CN114091975A - Method for evaluating operation life of energy storage power station - Google Patents
Method for evaluating operation life of energy storage power station Download PDFInfo
- Publication number
- CN114091975A CN114091975A CN202111487606.9A CN202111487606A CN114091975A CN 114091975 A CN114091975 A CN 114091975A CN 202111487606 A CN202111487606 A CN 202111487606A CN 114091975 A CN114091975 A CN 114091975A
- Authority
- CN
- China
- Prior art keywords
- layer
- output
- follows
- hidden layer
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 63
- 238000004146 energy storage Methods 0.000 title claims abstract description 35
- 238000013528 artificial neural network Methods 0.000 claims abstract description 42
- 239000002245 particle Substances 0.000 claims abstract description 28
- 230000036541 health Effects 0.000 claims abstract description 25
- 238000012549 training Methods 0.000 claims abstract description 17
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 7
- 230000008569 process Effects 0.000 claims description 35
- 239000000126 substance Substances 0.000 claims description 35
- 238000011156 evaluation Methods 0.000 claims description 29
- 230000006870 function Effects 0.000 claims description 27
- 210000002569 neuron Anatomy 0.000 claims description 27
- 230000004913 activation Effects 0.000 claims description 10
- 238000010277 constant-current charging Methods 0.000 claims description 6
- 238000012886 linear function Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 5
- 238000010280 constant potential charging Methods 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 238000009795 derivation Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 2
- 238000012546 transfer Methods 0.000 claims description 2
- 230000009286 beneficial effect Effects 0.000 abstract description 4
- 230000008034 disappearance Effects 0.000 abstract description 4
- 238000004880 explosion Methods 0.000 abstract description 4
- 208000028659 discharge Diseases 0.000 description 13
- 238000007600 charging Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 8
- 238000004590 computer program Methods 0.000 description 7
- 238000007599 discharging Methods 0.000 description 6
- 238000002474 experimental method Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 230000000875 corresponding effect Effects 0.000 description 4
- 238000013178 mathematical model Methods 0.000 description 4
- 238000003860 storage Methods 0.000 description 4
- GELKBWJHTRAYNV-UHFFFAOYSA-K lithium iron phosphate Chemical compound [Li+].[Fe+2].[O-]P([O-])([O-])=O GELKBWJHTRAYNV-UHFFFAOYSA-K 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 229910052799 carbon Inorganic materials 0.000 description 2
- 210000004027 cell Anatomy 0.000 description 2
- 230000002596 correlated effect Effects 0.000 description 2
- 230000005684 electric field Effects 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 230000032683 aging Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000005562 fading Methods 0.000 description 1
- 239000000446 fuel Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000006386 neutralization reaction Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000013083 solar photovoltaic technology Methods 0.000 description 1
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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Economics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Strategic Management (AREA)
- Evolutionary Computation (AREA)
- Development Economics (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- Educational Administration (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
An improved method for evaluating the service life of an energy storage power station is disclosed, which comprises the following steps: initializing a population; calculating the fitness value of the particle; speed update and location update; an end condition; optimizing initial weight and threshold of a neural network by a particle swarm algorithm; and training the neural network by adopting the exponential decay learning rate, and evaluating the health state of the battery and the residual life of the battery. The method not only ensures the accuracy and stability of the prediction result, and avoids the problems of gradient disappearance, explosion and the like, but also is beneficial to the rapid convergence of the neural network, enhances the stability of the network and improves the learning efficiency of the network.
Description
Technical Field
The invention belongs to the technical field of new energy storage, relates to an evaluation method based on a big data technology, and particularly relates to an evaluation method of the operation life of an energy storage power station.
Background
With the proposal of carbon peak-reaching carbon neutralization target, new energy and renewable energy tend to enter the mainstream of an energy system more quickly, and the revolutionary energy transformation guides the major innovation of energy knowledge and a technical system and promotes the breakthrough of basic theory, technical chain, industrial form and the like. The new energy storage technology, the solar photovoltaic technology and the solar fuel technology are the most promising technical subjects. Among them, the breakthrough and popularization of large-scale energy storage technology is a strong support for the development of renewable energy.
However, the new energy often has the characteristics of intermittence, fluctuation, randomness, unpredictability, non-continuity and the like, and only the large-scale direct access to the power grid can generate no small negative influence on the stability and reliability of the system and the power supply quality. Electrical energy is a process energy source and cannot be directly stored, and must be converted to other forms of energy for storage. The energy storage power station realizes the four-quadrant flexible operation of power through the power conversion device, realizes the active and reactive instantaneous balance of the energy storage power station, and improves the system stability. In the technical standard blueprint planning of the smart power grid, an energy storage power station is taken as an important part of a smart power grid system, and the energy storage power station can realize the following functions in the power grid: the system has the advantages of rotating for standby, managing load, guaranteeing stability of a power grid, achieving intelligent scheduling and load balancing, having a load shedding function and being supported by a system in a reactive mode.
The technical characteristics of the energy storage power station are the basis of mathematical model establishment, energy storage battery health state analysis and example boundary parameter setting; the health state of the energy storage battery is related to factors such as a charging and discharging curve of the energy storage power station, the technical characteristics of the energy storage battery, the capacity attenuation characteristics of the energy storage battery and the like, and the service life of the energy storage power station can be evaluated by setting the boundary conditions of the health state of the energy storage battery. The capacity fading characteristics of energy storage power stations in normal life cycles and the evaluation of the state of health (SoH) of batteries, which is defined as the ratio of the remaining available capacity of a battery to the initial dischargeable amount of a new battery at the time of commissioning, are of increasing interest.
The evaluation of the health state of the battery is an important link in the evaluation of the state of the power station. Along with the gradual lengthening of the operation time of the energy storage power station, the health state of the battery is gradually reduced due to the attenuation of the available capacity of the battery, and the residual chargeable and dischargeable times are reduced. Generally, when the battery capacity is less than 70% of the initial dischargeable amount, as the end of the battery life, the number of discharge cycles in the period from the current observation time to the end of the battery life is charged as the remaining life. Therefore, the evaluation of the health state and the residual life of the battery is beneficial to the evaluation of the operation life of the energy storage power station, and meanwhile, a user can know the operation condition of the current battery of the energy storage power station in time, so that the safe and efficient operation of the battery pack of the energy storage power station is ensured.
The method for evaluating the operation life of the energy storage power station mainly comprises an experimental method, a battery model-based method and a data-driven method.
The experimental method is that a mathematical model is established by measuring main factors influencing the battery capacity attenuation speed such as charge-discharge depth, charge-discharge cycle times, environmental temperature in an energy storage power station and the like through experiments to calculate the health state evaluation and the service life of the battery. And establishing a service life mathematical model of the energy storage power station under different charge states based on the weighting coefficients in the Tan Xingsheng. Huitong et al established a mathematical model of the service life of the energy storage power station by considering the correlation between the service life of the energy storage power station and the depth of discharge. Peak and the like research the correlation between the health state of the energy storage battery and the charging and discharging cycle times based on experimental detection and analysis, and analyze the influence of the discharging depth on the relational expression.
Methods based on battery models include kalman filtering, observer methods, and particle filtering methods. Shinwei et al proposed a method for estimating the cell SoH based on ohmic internal resistance, followed and predicted the change in ohmic internal resistance of the cell using an improved unscented particle filter algorithm, and verified the accuracy of the method using experimental data.
The method based on data driving gets rid of the dependence of a prediction algorithm on the internal structure of the research object and a battery capacity attenuation model to a greater extent, does not need to consider a specific attenuation mechanism, and obtains the characteristic parameters of the performance change of the research object through the technologies of characteristic quantity identification, selection, fusion and the like; when the training of the prediction model is completed, the functional relationship between the prediction model and the input/output parameters is established, so that the more accurate life prediction, especially the prediction of the residual life, is realized in the middle and later stages. The current wide range of predictive models and algorithms include: SVM, GPR, correlation vector machine, and neural network methods, among others.
However, due to time-varying uncertainty, individual difference, measurement and other factors, the situation that the prediction result is uncertain in multiple prediction processes of the data-driven life prediction method is easily caused, and the reliability and accuracy of life prediction are directly influenced. Therefore, it becomes very critical how to further improve the overall performance of the remaining life prediction method by optimizing and improving the data-driven life prediction model, especially on the basis of improving the accuracy and stability of the single-step prediction result.
In view of the above-mentioned drawbacks of the prior art, there is still a need to find an improved method for assessing the operational lifetime of energy storage power stations.
Disclosure of Invention
The invention aims to provide an improved method for evaluating the operating life of an energy storage power station. Compared with the prior art, the evaluation method provided by the invention not only ensures the accuracy and stability of the prediction result, and avoids the problems of gradient disappearance or explosion and the like, but also is beneficial to the rapid convergence of the neural network, and enhances the stability of the network and the learning efficiency of the network.
In order to solve the technical problems, the invention adopts the following technical scheme: an improved method for evaluating the operating life of an energy storage power station comprises the following specific steps:
the first step is as follows: population initialization, to initial positionX 0And velocityV 0Inertial weightωLearning factorc 1,c 2Initializing parameters;
the second step is that: calculating the fitness value of the particles, and taking the root mean square error as a fitness function;
the third step: speed update and location update
Wherein the content of the first and second substances,r 1andr 2is [0,1]]A random number in between;V i andX i is the current timeiThe velocity and position of the particle;P ibest is as followsiIndividual optimal solutions for individual particles;G best the optimal solution of the population is obtained;
in order to avoid falling into a local optimal solution, a variation factor is introduced in the processλWhen the variation condition is satisfied,
reinitializing the positions and the speeds of all particles of the population and storing the current time information;
the fourth step: and (4) finishing conditions: outputting when the algorithm reaches the maximum iteration times or the error is smaller than a set value; otherwise, returning to the second step and continuing optimization;
the fifth step: optimizing initial weights of neural networks via particle swarm optimizationω ij Andω jk and a threshold valuea j Andb k ;
and a sixth step: training the neural network by using exponential decay learning rate, and evaluating the health state of the batterySoHAnd remaining battery lifeRLEvaluation, can obtain
Wherein the content of the first and second substances,T threshold is composed ofSoHThe charge-discharge period corresponding to the failure threshold value is reduced,iin order to train the size of the sample,qis predictedRLThe value of the one or more of the one,Tindicating charge and dischargeNumber of electrical cycles.
The evaluation method according to the present invention, wherein the fitness function is:
wherein the content of the first and second substances,pto describe the number of particles of a population size,Y k is as followskThe ideal output value of each particle is calculated,O k is as followskAn
The actual output value of the particle.
The evaluation method according to the present invention, wherein the model of the neural network output is:
the evaluation method according to the present invention, wherein the exponentially decaying learning rate is:
wherein the initial learning rate isη 0=0.2,decay_rateIn order to be the rate of decay,global_steprepresenting the number of current iterations with a decay rate ofdecay_step=100。
The evaluation method according to the invention, wherein the model of the neural network output is obtained according to the following method:
(1) selecting three-layer neural network, selecting the number of neurons in hidden layer according to empirical formula
Wherein the content of the first and second substances,min order to imply the number of layer nodes,nas nodes of the input layerThe number of the first and second groups is,las the number of nodes of the output layer,a1 to 10
A constant of (d);
(2) in the neural network, the forward transfer process of information is from an input layer to a hidden layer, from the hidden layer to an output layer, and each layer is output after being processed by an activation function; the output of the hidden layer is:
wherein the content of the first and second substances,Xas input variablesHI1、HI2 andHI3,ω ij is as followsiInput layer neurons andjthe weight of each hidden layer neuron,a j is as followsjThe threshold of each of the hidden layer neurons,θ j in order to have the hidden layer input,H j is as followsjThe output of each of the hidden layer neurons,f 1is a logsig function, which is a hidden layer activation function, i.e. a
the output layer activation function is a linear function, so the output layer output can be expressed as
Wherein the content of the first and second substances,β k the output layer input is predicted for the network,ω jk is as followsjHidden layer neurons andkthe weight of each output layer neuron,b k is as followskThe threshold of each of the output layer neurons,O k is as followskAn output of the output layer neurons, whereinf 2Activating a function for purelin, which is a linear function;
(3) from the predicted output Ok and the expected output Yk of the BP neural network, the performance metric is expressed using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE), i.e.:
(4) the neural network reduces errors by adjusting the weight and deviation of each iteration, and the updating process of the weight and the threshold is as follows:
wherein the content of the first and second substances,Lfor the number of iterations of the neural network training process,ηis the learning rate;
(5) the error back propagation process is from the output layer to the hidden layer and from the hidden layer to the input layer; when the error is smaller than the preset error, the training process is stopped; updating the weight and the threshold according to a gradient descent principle and a chain type derivation rule; the updating process of the connection weight of the output layer and the hidden layer comprises the following steps:
wherein the content of the first and second substances,eis an errorY k -O k ;
The updating process of the output layer threshold and the hidden layer threshold is as follows:
the updating process of the weight values of the hidden layer and the input layer is as follows:
the hidden layer and input layer threshold value updating process comprises the following steps:
after the neural network is learned through the steps, the output model is as follows:
according to the evaluation method of the present invention, the number of nodes of the input layer is 3, and the number of nodes of the output layer is 1.
The evaluation method according to the present invention, wherein the mutation condition is:
the evaluation method according to the present invention, wherein,HI1 is the time for the voltage to reach the charge cut-off voltage for the first time in the constant current charging mode, and the unit is second;HI2 is the voltage corresponding to the 500 th second in the constant current charging mode;HIand 3 is the current change of the first 1000 seconds in the constant voltage charging mode.
The evaluation method according to the present invention, wherein the linear transformation of the raw data maps the result value between [0,1], and the normalization formula is:
wherein the content of the first and second substances,xis the data in the health index and is,x min andx max the minimum and maximum values in each health indicator curve, respectively.
Compared with the prior art, the evaluation method provided by the invention not only ensures the accuracy and stability of the prediction result, and avoids the problems of gradient disappearance or explosion and the like, but also is beneficial to the rapid convergence of the neural network, and enhances the stability of the network and the learning efficiency of the network.
Detailed Description
The invention will be further illustrated with reference to specific embodiments.
It should be understood that the detailed description of the invention is merely illustrative of the spirit and principles of the invention and is not intended to limit the scope of the invention.
In the embodiment of the invention, the simplified analog energy storage power station is composed of a battery charging and discharging machine, a constant temperature and humidity box, a plurality of lithium iron phosphate batteries and a host machine. The lithium iron phosphate battery is placed in a constant temperature and humidity box, the experiment is guaranteed to be carried out in the environment with the same temperature, then the battery charging and discharging machine is connected through a wire harness, and the upper host records data such as voltage, current and temperature of the battery during charging and discharging in real time.
The factory rated capacity of the lithium iron phosphate battery used in the experiment is 10Ah, the charge cut-off voltage is 3.65V, and the discharge cut-off voltage is 2.6V.
The charge-discharge scheme comprises the following steps: placing the battery in a constant temperature and humidity box, and setting the temperature to be 25 ℃; charging the battery with 1/3C constant current until the charge cut-off voltage is 3.65V; converting constant voltage charging, and standing for 0.5h when the charging current is reduced to 0.01C; entering a constant current discharge stage, performing constant current discharge at 1C, wherein the discharge cut-off voltage is 2.6V, and standing for 10min after the discharge is finished; this is one cycle; if the maximum available capacity of the battery drops below 70% of the rated capacity, the experiment is ended. Otherwise, the above cycle is repeated.
The health indicator HI is extracted from a voltage curve and a current curve of a charging process of the battery, and then the degree of correlation between the extracted health indicator HI and the capacity of the battery is analyzed. Wherein the health indexes are as follows:
HI1 is the time for the voltage to reach the charge cut-off voltage for the first time in the constant current charging mode, and the unit is second;HI2 is the voltage corresponding to the 500 th second in the constant current charging mode;HIand 3 is the current change of the first 1000 seconds in the constant voltage charging mode.
Respectively as follows:
wherein the content of the first and second substances,Iis the flow of electricity, and the temperature of the gas,Vis the voltage of the electric field generated by the electric field generator,mis the time in seconds of the electrical process of the charging process.
Linear transformation of the raw data, mapping the result value between [0,1], normalized formula:
wherein the content of the first and second substances,xis the data in the health index and is,x min andx max the minimum and maximum values in each health indicator curve, respectively.
The degree of correlation of the battery capacity with the health index was analyzed using the following correlation coefficient formula. If the absolute value of the correlation coefficient is greater than 0.6, the two variables are considered strongly correlated.
Wherein the content of the first and second substances,x i is a health index of the human body,x i ’is thatx i The position after the sorting is carried out,y i as the capacity of the battery, there is,y i ’is thaty i The sorted positions, n being the sample size,d i is the difference in the order of the positions,r xyis the correlation coefficient.
The calculation results show that the method has the advantages that,HI1 andHI3 is strongly positive with respect to battery capacity;HI2 is strongly negatively correlated to battery capacity.
The BP neural network is a multilayer feedforward network, has the characteristics of signal forward transmission and error backward propagation, and is one of basic neural network models. Neural networks have achieved great success in recent years, and are widely used for various problems due to their strong generalization ability. The BP neural network can obtain the mapping relation between variables through a training process, and belongs to a typical black box model. In the process of information forward transmission, an input signal is processed layer by layer from an input layer through a hidden layer to an output layer. The hidden layer and the output layer are processing layers with activation functions, if the output of the output layer can not meet the precision requirement, back propagation is carried out according to the prediction error, and therefore the network weight value and the threshold value are adjusted, and the neural network prediction output is enabled to continuously approach the expected output.
The invention selects three layers of neural networks, and the selection of the number of neurons in the hidden layer is based on an empirical formula
Wherein the content of the first and second substances,min order to imply the number of layer nodes,nin order to input the number of nodes of the layer,las the number of nodes of the output layer,a1 to 10
Is constant.
In one specific embodiment, the number of input layer nodes is 3 and the number of output layer nodes is 1.
In a neural network, the information is transferred from the input layer to the hidden layer and from the hidden layer to the output layer. And each layer is output after being processed by the activation function. The output of the hidden layer is:
wherein the content of the first and second substances,Xas input variablesHI1、HI2 andHI3,ω ij is as followsiInput layer neurons andjthe weight of each hidden layer neuron,a j is as followsjThe threshold of each of the hidden layer neurons,θ j in order to have the hidden layer input,H j is as followsjThe output of each of the hidden layer neurons,f 1is a logsig function, which is a hidden layer activation function, i.e. a
the output layer activation function is a linear function, so the output layer output can be expressed as
Wherein the content of the first and second substances,β k the output layer input is predicted for the network,ω jk is as followsjHidden layer neurons andkthe weight of each output layer neuron,b k is as followskThe threshold of each of the output layer neurons,O k is as followskAn output of the output layer neurons, whereinf 2The function is activated for purelin, which is a linear function.
From the predicted output Ok and the expected output Yk of the BP neural network, the performance metric is expressed using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE), i.e.:
the neural network reduces errors by adjusting the weight and deviation of each iteration, and the updating process of the weight and the threshold is as follows:
wherein the content of the first and second substances,Lfor the number of iterations of the neural network training process,ηis the learning rate.
The error is reversely propagated from the output layer to the hidden layer and from the hidden layer to the input layer when the error is less than the preset value
In case of a set error, the training process is stopped. According to gradient descent principle and chain derivation rule, weighting value and threshold value are calculated
And (6) updating. The updating process of the connection weight of the output layer and the hidden layer comprises the following steps:
wherein the content of the first and second substances,eis an errorY k -O k ;
The updating process of the output layer threshold and the hidden layer threshold is as follows:
the updating process of the weight values of the hidden layer and the input layer is as follows:
the hidden layer and input layer threshold value updating process comprises the following steps:
after the neural network is learned through the steps, the output model is as follows:
aiming at the learning rate, the invention provides a neural network model of a dynamic exponential decay learning rate. The whole network has larger learning rate in the early stage of training and can quickly learn from the neural networkLearning battery from health indicatorsRLKey features of (1); in the later training stage, the learning rate is reduced, and the neural network is prevented from falling into an over-fitting state;
meanwhile, the fixed value is kept unchanged in the training process of a certain stage, the stability of the network is enhanced, and the learning of the network is improved
Efficiency. The learning rate is:
wherein the initial learning rate isη 0=0.2,decay_rateIn order to be the rate of decay,global_steprepresenting the current iteration
Number of times, decay rate ofdecay_step=100。
If the initial values of the weight and the threshold of the neural network are not properly selected, the problems of gradient disappearance, explosion and the like in the training process of the neural network can be caused. The method comprises the following specific steps:
the first step is as follows: population initialization, to initial positionX 0And velocityV 0Inertial weightωLearning factorc 1,c 2Initializing parameters;
the second step is that: calculating the fitness value of the particle with the root mean square error as the fitness function
Wherein the content of the first and second substances,pto describe the number of particles of a population size,Y k is as followskThe ideal output value of each particle is calculated,O k is as followskAn
An actual output value of the particle;
the third step: speed update and location update
Wherein the content of the first and second substances,r 1andr 2is [0,1]]A random number in between;V i andX i is the current timeiThe velocity and position of the particle;P ibest is as followsiIndividual optimal solutions for individual particles;G best the optimal solution of the population is obtained;
in order to avoid falling into a local optimal solution, a variation factor is introduced in the processλWhen the variation condition is satisfied,
reinitializing the positions and speeds of all particles in the population, storing the current time information, and generating variation
The conditions of (a) are as follows:
the fourth step: and (4) finishing conditions: outputting when the algorithm reaches the maximum iteration times or the error is smaller than a set value; otherwise, returning to the second step and continuing optimization;
the fifth step: optimizing initial weights of neural networks via particle swarm optimizationω ij Andω jk and a threshold valuea j Andb k (ii) a The model of the neural network output is:
and a sixth step: training the neural network by using exponential decay learning rate, and evaluating the health state of the batterySoHAnd remaining battery lifeRLEvaluation, can obtain
Wherein the content of the first and second substances,T threshold is composed ofSoHThe charge-discharge period corresponding to the failure threshold value is reduced,iin order to train the size of the sample,qis predictedRLThe value of the one or more of the one,Tindicating the number of charge and discharge cycles.
In one embodiment, the training sample size of the battery aging data set is set to K = 100; the remaining life of the batteryRLThe deviation from the true value was about 3.2%.
For convenience of description, each part of the above apparatus is separately described as being functionally divided into various modules or units. Of course, the functionality of the various modules or units may be implemented in the same one or more pieces of software or hardware in practicing the invention.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only intended to illustrate the technical solution of the present invention and not to limit the same, and a person of ordinary skill in the art can make modifications or equivalent substitutions to the specific embodiments of the present invention with reference to the above embodiments, and any modifications or equivalent substitutions which do not depart from the spirit and scope of the present invention are within the protection scope of the present invention as claimed in the appended claims.
Claims (9)
1. An improved method for evaluating the operating life of an energy storage power station comprises the following specific steps:
the first step is as follows: population initialization, to initial positionX 0And velocityV 0Inertial weightωLearning factorc 1,c 2Initializing parameters;
the second step is that: calculating the fitness value of the particles, and taking the root mean square error as a fitness function;
the third step: speed update and location update
Wherein the content of the first and second substances,r 1andr 2is [0,1]]A random number in between;V i andX i is the current timeiThe velocity and position of the particle;P ibest is as followsiIndividual optimal solutions for individual particles;G best the optimal solution of the population is obtained;
in order to avoid falling into a local optimal solution, a variation factor is introduced in the processλWhen the variation condition is satisfied,
reinitializing the positions and the speeds of all particles of the population and storing the current time information;
the fourth step: and (4) finishing conditions: outputting when the algorithm reaches the maximum iteration times or the error is smaller than a set value; otherwise, returning to the second step and continuing optimization;
the fifth step: optimizing initial weights of neural networks via particle swarm optimizationω ij Andω jk and a threshold valuea j Andb k ;
and a sixth step: training the neural network by using exponential decay learning rate, and evaluating the health state of the batterySoHAnd remaining battery lifeRLEvaluation, can obtain
Wherein the content of the first and second substances,T threshold is composed ofSoHThe charge-discharge period corresponding to the failure threshold value is reduced,iin order to train the size of the sample,qis predictedRLThe value of the one or more of the one,Tindicating the number of charge and discharge cycles.
2. The evaluation method of claim 1, wherein the fitness function is:
wherein the content of the first and second substances,pto describe the number of particles of a population size,Y k is as followskThe ideal output value of each particle is calculated,O k is as followskAn
The actual output value of the particle.
5. The evaluation method of claim 1, wherein the model of the neural network output is derived as follows:
(1) selecting three-layer neural network, selecting the number of neurons in hidden layer according to empirical formula
Wherein the content of the first and second substances,min order to imply the number of layer nodes,nin order to input the number of nodes of the layer,las the number of nodes of the output layer,a1 to 10
A constant of (d);
(2) in the neural network, the forward transfer process of information is from an input layer to a hidden layer, from the hidden layer to an output layer, and each layer is output after being processed by an activation function; the output of the hidden layer is:
wherein the content of the first and second substances,Xas input variablesHI1、HI2 andHI3,ω ij is as followsiInput layer neurons andjthe weight of each hidden layer neuron,a j is as followsjThe threshold of each of the hidden layer neurons,θ j in order to have the hidden layer input,H j is as followsjThe output of each of the hidden layer neurons,f 1is a logsig function, which is a hidden layer activation function, i.e. a
the output layer activation function is a linear function, so the output layer output can be expressed as
Wherein the content of the first and second substances,β k the output layer input is predicted for the network,ω jk is as followsjHidden layer neurons andkthe weight of each output layer neuron,b k is as followskThe threshold of each of the output layer neurons,O k is as followskAn output of the output layer neurons, whereinf 2Activating a function for purelin, which is a linear function;
(3) from the predicted output Ok and the expected output Yk of the BP neural network, the performance metric is expressed using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE), i.e.:
(4) the neural network reduces errors by adjusting the weight and deviation of each iteration, and the updating process of the weight and the threshold is as follows:
wherein the content of the first and second substances,Lfor the number of iterations of the neural network training process,ηis the learning rate;
(5) the error back propagation process is from the output layer to the hidden layer and from the hidden layer to the input layer; when the error is smaller than the preset error, the training process is stopped; updating the weight and the threshold according to a gradient descent principle and a chain type derivation rule; the updating process of the connection weight of the output layer and the hidden layer comprises the following steps:
wherein the content of the first and second substances,eis an errorY k -O k ;
The updating process of the output layer threshold and the hidden layer threshold is as follows:
the updating process of the weight values of the hidden layer and the input layer is as follows:
the hidden layer and input layer threshold value updating process comprises the following steps:
after the neural network is learned through the steps, the output model is as follows:
6. the evaluation method according to claim 5, wherein the number of input layer nodes is 3 and the number of output layer nodes is 1.
8. the evaluation method according to claim 5,HI1 is the time for the voltage to reach the charge cut-off voltage for the first time in the constant current charging mode, and the unit is second;HI2 is the voltage corresponding to the 500 th second in the constant current charging mode;HIand 3 is the current change of the first 1000 seconds in the constant voltage charging mode.
9. The evaluation method of claim 5, wherein the linear transformation of the raw data maps the result values between [0,1], with a normalization formula of:
wherein the content of the first and second substances,xis the data in the health index and is,x min andx max the minimum and maximum values in each health indicator curve, respectively.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111487606.9A CN114091975B (en) | 2021-12-08 | 2021-12-08 | Method for evaluating operation life of energy storage power station |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111487606.9A CN114091975B (en) | 2021-12-08 | 2021-12-08 | Method for evaluating operation life of energy storage power station |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114091975A true CN114091975A (en) | 2022-02-25 |
CN114091975B CN114091975B (en) | 2022-06-21 |
Family
ID=80306829
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111487606.9A Active CN114091975B (en) | 2021-12-08 | 2021-12-08 | Method for evaluating operation life of energy storage power station |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114091975B (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106597311A (en) * | 2016-12-21 | 2017-04-26 | 哈尔滨工业大学 | Fused HI equivalence lithium ion battery degradation prediction method based on principal component analysis |
US20180143259A1 (en) * | 2016-11-22 | 2018-05-24 | Samsung Electronics Co., Ltd. | Method and apparatus for estimating state of battery |
CN108549032A (en) * | 2018-04-17 | 2018-09-18 | 北京智行鸿远汽车有限公司 | A kind of evaluation method of cell health state SOH |
US20190157891A1 (en) * | 2017-11-20 | 2019-05-23 | The Trustees Of Columbia University In The City Of New York | Neural-network state-of-charge estimation |
CN110687450A (en) * | 2019-08-28 | 2020-01-14 | 武汉科技大学 | Lithium battery residual life prediction method based on phase space reconstruction and particle filtering |
CN111948560A (en) * | 2020-07-30 | 2020-11-17 | 西安工程大学 | Lithium battery health state estimation method based on multi-factor evaluation model |
CN113094981A (en) * | 2021-03-28 | 2021-07-09 | 西北工业大学 | Lithium ion battery reliability evaluation method based on grey neural network model and self-service method |
CN113740736A (en) * | 2021-08-31 | 2021-12-03 | 哈尔滨工业大学 | Electric vehicle lithium battery SOH estimation method based on deep network self-adaptation |
-
2021
- 2021-12-08 CN CN202111487606.9A patent/CN114091975B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180143259A1 (en) * | 2016-11-22 | 2018-05-24 | Samsung Electronics Co., Ltd. | Method and apparatus for estimating state of battery |
CN106597311A (en) * | 2016-12-21 | 2017-04-26 | 哈尔滨工业大学 | Fused HI equivalence lithium ion battery degradation prediction method based on principal component analysis |
US20190157891A1 (en) * | 2017-11-20 | 2019-05-23 | The Trustees Of Columbia University In The City Of New York | Neural-network state-of-charge estimation |
CN108549032A (en) * | 2018-04-17 | 2018-09-18 | 北京智行鸿远汽车有限公司 | A kind of evaluation method of cell health state SOH |
CN110687450A (en) * | 2019-08-28 | 2020-01-14 | 武汉科技大学 | Lithium battery residual life prediction method based on phase space reconstruction and particle filtering |
CN111948560A (en) * | 2020-07-30 | 2020-11-17 | 西安工程大学 | Lithium battery health state estimation method based on multi-factor evaluation model |
CN113094981A (en) * | 2021-03-28 | 2021-07-09 | 西北工业大学 | Lithium ion battery reliability evaluation method based on grey neural network model and self-service method |
CN113740736A (en) * | 2021-08-31 | 2021-12-03 | 哈尔滨工业大学 | Electric vehicle lithium battery SOH estimation method based on deep network self-adaptation |
Non-Patent Citations (1)
Title |
---|
邓伟锋等: "基于GA优化BP神经网络的微电网蓄电池健康状态评估", 《电测与仪表》 * |
Also Published As
Publication number | Publication date |
---|---|
CN114091975B (en) | 2022-06-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Hannan et al. | Neural network approach for estimating state of charge of lithium-ion battery using backtracking search algorithm | |
CN111709186B (en) | Integrated estimation method for health state of retired power lithium battery | |
CN109143105A (en) | A kind of state-of-charge calculation method of lithium ion battery of electric automobile | |
Che et al. | SOC and SOH identification method of li-ion battery based on SWPSO-DRNN | |
CN111366848A (en) | Battery health state prediction method based on PSO-ELM algorithm | |
CN115632179B (en) | Intelligent quick charging method and system for lithium ion battery | |
Zhang et al. | Intelligent state of charge estimation of battery pack based on particle swarm optimization algorithm improved radical basis function neural network | |
CN106501728A (en) | A kind of battery equivalent model parameter identification method based on multi-objective genetic algorithm | |
Boujoudar et al. | Lithium-ion batteries modeling and state of charge estimation using artificial neural network | |
CN106777786A (en) | A kind of lithium ion battery SOC estimation method | |
Dong et al. | Estimation of power battery SOC based on improved BP neural network | |
CN107132490A (en) | A kind of method for realizing the estimation of lithium battery group state-of-charge | |
Dou et al. | Extreme learning machine model for state-of-charge estimation of lithium-ion battery using salp swarm algorithm | |
Geng et al. | State of charge estimation method for lithium-ion batteries in all-electric ships based on LSTM neural network | |
Zhao et al. | The li-ion battery state of charge prediction of electric vehicle using deep neural network | |
CN104090238B (en) | Lithium ion battery residual life Forecasting Methodology based on AR integrated study model | |
Yang et al. | A new self-learning TLBO algorithm for RBF neural modelling of batteries in electric vehicles | |
Elamine et al. | Multi-agent system based on fuzzy control and prediction using NN for smart microgrid energy management | |
Eleftheriadis et al. | Hyperband optimization of stacked bidirectional long short-term memory neural network for the state of charge estimation | |
CN110232432B (en) | Lithium battery pack SOC prediction method based on artificial life model | |
Jusoh et al. | Accurate battery model parameter identification using heuristic optimization | |
Zhang et al. | Research for SOC prediction of lithium battery based on GA-ESN | |
Fan et al. | An adaptive neuro-fuzzy inference system (ANFIS) based model for the temperature prediction of lithium-ion power batteries | |
CN114091975B (en) | Method for evaluating operation life of energy storage power station | |
Piao et al. | An improved model based on artificial neural networks and Thevenin model for nickel metal hydride power battery |
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 |