Single-phase photovoltaic inverter on-line condition monitoring and method for predicting residual useful life
Technical field
The present invention relates to power conversion circuit prognostic and health management technical field, more particularly to a kind of single-phase photovoltaic
Inverter on-line condition monitoring and method for predicting residual useful life.
Background technology
As the energy and environment crisis that gradually shortage and climate warming trigger of global fossil energy becomes increasingly conspicuous, the sun
Can photovoltaic generation with its aboundresources, widely distributed, exploitation prospect is wide etc., and advantage becomes most promising renewable
The energy.Photovoltaic DC-to-AC converter is the core component of whole photovoltaic generating system, and the direct current power that it sends photovoltaic module is reverse into
Required alternating electromotive force, while also there is maximal power tracing control function, error protection etc..Rui Tuo groups of the U.S. and
Sang Diya laboratory researches find that photovoltaic DC-to-AC converter is the main reason for causing photovoltaic generating system failure, and its crash rate
Up to 51%.In addition, inverter is generally positioned over open air jointly with system, the pole cold pole heat for being subjected to seasonal weather change for a long time is disliked
There are the influence of the factors such as overcurrent, overvoltage, frequency disturbance when bad environment and working status.Therefore, to realize photovoltaic generation
Stable, reliable, safety the longtime running of system, it is necessary to implement on-line condition monitoring and predicting residual useful life to it, consequently facilitating
Condition maintenarnce is carried out to it, to avoid or reduce failure caused by risk and maintenance cost.
At present, on-line condition monitoring and predicting residual useful life phase of the domestic and international related research institutes to single-phase photovoltaic inverter
To less, the status assessment or predicting residual useful life of offline mode are carried out primarily directed to the single component in circuit, it is domestic
Related research institutes are more to carry out fail-safe analysis by the way of Performance Degradation Data.In addition, single-phase photovoltaic inverter is in reality
Under border operation situation, influenced be subject to many extraneous factors, such as environment temperature, input voltage disturb.Therefore, how to determine single-phase
The state parameter of photovoltaic DC-to-AC converter and how to predict the problem of its remaining life becomes urgent need to resolve.
The present invention combines the more performance parameters of single-phase photovoltaic inverter and considers that the junction temperature of power MOSFET influences, using depth
Degree neural network model carries out it presence assessment, meanwhile, the historical time sequence according to power MOSFET ON resistance
Establish Gaussian process regression model and carry out multi-step prediction, realize predicting residual useful life.
The content of the invention
It is an object of the invention to provide a kind of single-phase photovoltaic inverter on-line condition monitoring and method for predicting residual useful life,
The input voltage of single-phase photovoltaic inverter, input current, output voltage, output current and power during by gathering health status
The drain-source electrode current and junction temperature of MOSFET, establishes deep neural network model, determines the state parameter base of single-phase photovoltaic inverter
Quasi- value;And power MOSFET ON Resistance standard value is calculated, determine the failure threshold of single-phase photovoltaic inverter.Based on depth nerve
Network model carries out presence assessment to tested single-phase photovoltaic inverter, meanwhile, using Gaussian process regression model to power
MOSFET conducting resistances carry out multi-step prediction, when conducting resistance predicted value reaches failure threshold first, you can obtain single-phase light
Lie prostrate the remaining life of inverter.
To achieve the above object, single-phase photovoltaic inverter on-line condition monitoring and method for predicting residual useful life of the invention,
Specifically include following steps:
State parameter a reference value and single-phase photovoltaic inverter when step 1, acquisition single-phase photovoltaic inverter are health status
Failure threshold, it is comprised the following steps that:
Step 1.1, to be defined on environment temperature be 25 DEG C, under the conditions of input voltage is rated operational voltage, and performance meets will
The state for the single-phase photovoltaic inverter asked is health status;Constant duration continuous acquisition single-phase photovoltaic inverter is p moment
The input voltage V of pointi(n), input current Ii(n), output voltage Vo(n), output current Io(n)And the hourglass source electrode of power MOSFET
Electric current IDS(n)With junction temperature Tj(n), wherein, n=1,2 ..., p;
Step 1.2, using depth belief network build deep neural network model, for the connection of input layer v and hidden layer h
Probability distribution is closed to be expressed as:P (v, h1..., hu)=P (v | h1)P(h1|h2)…P(hu-2|hu-1)P(hu-1|hu), wherein, u is
Hidden layer number;The input voltage V gathered according to step 1.1i(n), input current Ii(n), output voltage Vo(n), output current
Io(n)With the junction temperature T of power MOSFETj(n)As mode input sample data, using successively greedy algorithm to deep neural network
Model carries out unsupervised mode training;The input dimension for determining deep neural network model is 5 and output dimension is 1;Then, adopt
By the use of supervised learning mode using input sample data as monitoring data, according to maximum likelihood function, deep neural network is carried out
Fine setting;Finally, according to the deep neural network model determined, the status parameter values SC of single-phase photovoltaic inverter is obtained(n);
State parameter a reference value when step 1.3, calculating single-phase photovoltaic inverter are health status
Step 1.4, the conducting resistance a reference value for obtaining power MOSFET when single-phase photovoltaic inverter is health statusWhereinAnd RDS(n)For the conducting resistance of the n-th moment point
Value;Definition increases to its a reference value R when power MOSFET ON resistance valueDS01.2 times when, the residue of single-phase photovoltaic inverter
Service life is zero, i.e. the failure threshold F=1.2R of single-phase photovoltaic inverterDS0;
Step 2, according to the deep neural network model established in step 1, obtain tested single-phase photovoltaic inverter in tmWhen
The status parameter values at quarterAnd compared with state parameter a reference value, judge tmMoment is tested single-phase photovoltaic inverter
State, its status assessment criterion are as follows:
(1) whenWhen, it is normal condition to be tested single-phase photovoltaic inverter;
(2) whenWhen, it is faint degenerate state to be tested single-phase photovoltaic inverter;
(3) whenWhen, it is serious degenerate state to be tested single-phase photovoltaic inverter;
(4) whenWhen, tested single-phase photovoltaic inverter is to close on failure state;
Step 3, with power MOSFET in t1, t2..., tmThe conducting resistance at momentAs historical time sequence number
According to wherein z=1,2 ..., m;Kernel function of square exponential function as Gauss regression process model is selected, and hyper parameter is sought
Excellent setting;The Embedded dimensions for determining Gauss regression process model are w and time delay is τ, build the input vector of the modelOutput vectorSo as to predict that step number is τ;When pre-
Measure tm+qThe conducting resistance predicted value at momentWhen being more than failure threshold F first, then it can obtain single-phase photovoltaic inverter and exist
tmThe remaining life at moment is
The present invention combines the more performance parameters of single-phase photovoltaic inverter and considers that the junction temperature of power MOSFET influences, and builds respectively
Vertical deep neural network model carries out it on-line condition monitoring and Gaussian process regression model carries out multi-step prediction, realizes remaining
Life prediction, can carry out photovoltaic generating system effective and reasonable condition maintenarnce, reduce failure risk and maintenance and repair cost, be
Its long-term safety, stabilization, Effec-tive Function provide safeguard.
Brief description of the drawings
Fig. 1 is single-phase photovoltaic inverter schematic diagram;
Fig. 2 is single-phase photovoltaic inverter on-line condition monitoring and predicting residual useful life flow chart;
Embodiment
Embodiment
Technical scheme is described in detail with reference to Figure of description.
As shown in Fig. 2, the input voltage of single-phase photovoltaic inverter, input current, output electricity during by gathering health status
The drain-source electrode current and junction temperature of pressure, output current and power MOSFET, establish deep neural network model, determine single-phase photovoltaic
The state parameter a reference value of inverter, and power MOSFET ON Resistance standard value is calculated, determine the mistake of single-phase photovoltaic inverter
Imitate threshold value.Presence assessment is carried out to tested single-phase photovoltaic inverter based on deep neural network model, meanwhile, using Gauss
Process regression model carries out multi-step prediction to power MOSFET ON resistance, when conducting resistance predicted value reaches failure threshold first
When, you can obtain the remaining life of single-phase photovoltaic inverter.
Single-phase photovoltaic inverter on-line condition monitoring and method for predicting residual useful life, embodiment are as follows:
State parameter a reference value and single-phase photovoltaic inverter when step 1, acquisition single-phase photovoltaic inverter are health status
Failure threshold, it is comprised the following steps that:
Step 1.1, to be defined on environment temperature be 25 DEG C, under the conditions of input voltage is rated operational voltage, and performance meets will
The state for the single-phase photovoltaic inverter asked is health status;Constant duration continuous acquisition single-phase photovoltaic inverter is p moment
The input voltage V of pointi(n), input current Ii(n), output voltage Vo(n), output current Io(n)And the hourglass source electrode of power MOSFET
Electric current IDS(n)With junction temperature Tj(n), wherein, n=1,2 ..., p;
Step 1.2, using depth belief network build deep neural network model, for the connection of input layer v and hidden layer h
Probability distribution is closed to be expressed as:P (v, h1..., hu)=P (v | h1)P(h1|h2)…P(hu-2|hu-1)P(hu-1|hu), wherein, u is
Hidden layer number;The input dimension for determining deep neural network model is 5 and output dimension is 1, is gathered according to step 1.1 defeated
Enter voltage Vi(n), input current Ii(n), output voltage Vo(n), output current Io(n)With the junction temperature T of power MOSFETj(n)As model
Input sample data, unsupervised mode training is carried out using successively greedy algorithm to deep neural network model;Then, using prison
Mode of learning is superintended and directed using input sample data as monitoring data, according to maximum likelihood function, deep neural network is finely adjusted;
Finally, according to the deep neural network model determined, the status parameter values SC of single-phase photovoltaic inverter is obtained(n);
State parameter a reference value when step 1.3, calculating single-phase photovoltaic inverter are health status
Power MOSFET in step 1.4, data the n-th moment point single-phase photovoltaic inverter of calculating gathered according to step 1.1
Conduction resistance valueWhereinIt is healthy shape so as to obtain photovoltaic DC-to-AC converter
The conducting resistance a reference value of power MOSFET during stateDefinition increases when power MOSFET ON resistance value
To its a reference value RDS01.2 times when, the remaining life of single-phase photovoltaic inverter is zero, i.e. the failure threshold of single-phase photovoltaic inverter
Value F=1.2RDS0;
Step 2, obtain tested single-phase photovoltaic inverter in tmThe input voltage at momentInput currentOutput electricity
PressureOutput currentAnd the drain-source electrode current of power MOSFETAnd junction temperatureBy input voltageIt is defeated
Enter electric currentOutput voltageOutput currentWith the junction temperature of power MOSFETAs established in step 1
The input quantity of deep neural network model, obtains tmThe status parameter values of moment single-phase photovoltaic inverterAccording to following shape
State judgment criterion assesses the state of current time single-phase photovoltaic inverter:
(1) whenWhen, it is normal condition to be tested single-phase photovoltaic inverter;
(2) whenWhen, it is faint degenerate state to be tested single-phase photovoltaic inverter;
(3) whenWhen, it is serious degenerate state to be tested single-phase photovoltaic inverter;
(4) whenWhen, tested single-phase photovoltaic inverter is to close on failure state;
Power MOSFET is in t in step 3, the tested single-phase photovoltaic inverter of calculating1, t2..., tmThe conducting resistance at momentWhereinEstablish Gauss
Process regression model carries out multi-step prediction to it, realizes predicting residual useful life, it is comprised the following steps that:
(3.1) with power MOSFET in t1, t2..., tmThe conducting resistance at momentAs historical time sequence data;
(3.2) kernel function of square exponential function as Gauss regression process model is selected, and square exponential function is set
Hyper parameter initial value be zero;The optimal value of hyper parameter, wherein iteration are solved using conjugated gradient optimum algorithm iterative learning
Number is arranged to 100 times;
(3.3) Embedded dimensions for determining Gauss regression process model are w and time delay is τ, build the input of the model
VectorOutput vectorSo as to predict that step number is τ;
(3.4) multi-step prediction is carried out to power MOSFET ON resistance using established Gaussian process regression model, when
Predict tm+qThe conducting resistance predicted value at momentWhen being more than failure threshold F first, then single-phase photovoltaic inverter is can obtain
In tmThe remaining life at moment is