim 1, wherein in step S3, the network parameters of the BP neural network are: the number of nodes of the input layer is 24, the number of nodes of the output layer is 1, the number of nodes of the hidden layer is set to be a single layer, the number of nodes of the hidden layer is set to be 13, the transfer function of the hidden layer is selected from logsig, and the transfer function of the output layer is selected from purelin. The iteration method uses Levenberg-Marquardt algorithms (LM).
5. The BP neural network-based energy prediction method for the capacitatable sensor nodes according to claim 1, wherein the training model of the BP neural network in step S3 has an elastic adaptive characteristic, and the weight, the threshold and the learning rate of the network are adjusted according to an error in the training process. The learning rate adjustment formula is:
Figure FDA0002274259870000025
η denotes the learning rate used in the training process, η' denotes the updated value of the learning rate after the training, c is a constant greater than 1, d is a constant greater than 0 and less than 1, ekError obtained after the last training process, ek' is the error obtained after the training process is finished.
The error calculation formula after each training process is as follows:
ek＝Yk-Ok
k＝1,2,...,m
in the formula YkTo desired output, OkIs the prediction output of the BP neural network. The weight value updating formula is as follows:
Figure FDA0002274259870000026
Wjk′＝Wjk+ηHjek
j＝1,2,...,l；k＝1,2,...,m
where n is the number of nodes in the input layer, l is the number of nodes in the hidden layer, m is the number of nodes in the output layer, HjFor the jth node output of the hidden layer, XiAs input value of BP neural network, WijIs the connection weight between the input layer and the hidden layer, WjkIs the connection weight, W, between the hidden layer and the output layerij＇、Wjk"is the updated weight. The threshold update formula is:
Figure FDA0002274259870000031
bk′＝bk+ek
k＝1,2,...,m
in the formula, ajAs a hidden layer threshold, bkAs output layer threshold, aj＇、bk"is the updated threshold.
6. The energy prediction method of the capacitatable sensor node based on the BP neural network as claimed in claim 1, wherein said step S4 collects the solar energy collection power 24 hours before the prediction time as the input vector of the BP neural network, excludes abnormal data, and performs normalization processing to obtain the normalized input vector.
7. The energy prediction method of the capacitatable sensor node based on the BP neural network as claimed in claim 1, wherein the output vector of the neural network of step S5BP is subjected to an inverse normalization process to obtain the predicted solar energy collection power at that moment, and the inverse normalization process formula is:
Figure FDA0002274259870000032
Figure FDA0002274259870000033
predicting and outputting the k component, y before the inverse normalization processing for the BP neural networkkFor solar energy collection power after reverse normalization, yk,max、yk,minRespectively the maximum value and the minimum value of the k component in the original output vector before normalization processing. m is the number of output layer nodes of the BP neural network, and m is set to be 1 in the invention.
CN201911116683.6A 2019-11-15 2019-11-15 Energy prediction method for energy-available sensor node based on BP neural network Pending CN110852514A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911116683.6A CN110852514A (en) 2019-11-15 2019-11-15 Energy prediction method for energy-available sensor node based on BP neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911116683.6A CN110852514A (en) 2019-11-15 2019-11-15 Energy prediction method for energy-available sensor node based on BP neural network

Publications (1)

Publication Number Publication Date
CN110852514A true CN110852514A (en) 2020-02-28

Family

ID=69600844

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911116683.6A Pending CN110852514A (en) 2019-11-15 2019-11-15 Energy prediction method for energy-available sensor node based on BP neural network

Country Status (1)

Country Link
CN (1) CN110852514A (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103886223A (en) * 2014-04-14 2014-06-25 中国科学院声学研究所 Method and system for predicting power
CN104219682A (en) * 2014-08-20 2014-12-17 北京农业信息技术研究中心 Method and system for constructing network layers of hybrid power supply wireless sensor networks of farmlands
CN109669017A (en) * 2017-10-17 2019-04-23 中国石油化工股份有限公司 Refinery's distillation tower top based on deep learning cuts water concentration prediction technique

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103886223A (en) * 2014-04-14 2014-06-25 中国科学院声学研究所 Method and system for predicting power
CN104219682A (en) * 2014-08-20 2014-12-17 北京农业信息技术研究中心 Method and system for constructing network layers of hybrid power supply wireless sensor networks of farmlands
CN109669017A (en) * 2017-10-17 2019-04-23 中国石油化工股份有限公司 Refinery's distillation tower top based on deep learning cuts water concentration prediction technique

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
李玲生等: ""基于BP神经网络的太阳能收集功率预测研究"", 《软件导刊》 *
江良征: ""电弧焊熔透视觉传感与识别方法研究"", 《中国优秀硕士学位论文全文数据库 工程科技I辑》 *

Similar Documents

Publication Publication Date Title
CN109214592B (en) Multi-model-fused deep learning air quality prediction method
CN110705743B (en) New energy consumption electric quantity prediction method based on long-term and short-term memory neural network
CN105574615B (en) wavelet-BP neural network wind power prediction method based on spatial correlation and GA
CN112733462B (en) Ultra-short-term wind power plant power prediction method combining meteorological factors
CN111626518A (en) Urban daily water demand online prediction method based on deep learning neural network
CN111414717A (en) XGboost-L ightGBM-based unit power prediction method
CN112101669B (en) Photovoltaic power interval prediction method based on improved extreme learning machine and quantile regression
CN111833202B (en) Farmland evapotranspiration short-term prediction method considering crop coefficient dynamic change and rainfall
WO2019205528A1 (en) Antenna data-based algorithm for predicting solar energy collection power of wireless sensor network node
CN111598225B (en) Air conditioner cold load prediction method based on self-adaptive deep confidence network
CN112116144A (en) Regional power distribution network short-term load prediction method
CN112819189A (en) Wind power output prediction method based on historical predicted value
CN116167531A (en) Photovoltaic power generation prediction method based on digital twin
CN115759465A (en) Wind power prediction method based on multi-target collaborative training and NWP implicit correction
CN115689068A (en) Thermal load prediction method based on room temperature correction and CNN-BilSTM
CN112927097A (en) Photovoltaic power generation short-term prediction method based on GRA-ABC-Elman model
CN114429233A (en) Cluster algorithm-based station-level optimal power control method and system
CN116187540B (en) Wind power station ultra-short-term power prediction method based on space-time deviation correction
CN107330518A (en) Energy management control method and system based on temperature adjustment load prediction
CN109447843B (en) Photovoltaic output prediction method and device
CN110852514A (en) Energy prediction method for energy-available sensor node based on BP neural network
CN111401659A (en) Ultra-short-term or short-term photovoltaic power generation power prediction method based on case reasoning
Tai et al. Power prediction of photovoltaic power generation based on LSTM model with additive Attention mechanism
CN114936640A (en) Online training method for new energy power generation intelligent prediction model
Kaysal et al. Comparison the performance of different optimization methods in artificial intelligence based electricity production forecasting

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20200228