CN112910288B - Over-temperature early warning method based on inverter radiator temperature prediction - Google Patents
Over-temperature early warning method based on inverter radiator temperature prediction Download PDFInfo
- Publication number
- CN112910288B CN112910288B CN202011442899.4A CN202011442899A CN112910288B CN 112910288 B CN112910288 B CN 112910288B CN 202011442899 A CN202011442899 A CN 202011442899A CN 112910288 B CN112910288 B CN 112910288B
- Authority
- CN
- China
- Prior art keywords
- temperature
- early warning
- inverter
- distribution
- prediction
- 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
- 238000000034 method Methods 0.000 title claims abstract description 65
- 238000009826 distribution Methods 0.000 claims abstract description 93
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 17
- 230000008569 process Effects 0.000 claims abstract description 4
- 238000005259 measurement Methods 0.000 claims abstract description 3
- 230000006870 function Effects 0.000 claims description 20
- 238000012549 training Methods 0.000 claims description 17
- 238000013528 artificial neural network Methods 0.000 claims description 16
- 230000007246 mechanism Effects 0.000 claims description 15
- 210000004027 cell Anatomy 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 7
- 238000004140 cleaning Methods 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 6
- 238000012423 maintenance Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000000630 rising effect Effects 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 5
- 210000004205 output neuron Anatomy 0.000 claims description 5
- 230000006403 short-term memory Effects 0.000 claims description 5
- GNFTZDOKVXKIBK-UHFFFAOYSA-N 3-(2-methoxyethoxy)benzohydrazide Chemical compound COCCOC1=CC=CC(C(=O)NN)=C1 GNFTZDOKVXKIBK-UHFFFAOYSA-N 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 230000001413 cellular effect Effects 0.000 claims description 3
- 230000010354 integration Effects 0.000 claims description 3
- 230000004044 response Effects 0.000 claims description 3
- FGUUSXIOTUKUDN-IBGZPJMESA-N C1(=CC=CC=C1)N1C2=C(NC([C@H](C1)NC=1OC(=NN=1)C1=CC=CC=C1)=O)C=CC=C2 Chemical compound C1(=CC=CC=C1)N1C2=C(NC([C@H](C1)NC=1OC(=NN=1)C1=CC=CC=C1)=O)C=CC=C2 FGUUSXIOTUKUDN-IBGZPJMESA-N 0.000 claims description 2
- 238000000342 Monte Carlo simulation Methods 0.000 claims description 2
- 238000013459 approach Methods 0.000 claims description 2
- 238000013507 mapping Methods 0.000 claims description 2
- 239000000203 mixture Substances 0.000 claims description 2
- 238000011056 performance test Methods 0.000 claims description 2
- 238000012163 sequencing technique Methods 0.000 claims description 2
- 230000015654 memory Effects 0.000 abstract description 12
- 230000008859 change Effects 0.000 abstract description 4
- 238000010248 power generation Methods 0.000 description 11
- 238000013461 design Methods 0.000 description 4
- 238000005070 sampling Methods 0.000 description 4
- 230000007613 environmental effect Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000017525 heat dissipation Effects 0.000 description 2
- 230000009191 jumping Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000009529 body temperature measurement Methods 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000013341 scale-up Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000010998 test method Methods 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02M—APPARATUS FOR CONVERSION BETWEEN AC AND AC, BETWEEN AC AND DC, OR BETWEEN DC AND DC, AND FOR USE WITH MAINS OR SIMILAR POWER SUPPLY SYSTEMS; CONVERSION OF DC OR AC INPUT POWER INTO SURGE OUTPUT POWER; CONTROL OR REGULATION THEREOF
- H02M7/00—Conversion of ac power input into dc power output; Conversion of dc power input into ac power output
- H02M7/42—Conversion of dc power input into ac power output without possibility of reversal
- H02M7/44—Conversion of dc power input into ac power output without possibility of reversal by static converters
- H02M7/48—Conversion of dc power input into ac power output without possibility of reversal by static converters using discharge tubes with control electrode or semiconductor devices with control electrode
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Power Engineering (AREA)
- Inverter Devices (AREA)
Abstract
The invention discloses an over-temperature early warning method based on inverter radiator temperature prediction, which relates to the technical field of electronic power electronics, and firstly removes noise generated in a data acquisition process by introducing an empirical mode decomposition method; then, based on a Bayes long-time memory network, a model for predicting the temperature of the inverter radiator is provided, and the model comprises the following steps: (1) constructing a traditional long-time and short-time memory network, and extracting data time sequence characteristics; (2) introducing Bayes thought, performing approximate inference by using a Monte Carlo dropout method, and learning the network weight by minimizing KL divergence between the approximate distribution and posterior distribution of the network weight; (3) and according to the distribution condition of the temperature prediction result of the inverter, calculating the deviation condition between the predicted value and the actual value of the model by adopting two measurement modes of the square Mahalanobis distance and the local density ratio respectively, and adjusting the network weight. The method effectively grasps the temperature change trend of the radiator of the inverter and realizes the over-temperature early warning of the inverter.
Description
Technical Field
The invention relates to the technical field of power electronics, in particular to an over-temperature early warning method for an inverter.
Background
With the rapid development of economy, fossil energy consumption is continuously increasing, and the world is facing increasingly serious energy shortage and environmental destruction problems. Photovoltaic power generation is one of the most promising power generation modes with scale-up and commercialization development prospects in renewable energy sources at present as an environment-friendly renewable energy power generation technology, and is receiving more and more attention. Because the photovoltaic power station is often built in remote areas with severe natural geographic environments such as Gobi and islands, it is especially important to ensure reliable and stable operation of the core device photovoltaic grid-connected inverter. Over-temperature faults are a type of fault that often occurs with inverters. The failure is caused by many reasons, such as environmental temperature rise, failure of the heat dissipation fan of the inverter, overhigh power generation of the inverter, aging of the inverter, over-limit of current and voltage of the inverter, and the like. Over-temperature of the inverter can cause the generated power to be derated, and the inverter can be directly stopped when the generated power is serious, so that huge economic loss of power generation is caused. Therefore, the temperature of the radiator of the inverter is quickly and accurately predicted, an over-temperature early warning is necessary to the inverter in advance, a power supply plan is adjusted timely, and the economic benefit of operation of a photovoltaic power station is improved.
The existing invention about inverter temperature prediction mainly carries out mechanism modeling based on internal physical parameters of an inverter. Such schemes will typically first establish a photovoltaic inverter component temperature prediction equation, and the parametric information required for the prediction equation includes ambient temperature, heat sink temperature rise, inverter component temperature rise, and inverter component temperature. Wherein the ambient temperature is measured using a temperature sensor or a meteorological monitor. Next, such a scheme would calculate the inverter heat sink temperature rise by establishing a thermal equilibrium equation of state for the photovoltaic inverter heat sink. Then, an equation of the temperature difference between the IGBT and the radiator in a stable state, namely the temperature rise of the inverter element, is established by using the heat dissipation coefficient and the power consumption of the Insulated Gate Bipolar Transistor (IGBT). And finally, calculating the predicted value of the temperature of the photovoltaic inverter element by combining the obtained ambient temperature, the temperature rise of the radiator, the temperature rise of the inverter element and the temperature of the photovoltaic inverter element according to the established temperature prediction equation of the photovoltaic inverter element.
Another scheme of the prior invention is to establish a weather-related photovoltaic module working temperature prediction method through a mathematical statistics method according to a large amount of data. According to the scheme, a nonlinear model of the working temperature, the ambient temperature, the radiation intensity and the wind speed of the photovoltaic module is established. And finally, carrying out linear fitting according to a large amount of data by a mathematical statistics method to finally obtain a linear temperature prediction model.
The existing invention also has an inverter temperature prediction scheme based on a neural network. The scheme firstly builds a Back Propagation (BP) neural network for IGBT junction temperature prediction, wherein the number of input layers is 1, the neural network comprises three neurons which are respectively used for inputting a phase current peak value, a switching frequency and an ambient temperature; the number of the output layers is 1, and the output layers comprise a neuron for outputting IGBT junction temperature. Then, a 3D thermal simulation model of the inverter is built by adopting ANSYS Icepak software, and a plurality of groups of IGBT junction temperatures and corresponding junction temperature characteristics are collected by changing environmental information and other parameter information, wherein the junction temperature characteristics comprise: the method comprises the steps that phase current peak values, switching frequency and ambient temperature are preprocessed and used as training samples of the BP neural network, and an IGBT junction temperature prediction model based on the BP neural network is obtained through training by dividing a training set and a test set. And finally, inputting the collected actual junction temperature characteristics into a pre-trained IGBT junction temperature prediction model to obtain the junction temperature of the IGBT to be measured.
The prior invention method considers the radiator temperature of the inverter as a known quantity or assumes that the quantity can be measured by a temperature sensor, however, the inverter installed in the actual field rarely has the radiator temperature measurement function due to the cost. Meanwhile, the existing scheme ignores the problem of preprocessing field acquired data, for example, the three-phase current and three-phase voltage data contain a lot of interference noise, which can cause great influence on the accuracy of the prediction result. In addition, the existing prediction scheme based on the neural network adopts the most basic BP neural network, a multi-layer network prediction method under deep learning is not considered, point prediction is only carried out on temperature values of the inverter at each moment, and posterior distribution information of the temperature of the inverter is not utilized.
Therefore, based on the analysis, in a large-scale ground photovoltaic power station, an effective data preprocessing and accurate inverter temperature prediction method is urgently needed, the calculation cost of the system is minimized, the time delay requirement that the data of the photovoltaic sensor is stored and processed in real time is met, meanwhile, the temperature change trend of the inverter radiator is timely and effectively grasped according to the inverter temperature prediction value, and the over-temperature early warning of the inverter is realized.
Therefore, technicians in the field are dedicated to developing an inverter over-temperature early warning method based on Empirical Mode Decomposition (EMD) and Bayesian Long Short-Term Memory network (BLSTM) in a large-scale ground photovoltaic power station, so that an over-temperature early warning can be sent out to the inverter in advance, and the method has important significance for avoiding huge power generation economic loss and major safety accidents caused by over-temperature derating and shutdown of the inverter.
Disclosure of Invention
In view of the above defects in the prior art, the technical problem to be solved by the invention is how to quickly and accurately predict the temperature of the inverter radiator by data preprocessing and mining the time sequence information and posterior distribution information of predicted data of the inverter by adopting a multilayer neural network, and designing an inverter over-temperature early warning mechanism, so that the inverter over-temperature early warning function is realized, and the huge power generation economic loss and the occurrence of serious safety accidents of a large-scale ground photovoltaic power station caused by the over-temperature derating and shutdown of the inverter are avoided.
In order to achieve the purpose, the invention provides an over-temperature early warning method based on inverter radiator temperature prediction, which comprises the steps of firstly, removing noise generated in the data acquisition process by introducing an Empirical Mode Decomposition (EMD) method; then based on Bayesian Long Short-Term Memory network (BLSTM), a prediction model of inverter radiator temperature is provided, which comprises: (1) constructing a traditional Long Short-Term Memory (LSTM) network, and extracting data time sequence characteristics; (2) introducing Bayes thought, performing approximate inference by using a Monte Carlo dropout method, and learning the network weight by minimizing KL divergence between the approximate distribution and posterior distribution of the network weight; (3) and according to the distribution condition of the temperature prediction result of the inverter, calculating the deviation condition between the predicted value and the actual value of the model by adopting two measurement modes of the square Mahalanobis distance and the local density ratio respectively, and adjusting the network weight.
Further, the over-temperature early warning method based on inverter radiator temperature prediction comprises the following steps:
step a, data cleaning, namely cleaning original data by selecting empirical mode decomposition, wherein three-phase current and three-phase voltage data collected by an actual inverter contain more noise;
b, preprocessing data, namely converting the data into effective data which can be processed by a neural network;
step c, establishing LSTM;
step d, establishing BLSTM;
step e, realizing BLSTM based on the variation dropout;
f, multi-step prediction is carried out, an early warning mechanism under various time scales of ultra-short time, long time and the like is established for the inverter, and the output dimension of the designed BLSTM is adjustable;
step g, quantizing the approximate posterior distribution;
and h, early warning of the over-temperature of the inverter.
Further, the step a further comprises:
a1, signal decomposition, namely decomposing the original three-phase current and three-phase voltage data into 14 Intrinsic Mode Function (IMF) components respectively by using the empirical Mode method;
step a2, filtering signals, then sequencing 14 decomposed IMF components from low to high according to frequency, and eliminating the last four high-frequency IMF components which are regarded as pseudo components, namely high-frequency noise;
a3, signal reconstruction, namely respectively superposing 10 effective IMF components of the three-phase current and the three-phase voltage after signal filtration to obtain the reconstructed three-phase current and three-phase voltage data;
further, the step b further comprises:
b1, dividing a data set, namely firstly setting proportion division parameters of a training set and a test set, dividing 70% of historical operation data of the inverter into the training set for training an inverter radiator temperature prediction model, and taking 30% of the historical operation data of the inverter as the test set for model performance test;
b2, normalizing the data, and processing the data by using a MinMaxScaler method, wherein the mathematical expression is as follows:
wherein x max Representing the maximum value, x, in the input sample min Represents the minimum value in the sample, and X is the normalized result, ranging from 0 to 1.
Further, in step c, in order to learn the parameters of the LSTM network, the loss function is typically selected as a mean square error loss:
or selecting a cross entropy loss function:
where Θ represents the parameter set of the LSTM,
is the desired output value of the network; in addition, the invention further introduces an L2 regularization term to prevent neural network overfitting:
L(Θ)＝J(Θ)+λ(||W f || 2 +||W i || 2 +||W c || 2 +||W o || 2 )
where λ is the regularization parameter.
Further, the BLSTM obtains a probability model of input and output mapping by integrating the statistical modeling of the LSTM network parameters; the probability model solving method comprises the following steps:
C t ＝f i ω (x t ,h t-1 )
where the indices i and o denote the indices of the hidden and output layer nodes, respectively, f i ω And
respectively representing two nonlinear operators;
the probability of each data point output is:
where τ is an accuracy parameter reflecting the inherent noise of the data;
step 2, a training data set (historical operation data of the inverter stored in a large-scale ground photovoltaic power station) containing X (the three-phase current and the three-phase voltage of the inverter, the air temperature in the inverter and the transformer temperature) and Y (the actual temperature of a radiator of the inverter in a photovoltaic system) is given, and a posterior distribution p (omega | X, Y) needing to be estimated is learned in a parameter space; obtaining a predicted output y of the inverter radiator temperature by integration using the updated distribution * Distribution of (a):
p(y * ∣x * ,X,Y)＝∫p(y * ∣x * ,ω)p(ω∣X,Y)dω
wherein x is * Representing a new observation, for the prior distribution, a standard zero-mean gaussian prior on the weight matrix p (w) is usually chosen, and the uncertainty of the prediction will be directly reflected in the posterior distribution p (y) * ∣x * ,X,Y)。
Further, in the step e, a refractory posterior distribution p (ω | X, Y) is first approximated by a simple parameterized distribution q (ω), and then approximated by a Monte Carlo (MC) integral of q (ω), which specifically includes the following steps:
step e1, obtaining an approximate distribution by weight matrix decomposition, for w k For each row of (a), the variation dropout will impose a distribution of variation, i.e. the approximate distribution q (ω) is a mixture of two gaussian distributions with small variance:
where p is a predefined dropout probability, σ 2 Is a small precision parameter, m k Is a variation parameter;
step e2, learning the weights of the network by minimizing the KL divergence between the approximated distribution and the posterior distribution, so that the approximated distribution q (ω) in the variation inference is as close as possible to the true posterior distribution p (ω | X, Y), specifically, the following objective function is minimized:
KL(q(ω)||p(ω∣X,Y)))
further, the step f is implemented by adjusting the number of output neurons of the BLSTM in the step d to realize multi-step prediction, wherein each neuron corresponds to a prediction step.
Further, step g introduces two methods to quantify the deviation of the actual inverter heat sink temperature value from its corresponding predicted distribution, specifically including:
(1) the squared mahalanobis distance method suitable for gaussian prediction distribution: if the predicted distribution obtained by the Monte Carlo method is Gaussian or approximately Gaussian, the squared Mahalanobis distance can be used to characterize the deviation of the predicted value from the actual value. First, a Monte Carlo sample at time t is used to predict the distribution
To approximate the sample mean value mu t Sum covariance S t ；
When observing true value X t When available, the squared mahalanobis distance is determined by the following equation:
(2) local density ratio method for non-gaussian distributions: if the predicted distribution is not well described as a Gaussian distribution, then it is necessary to quantify the anomalies for each observation using a non-parametric approach. For these cases, a Local Density Ratio (LDR) method is introduced that is closely related to the local outlier factor; LDR statistics quantify deviation of each new observation from its predicted distribution using an estimate of the density of the nearest k neighbor (k-NNs) observations around it; local density estimation based on k-NNs
Can be calculated from the following formula:
wherein,
representing the nearest k predictors around x,
d (p, x) represents the Euclidean distance between the predicted value x and another predicted value p,
then, the value x is observed t Local density ofThe degree ratio is defined as:
that is, x t The local density of the nearest k surrounding predictors is averaged and divided by x t The local density of (a).
Further, in the step h, according to the proportion of the predicted temperature to the upper limit of the inverter temperature, the inverter over-temperature early warning is divided into three levels, namely a mild early warning (70% -80%), a moderate early warning (80% -90%) and a severe early warning (90% or more). Meanwhile, aiming at different early warning levels, the invention designs a corresponding response mechanism, which can be processed by the following conditions:
s1, a severe early warning is generated, and at any time, when the temperature prediction result of a certain time reaches the severe early warning, the system immediately sends out an over-temperature early warning to prompt power station operation and maintenance personnel to take corresponding maintenance measures on the inverter;
and S2, a moderate early warning is generated, when the temperature prediction result of a certain time is the moderate early warning, the system does not immediately send an over-temperature warning, but continuously tracks the prediction results of 2 times (namely 5 minutes and 10 minutes) in the future, and if the prediction results are the moderate early warnings and show a temperature rising trend (the temperature prediction value at 10 minutes is higher than the temperature prediction value at 5 minutes), the system sends the over-temperature warning. Specially, when severe early warning appears in the two future prediction results, an alarm is given immediately;
and S3, a mild early warning is generated, when the temperature prediction result of a certain time is the mild early warning, the system does not immediately send out an over-temperature warning, but continuously tracks the prediction results of 5 times (namely 5 minutes, 10 minutes, 15 minutes, 20 minutes and 25 minutes) in the future continuously, and if the prediction results are mild early warnings and the temperature is in an increasing trend, the system sends out the over-temperature warning. Specially, when severe early warning appears in the prediction results of the next 5 times, the alarm is given immediately;
and S4, jumping between the mild early warning and the moderate early warning, and performing subsequent judgment according to the moderate early warning rule when the temperature is predicted to rise from the mild early warning to the moderate early warning, wherein the temperature is in a rising trend. When the predicted temperature is reduced from the moderate early warning to the mild early warning, the temperature is in a descending trend, and subsequent judgment is carried out according to mild early warning rules. Particularly, when severe early warning appears in the prediction result, the alarm is given immediately.
Aiming at the problems that the temperature of an internal module of a large-scale ground photovoltaic power station inverter is difficult to measure and predict through mechanism modeling, and the power generation power is derated and stopped to cause huge power generation economic loss due to easy over-temperature fault, the method effectively improves the quality of original data by utilizing an EMD (empirical mode decomposition) method, realizes quick and accurate prediction of the temperature of an inverter radiator based on a Bayesian long-time memory network, effectively grasps the temperature change trend of the inverter radiator, and realizes early warning of the over-temperature of the inverter.
The conception, the specific structure and the technical effects of the present invention will be further described with reference to the accompanying drawings to fully understand the objects, the features and the effects of the present invention.
Drawings
FIG. 1 is an overall flow diagram of a preferred embodiment of the present invention;
FIG. 2 is an LSTM bulk operation mechanism of a preferred embodiment of the present invention;
FIG. 3 is a block diagram of a BLSTM structure of a preferred embodiment of the present invention;
fig. 4 is a cloud-edge coordination-based inverter over-temperature warning mechanism according to a preferred embodiment of the present invention.
Detailed Description
The technical contents of the preferred embodiments of the present invention will be more clearly and easily understood by referring to the drawings attached to the specification. The present invention may be embodied in many different forms of embodiments and the scope of the invention is not limited to the embodiments set forth herein.
As shown in fig. 4, the invention provides an inverter over-temperature early warning system based on empirical mode decomposition and bayes long-time memory network for a large-scale ground photovoltaic power station.
The system acquires real-time data information from the inverter at each sampling moment, and three-phase voltage and three-phase current data in the data are cleaned by using an EMD method to remove high-frequency noise mixed in the data. And then, the local computing equipment processes the cleaned data by utilizing a pre-trained inverter radiator temperature prediction module and outputs a prediction interval and a point prediction value of the inverter radiator temperature. Finally, the multi-stage over-temperature early warning mechanism established by the invention can effectively avoid the missing judgment and the misjudgment.
The invention will be further described with reference to fig. 1.
Aiming at the problems that the temperature of an internal module of a large-scale ground photovoltaic power station inverter is difficult to measure and predict through mechanism modeling, and the power generation power derating shutdown is caused to cause huge power generation economic loss due to easy over-temperature fault, the invention effectively improves the quality of original data by utilizing an EMD (empirical mode decomposition) method, realizes the rapid and accurate prediction of the temperature of an inverter radiator based on a Bayesian long-time memory network, effectively grasps the temperature change trend of the inverter radiator, realizes the over-temperature early warning of the inverter, and specifically comprises the following steps:
1. data cleaning: considering that three-phase current and three-phase voltage data collected by an actual inverter contain more noises, the method adopts empirical mode decomposition to clean the original data, and improves the data quality. Firstly, signal decomposition is carried out, and the original three-phase current and three-phase voltage data are respectively decomposed into 14 Intrinsic Mode Function (IMF) components by utilizing EMD. Secondly, signal filtering is carried out, 14 IMF components obtained by decomposition are sorted from low to high according to frequency, and finally four high-frequency IMF components are regarded as pseudo components, namely high-frequency noise and are removed. And finally, signal reconstruction, namely respectively superposing 10 effective IMF components of the three-phase current and the three-phase voltage after signal filtering to obtain the data of the three-phase current and the three-phase voltage after reconstruction.
2. Data preprocessing: after data cleaning, the invention further preprocesses the data in order to convert the data into effective data which can be processed by a neural network. Firstly, data set division is carried out, the proportion division parameters of a training set and a test set are set, and the size of the parameters can be 0-1 (more than 0 and less than 1 need to be ensured). The division ratio of the data set selected by the method is 0.7, namely 70% of the historical operation data volume of the inverter is divided into a training set for training the inverter radiator temperature prediction model, and the other 30% of the historical operation data volume of the inverter is used for testing the model performance. And then, data normalization is carried out, historical data recorded by the inverter comprise a plurality of different types and have different data scales, for example, the numerical values of three-phase voltage and three-phase current are all three to four hundred, but the temperature numerical value is only dozens, the inverter efficiency is small, and therefore multi-source heterogeneous data need to be subjected to standardization processing. The invention selects a MinMaxScaler method to carry out normalization processing on data, and the mathematical expression is as follows:
wherein x max Representing the maximum value, x, in the input sample min Represents the minimum value in the sample, and X is the normalized result, ranging from 0 to 1.
3. Establishing a long-time memory network: LSTM removes or adds information to the state of the cell through well-designed structures called "gates". The gate is a method for determining whether information passes through, and comprises a Sigmoid neural network layer and a bit-wise multiplication operation. The mathematical formula of the Sigmoid function is as follows:
where x represents the input, the output value y of the Sigmoid function is a value between 0 and 1, 0 represents "no amount is allowed to pass" and 1 means "any amount is allowed to pass". The LSTM has three gates, an input gate, a forgetting gate and an output gate, for protecting and controlling the cell state.
Wherein, the forgetting gate can read the output value h of the last moment of the cell t-1 And the input value x at the current time t Outputting a value f between 0 and 1 t Given cell state C t-1 。
f t ＝σ(W f ×[C t-1 ,h t-1 ,x t ]+b f )
Wherein W f And b f Are all weighting parameters of the network and σ is the Sigmoid activation function.
The input gate (input gate) is used to update the cell information.
i t ＝σ(W i ·[h t-1 ,x t ]+b i )
On the other hand, a candidate value vector (cell) is constructed:
the candidate vector is then multiplied by the entry gate to select the information to be updated.
Wherein, W i 、b i 、X C 、b C Is the weight parameter of the network and tanh is the activation function.
Afterwards, LSTM renews cell state C t-1 Is C t ，f t Dot-by-dot C t-1 Representing that forgotten information is to be discarded.
Dot multiplied by i t Representing the information to be updated in the candidate vector.
Finally, LSTM determines the output h of the network by means of output gates t And obtaining a final output value.
o t ＝σ(W o [h t-1 ,x t ]+b o )
h t ＝o t *tanh(C t )
Wherein W o 、b o Is a network weight parameter.
To optimize the parameters of the LSTM network, the mean square error is typically chosen as a loss function:
or selecting a cross entropy loss function:
where Θ represents the parameter set of the LSTM model,
the expected output value of the network, namely the temperature value of the tag inverter radiator in the training set in the invention. In addition, the present invention further introduces an L2 regularization term to prevent neural networks from overfitting:
L(Θ)＝J(Θ)+λ(||W f || 2 +||W i || 2 +||W c || 2 +||W o || 2 )
where λ is the regularization parameter.
4. Establishing a Bayesian long-time memory network: model parameter ω ═ W f ,W i ,W c ,W o ,b f ,b y ,b c ,b o As a random variable of the prior distribution p (ω). Thus, the cellular state and output of the LSTM network can be re-expressed as:
C t ＝f i ω (x t ,h t-1 )
where the indices i and o denote the indices of the hidden and output layer nodes, respectively, f i ω And
respectively representing two non-linear operators.
The probability of each data point output is.
Where τ is an accuracy parameter reflecting the inherent noise of the data, in the present invention, for the convenience of calculation, the likelihood function is assumed to satisfy the normal distribution, and the likelihood function is forwarded through the LSTM network.
Then, a training data set (historical inverter operation data stored in a large-scale ground photovoltaic power station) containing X (specifically, electrical parameter information such as three-phase current and three-phase voltage of the inverter and environment information such as air temperature and transformer temperature in the inverter) and Y (specifically, actual temperature of an inverter radiator in a photovoltaic system) is given, and the posterior distribution p (omega | X, Y) needing to be estimated is learned on a parameter space. With the updated distribution, the predicted output y of the inverter radiator temperature can be obtained by integration * Distribution of (a):
p(y * ∣x * ,X,Y)＝∫p(y * ∣x * ,ω)p(ω∣X,Y)dω
wherein x is * A new observation is represented where dependencies on the precision parameters, hidden layer states and past inputs are ignored. For prior distributions, a standard zero-mean Gaussian prior on the weight matrix p (W) is usually chosen, and the uncertainty of the prediction will be directly reflected in the posterior distribution p (y) * ∣x * ,X,Y)。
5. The Monte Carlo (MC) dropout technique was introduced: the invention uses the variation dropout in the variation reasoning of Bayesian long-time memory network. Variational inference is a technique that approximates a difficult a posteriori distribution p (ω | X, Y) using a simple parameterized distribution q (ω). At this time, the integral term can be approximated by a monte carlo integral of q (ω). Specifically, the approximate distribution is obtained by weight matrix decomposition. For w k Each row ofThe variational dropout applies a variation distribution, i.e. the approximate distribution q (ω) can be integrated from two gaussian distributions with small variance:
where p is a predefined dropout probability, σ 2 Is a small precision parameter, m k Is a variation parameter. In order to make the approximate distribution q (ω) in the variation inference maximally approximate the true posterior distribution p (ω | X, Y), the inventive method learns the weights of the network by minimizing the KL-divergence between the approximate distribution and the posterior distribution, in particular, minimizing the following objective function:
KL(q(ω)||p(ω∣X,Y)))
it is noted that the variable length memory network uses a fixed dropout mask at each time step, including the loop layer. Input, output and loop connections are optionally dropped at each time step. This is in contrast to the prior art, where different neural network elements are discarded at different time steps, and the fully-connected layer is not discarded.
The variational dropout method used in the test procedure can be regarded as an approximation of the posterior prediction distribution p (ω | X, Y) monte carlo sample. Given a new observation x * N samples of approximate prediction posteriori can be collected by forwarding N samples of random model
The corresponding empirical estimates of the posterior predicted mean, standard deviation, and covariance are:
wherein τ can be estimated as
A predefined regularization/weight decay parameter lambda is given.
6. Multi-step prediction: according to the invention, the output dimensionality of the designed Bayes long-time memory network is adjustable in consideration of the fact that early warning mechanisms under various time scales of ultra-short time, long time and the like possibly need to be established for the inverter on the actual photovoltaic field.
The present invention considers the sampling time interval of the actual inverter data as one step, for example, the inverter data is collected every 5 minutes, the single-step prediction means predicting the inverter radiator temperature of the next sampling (after 5 minutes), and the six-step prediction means predicting the inverter radiator temperature of the next six sampling (after 5 minutes, after 10 minutes, after 15 minutes, after 20 minutes, after 25 minutes, and after 30 minutes).
Specifically, the invention realizes multi-step prediction by adjusting the number of output neurons of a Bayesian long-time memory network in the fourth step, wherein each neuron corresponds to one prediction step length. For example, the number of output neurons of the network is set to 1 in the case of single-step prediction, and is set to 6 in the case of six steps. In the invention, the number of output neurons of the Bayesian long-time memory network is adjusted by only modifying a few model parameters, and then the model is retrained by using historical data to be deployed to an actual application site, so that the over-temperature early warning of the inverter under any time scale can be efficiently realized
7. Quantizing the approximate posterior distribution, and optimizing the network weight: the invention introduces two methods to quantify the deviation of the actual inverter radiator temperature predicted value and the actual value.
(1) The squared mahalanobis distance method suitable for gaussian prediction distribution: if the predicted distribution is a Gaussian distribution, orApproximating a gaussian distribution, the squared mahalanobis distance can be used to characterize the magnitude of the deviation of the predicted value from the actual value. First, a Monte Carlo sample at time t is used to predict the distribution
To approximate the sample mean value mu t Sum covariance S t 。
When observing true value X t When available, the squared mahalanobis distance is determined by the following equation:
the fact that the Mahalanobis distance is large indicates that the deviation between the actual observed value and the predicted posterior distribution is large, and indicates that the inverter radiator temperature prediction model is not accurate enough, and parameters in the model need to be further adjusted.
(2) Local density ratio method for non-gaussian distributions: if the predicted distribution is not well described as a Gaussian distribution, then the abnormal condition of each observation is quantified using a non-parameterized Local Density Ratio (LDR) method. The LDR statistic quantifies the deviation of each new observation from its predicted distribution using an estimate of the density of the nearest k neighbor (k-NNs) observations around it.
Local density estimation based on k-NNs
Can be calculated from the following formula:
wherein,
representing the nearest k predictors around x,
d (p, x) represents the Euclidean distance between the predicted value x and another predicted value p,
then, the value x is observed t The local density ratio of (a) is defined as:
that is, x t The local density of the nearest k surrounding predictors is averaged and divided by x t The local density of (a). When the LDR is large, the deviation between the posterior distribution and the predicted value is large, the inverter radiator temperature prediction model is not accurate enough, and parameters in the model need to be further adjusted. Meanwhile, a proper k value needs to be selected, a smaller k can cause larger fluctuation of the predicted value of the model, and a larger k can reduce the prediction accuracy of the model.
9. Inverter over-temperature early warning: considering that the operation conditions of inverters in different photovoltaic systems are different and the upper limits of the working temperatures of different inverters are different, the invention does not design an inverter over-temperature early warning mechanism based on absolute temperature (for example, the over-temperature early warning is sent out when the temperature exceeds a specific numerical value), but designs a multi-stage over-temperature early warning mechanism according to the temperature proportion.
According to the method, the over-temperature early warning of the inverter is divided into three levels of mild early warning (70-80%), moderate early warning (80-90%) and severe early warning (90% or more) according to the proportion of the predicted temperature to the upper limit of the temperature of the inverter. Meanwhile, aiming at different early warning levels, the invention designs a corresponding response mechanism, thereby effectively avoiding misjudgment and missed judgment.
Specifically, the following cases can be handled:
(1) and when the severe early warning is generated, the system immediately sends out over-temperature early warning to prompt power station operation and maintenance personnel to take corresponding maintenance measures for the inverter whenever the temperature prediction result reaches the severe early warning.
(2) When the temperature prediction result of a certain time is moderate early warning, the system does not immediately send out over-temperature warning, but continuously tracks the prediction results of 2 times (namely 5 minutes and 10 minutes in the future) in the future, and if the prediction results are moderate early warnings and show a temperature rising trend (the temperature prediction value at 10 minutes is higher than the temperature prediction value at 5 minutes), the system sends out the over-temperature warning. Particularly, when severe early warning occurs in the two future prediction results, the alarm is given immediately.
(3) When a temperature prediction result of a certain time is mild early warning, the system does not immediately send out an over-temperature warning, but continuously tracks the prediction results of 5 continuous times (namely 5 minutes, 10 minutes, 15 minutes, 20 minutes and 25 minutes) in the future, and only if the prediction results are mild early warnings and the temperature is in an increasing trend, the system sends out the over-temperature warning. Specifically, when severe early warning occurs in the prediction results of 5 times in the future, the warning is immediately given.
(4) And jumping between the mild early warning and the moderate early warning, and when the temperature is predicted to rise from the mild early warning to the moderate early warning, the temperature is in a rising trend, and subsequent judgment is carried out according to a moderate early warning rule. When the predicted temperature is reduced from the moderate early warning to the mild early warning, the temperature is in a descending trend, and subsequent judgment is carried out according to mild early warning rules. Particularly, when severe early warning appears in the prediction result, the alarm is given immediately.
The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the prior art according to the concepts of the present invention should be within the scope of protection determined by the claims.
Claims (7)
1. An over-temperature early warning method based on inverter radiator temperature prediction is characterized in that firstly, an Empirical Mode Decomposition (EMD) method is introduced to remove noise generated in a data acquisition process; then, based on a Bayesian Long Short-Term Memory network (BLSTM), a prediction model of the temperature of the inverter radiator is provided, which comprises the following steps: (1) constructing a traditional Long Short-Term Memory (LSTM) network, and extracting data time sequence characteristics; (2) introducing Bayes thought, performing approximate inference by using a Monte Carlo dropout method, and learning the network weight by minimizing KL divergence between the approximate distribution and posterior distribution of the network weight; (3) according to the distribution condition of the temperature prediction result of the inverter, calculating the deviation condition between the predicted value and the actual value of the model by adopting two measurement modes of the square Mahalanobis distance and the local density ratio respectively, and adjusting the network weight;
the early warning method comprises the following steps:
step a, data cleaning, namely cleaning original data by selecting empirical mode decomposition, wherein three-phase current and three-phase voltage data collected by an actual inverter contain more noise;
b, preprocessing data, namely converting the data into effective data which can be processed by a neural network;
step c, establishing LSTM;
step d, establishing BLSTM;
step e, realizing BLSTM based on the variation dropout;
f, multi-step prediction is carried out, an early warning mechanism under various time scales of ultra-short time, short time and long time is established for the inverter, and the output dimension of the designed BLSTM is adjustable;
step g, quantizing the approximate posterior distribution;
h, early warning of the over-temperature of the inverter;
the step a further comprises:
a1, signal decomposition, namely decomposing the original three-phase current and three-phase voltage data into 14 Intrinsic Mode Function (IMF) components respectively by using the empirical Mode method;
step a2, filtering signals, then sequencing 14 decomposed IMF components from low to high according to frequency, and eliminating the last four high-frequency IMF components which are regarded as pseudo components, namely high-frequency noise;
a3, signal reconstruction, and finally, overlapping 10 effective IMF components of the three-phase current and the three-phase voltage respectively after signal filtration to obtain the three-phase current and the three-phase voltage data after reconstruction;
the BLSTM obtains a probability model of input and output mapping by integrating the statistical modeling of LSTM network parameters; the probability model solving method comprises the following steps:
step 1, parameter ω ═ W f ,W i ,W c ,W o ,b f ,b y ,b c ,b o As random variables of the prior distribution p (ω), the cellular state and output of the LSTM can therefore be re-expressed as:
C t ＝f i ω (x t ,h t-1 )
where the indices i and o denote the indices of the hidden and output layer nodes, respectively, f i ω And
respectively representing two nonlinear operators;
the probability of each data point output is:
where τ is the accuracy parameter reflecting the inherent noise of the data, parameter ω is the model parameter, C t Is thinCurrent time state of cell, C t-1 State on the cell at a moment, I D Representing an identity matrix;
step 2, a training data set containing X, Y is given, and X is the three-phase current and the three-phase voltage of the inverter, and the air temperature and the transformer temperature in the inverter; learning a posterior distribution p (omega | X, Y) to be estimated on a parameter space by the aid of the training data set, namely the actual temperature of the inverter radiator in the photovoltaic system, and the historical operating data of the inverter stored in the large-scale ground photovoltaic power station; obtaining a predicted output y of the inverter radiator temperature by integration using the updated distribution * Distribution of (a):
p(y * ∣x * ,X,Y)＝∫p(y * ∣x * ,ω)p(ω∣X,Y)dω
wherein x is * Representing a new observation, selecting a standard zero mean Gaussian prior on the weight matrix p (W) for prior distribution, and directly reflecting the prediction uncertainty in the prior distribution p (y) * ∣x * ,X,Y)。
2. The inverter radiator temperature prediction-based over-temperature warning method according to claim 1, wherein the step b further comprises:
b1, dividing a data set, namely firstly setting proportion division parameters of a training set and a test set, dividing 70% of historical operation data of the inverter into the training set for training an inverter radiator temperature prediction model, and taking 30% of the historical operation data of the inverter as the test set for model performance test;
b2, normalizing the data, and processing the data by using a MinMaxScaler method, wherein the mathematical expression is as follows:
wherein x max Representing the maximum value, x, in the input sample min Represents the minimum value in the sample, X is normalizedAs a result, the range is between 0 and 1.
3. The inverter radiator temperature prediction based over-temperature warning method of claim 2, wherein in the step c, in order to learn the parameters of the LSTM network, the loss function is selected as a mean square error loss:
or selecting a cross entropy loss function:
where Θ represents the parameter set of the LSTM,
is the desired output value of the network; in addition, an L2 regularization term is further introduced to prevent neural networks from overfitting:
L(Θ)＝J(Θ)+λ(||W f || 2 +||W i || 2 +||W c || 2 +||W o || 2 )
where λ is the regularization parameter.
4. The method of claim 3, wherein in the step e, a refractory a-posteriori distribution p (ω | X, Y) is first approximated by a simple parameterized distribution q (ω) and then approximated by a Monte Carlo (MC) integral of q (ω), comprising the steps of:
step e1, obtaining an approximate distribution by weight matrix decomposition, for w k For each row of (a), the variation dropout will impose a distribution of variation, i.e. the approximate distribution q (ω) is a mixture of two gaussian distributions with small variance:
where p is a predefined dropout probability, σ 2 Is a small precision parameter, m k Is a variation parameter;
step e2, learning the weights of the network by minimizing the KL divergence between the approximated distribution and the posterior distribution, so that the approximated distribution q (ω) in the variation inference is as close as possible to the true posterior distribution p (ω | X, Y), specifically, the following objective function is minimized:
KL(q(ω)||p(ω∣X,Y)))。
5. the inverter radiator temperature prediction-based over-temperature warning method according to claim 4, wherein the step f is implemented by adjusting the number of output neurons of the BLSTM in the step d to realize multi-step prediction, wherein each neuron corresponds to one prediction step.
6. The over-temperature early warning method based on inverter radiator temperature prediction of claim 5, wherein the step g introduces two methods to quantify the deviation of the actual inverter radiator temperature value from its corresponding prediction distribution, specifically comprising:
(1) the squared mahalanobis distance method suitable for gaussian prediction distribution: if the prediction distribution obtained by the Monte Carlo method is Gaussian distribution or approximate Gaussian distribution, the square Mahalanobis distance can be used for representing the deviation between the predicted value and the actual value; first, a Monte Carlo sample at time t is used to predict the distribution
To approximate the sample mean value mu t Sum covariance S t ；
When observing true value X t When available, the squared mahalanobis distance is determined by the following equation:
(2) local density ratio method for non-gaussian distributions: if the predicted distribution is not well described as a Gaussian distribution, then it is necessary to quantify the anomalies for each observation using a non-parametric approach; for these cases, a Local Density Ratio (LDR) method is introduced that is closely related to the local outlier factor; LDR statistics quantify deviation of each new observation from its predicted distribution using an estimate of the density of the nearest k neighbor (k-NNs) observations around it; local density estimation based on k-NNs
Can be calculated from the following formula:
wherein,
representing the nearest k predictors around x,
d (p, x) represents the Euclidean distance between the predicted value x and another predicted value p,
then, the observed valuex t The local density ratio of (a) is defined as:
that is, x t The local density of the nearest k surrounding predictors is averaged and divided by x t The local density of (a).
7. The inverter radiator temperature prediction-based over-temperature early warning method as claimed in claim 6, wherein in the step h, the inverter over-temperature early warning is divided into three levels of mild early warning, moderate early warning and severe early warning according to the proportion of the predicted temperature to the upper limit of the inverter temperature; the mild early warning is 70-80%; the moderate early warning is 80% -90%; the severe early warning is more than 90%; meanwhile, aiming at different early warning levels, a corresponding response mechanism is designed, and the following conditions can be processed:
s1, a severe early warning is generated, and at any time, when the temperature prediction result of a certain time reaches the severe early warning, the system immediately sends out an over-temperature early warning to prompt power station operation and maintenance personnel to take corresponding maintenance measures on the inverter;
s2, moderate early warning occurs, when the temperature prediction result of a certain time is moderate early warning, the system does not immediately send out over-temperature warning, but continuously tracks 2 times of future continuous prediction, namely the prediction results of 5 minutes and 10 minutes in the future, if the prediction results are moderate early warning and show a temperature rising trend, namely the temperature prediction value at 10 minutes is higher than the temperature prediction value at 5 minutes, the system can send out over-temperature warning; when severe early warning occurs in the two future prediction results, an alarm is given immediately;
s3, mild early warning occurs, when the temperature prediction result of a certain time is mild early warning, the system does not immediately send out overtemperature warning, but continuously tracks the prediction results of 5 times in the future, wherein the 5 times in the future are respectively 5 minutes, 10 minutes, 15 minutes, 20 minutes and 25 minutes in the future, and if the prediction results are mild early warning and above and the temperature is in an increasing trend, the system sends out overtemperature warning; when severe early warning occurs in the prediction results of the next 5 times, the warning is immediately given;
s4, a mild early warning and a moderate early warning jump occur, when the temperature is predicted to rise from the mild early warning to the moderate early warning, the temperature is in a rising trend, and subsequent judgment is carried out according to a moderate early warning rule; when the predicted temperature is reduced from the moderate early warning to the mild early warning, the temperature is in a descending trend, and subsequent judgment is carried out according to mild early warning rules; and when the prediction result shows severe early warning, the alarm is given immediately.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011442899.4A CN112910288B (en) | 2020-12-08 | 2020-12-08 | Over-temperature early warning method based on inverter radiator temperature prediction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011442899.4A CN112910288B (en) | 2020-12-08 | 2020-12-08 | Over-temperature early warning method based on inverter radiator temperature prediction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112910288A CN112910288A (en) | 2021-06-04 |
| CN112910288B true CN112910288B (en) | 2022-08-09 |
Family
ID=76111434
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011442899.4A Active CN112910288B (en) | 2020-12-08 | 2020-12-08 | Over-temperature early warning method based on inverter radiator temperature prediction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112910288B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114710114B (en) * | 2022-05-23 | 2022-09-13 | 北京华清未来能源技术研究院有限公司 | Photovoltaic inverter fault prediction method |
| CN115936060B (en) * | 2022-12-28 | 2024-03-26 | 四川物通科技有限公司 | Substation capacitance temperature early warning method based on depth deterministic strategy gradient |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107766968A (en) * | 2017-09-26 | 2018-03-06 | 河海大学 | Short-term wind speed forecasting method based on CAPSO RVM built-up patterns |
| EP3407267A1 (en) * | 2017-05-25 | 2018-11-28 | Hitachi, Ltd. | Deep learning network architecture optimization for uncertainty estimation in regression |
| CN109583346A (en) * | 2018-11-21 | 2019-04-05 | 齐鲁工业大学 | EEG feature extraction and classifying identification method based on LSTM-FC |
| CN110059844A (en) * | 2019-02-01 | 2019-07-26 | 东华大学 | Energy storage device control method based on set empirical mode decomposition and LSTM |
| CN110751199A (en) * | 2019-10-15 | 2020-02-04 | 南京航空航天大学 | Satellite anomaly detection method based on Bayesian neural network |
| CN110881037A (en) * | 2019-11-19 | 2020-03-13 | 北京工业大学 | Network intrusion detection method and training method and device of model thereof, and server |
| CN111626506A (en) * | 2020-05-27 | 2020-09-04 | 华北电力大学 | Regional photovoltaic power probability prediction method based on federal learning and cooperative regulation and control system thereof |
| WO2020181240A1 (en) * | 2019-03-06 | 2020-09-10 | Gritstone Oncology, Inc. | Identification of neoantigens with mhc class ii model |
| WO2020191004A1 (en) * | 2019-03-18 | 2020-09-24 | Georgia Tech Research Corporation | Tracking and alert method and system for worker productivity and safety |
| CN112765873A (en) * | 2020-11-30 | 2021-05-07 | 中国南方电网有限责任公司超高压输电公司广州局 | LSTM algorithm-based power transformation equipment heating prediction method |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102880170B (en) * | 2012-10-08 | 2015-03-25 | 南京航空航天大学 | System failure early warning method based on baseline model and Bayesian factor |
| CN107729716B (en) * | 2017-11-27 | 2020-10-27 | 西安建筑科技大学 | Coal mine water inrush prediction method based on long-time and short-time memory neural network |
| US10599769B2 (en) * | 2018-05-01 | 2020-03-24 | Capital One Services, Llc | Text categorization using natural language processing |
| CN109948861A (en) * | 2019-03-26 | 2019-06-28 | 西南交通大学 | A kind of urban rail based on mode decomposition and deep learning passenger flow forecasting in short-term |
| CN111950759B (en) * | 2020-05-18 | 2024-06-21 | 武汉科技大学 | Short-term wind speed prediction method based on two-stage decomposition, LSTM and AT |
| CN111695724B (en) * | 2020-06-01 | 2023-05-05 | 浙江大学 | Wind speed prediction method based on hybrid neural network model |
| CN111815065B (en) * | 2020-07-21 | 2023-08-29 | 东北大学 | Short-term power load prediction method based on long-short-term memory neural network |
-
2020
- 2020-12-08 CN CN202011442899.4A patent/CN112910288B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3407267A1 (en) * | 2017-05-25 | 2018-11-28 | Hitachi, Ltd. | Deep learning network architecture optimization for uncertainty estimation in regression |
| CN107766968A (en) * | 2017-09-26 | 2018-03-06 | 河海大学 | Short-term wind speed forecasting method based on CAPSO RVM built-up patterns |
| CN109583346A (en) * | 2018-11-21 | 2019-04-05 | 齐鲁工业大学 | EEG feature extraction and classifying identification method based on LSTM-FC |
| CN110059844A (en) * | 2019-02-01 | 2019-07-26 | 东华大学 | Energy storage device control method based on set empirical mode decomposition and LSTM |
| WO2020181240A1 (en) * | 2019-03-06 | 2020-09-10 | Gritstone Oncology, Inc. | Identification of neoantigens with mhc class ii model |
| WO2020191004A1 (en) * | 2019-03-18 | 2020-09-24 | Georgia Tech Research Corporation | Tracking and alert method and system for worker productivity and safety |
| CN110751199A (en) * | 2019-10-15 | 2020-02-04 | 南京航空航天大学 | Satellite anomaly detection method based on Bayesian neural network |
| CN110881037A (en) * | 2019-11-19 | 2020-03-13 | 北京工业大学 | Network intrusion detection method and training method and device of model thereof, and server |
| CN111626506A (en) * | 2020-05-27 | 2020-09-04 | 华北电力大学 | Regional photovoltaic power probability prediction method based on federal learning and cooperative regulation and control system thereof |
| CN112765873A (en) * | 2020-11-30 | 2021-05-07 | 中国南方电网有限责任公司超高压输电公司广州局 | LSTM algorithm-based power transformation equipment heating prediction method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112910288A (en) | 2021-06-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Qin et al. | Macroscopic–microscopic attention in LSTM networks based on fusion features for gear remaining life prediction | |
| Deng et al. | Wind power forecasting methods based on deep learning: A survey | |
| Cheng et al. | A hybrid remaining useful life prognostic method for proton exchange membrane fuel cell | |
| Xiao et al. | Multi-step wind speed forecasting based on a hybrid forecasting architecture and an improved bat algorithm | |
| Tan et al. | Prediction of water quality time series data based on least squares support vector machine | |
| Wang et al. | A novel non-linear combination system for short-term wind speed forecast | |
| Niu et al. | Uncertainty modeling for chaotic time series based on optimal multi-input multi-output architecture: Application to offshore wind speed | |
| CN112910288B (en) | Over-temperature early warning method based on inverter radiator temperature prediction | |
| Mei et al. | Day-ahead nonparametric probabilistic forecasting of photovoltaic power generation based on the LSTM-QRA ensemble model | |
| Samet et al. | Quantizing the deterministic nonlinearity in wind speed time series | |
| Jiang et al. | Solar irradiance capturing in cloudy sky days–a convolutional neural network based image regression approach | |
| Ranganayaki et al. | An intelligent ensemble neural network model for wind speed prediction in renewable energy systems | |
| Zhu et al. | Study of joint temporal-spatial distribution of array output for large-scale photovoltaic plant and its fault diagnosis application | |
| Zhang et al. | A novel hybrid approach based on cuckoo search optimization algorithm for short‐term wind speed forecasting | |
| Xue et al. | An improved generic hybrid prognostic method for RUL prediction based on PF-LSTM learning | |
| Lyu et al. | Lebesgue-time–space-model-based diagnosis and prognosis for multiple mode systems | |
| Jayawardene et al. | Comparison of adaptive neuro-fuzzy inference systems and echo state networks for PV power prediction | |
| CN112069666A (en) | Power grid short-term reliability evaluation method based on probabilistic power flow method | |
| Qu et al. | A D-stacking dual-fusion, spatio-temporal graph deep neural network based on a multi-integrated overlay for short-term wind-farm cluster power multi-step prediction | |
| CN117195713A (en) | Lithium battery module temperature field reconstruction method based on deep operator learning | |
| CN105205560B (en) | Photovoltaic power supply power prediction method based on positive and negative error variable weights | |
| Li et al. | A framework for predicting network security situation based on the improved LSTM | |
| CN116706907A (en) | Photovoltaic power generation prediction method based on fuzzy reasoning and related equipment | |
| Saha et al. | A Bayesian Framework for Remaining Useful Life Estimation. | |
| Ghasvarian Jahromi et al. | Wind power prediction based on wind speed forecast using hidden Markov model |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |