CN117150334A - Lithium battery multi-condition prediction method and device based on optimized BiLSTM neural network - Google Patents

Lithium battery multi-condition prediction method and device based on optimized BiLSTM neural network Download PDF

Info

Publication number
CN117150334A
CN117150334A CN202310713306.0A CN202310713306A CN117150334A CN 117150334 A CN117150334 A CN 117150334A CN 202310713306 A CN202310713306 A CN 202310713306A CN 117150334 A CN117150334 A CN 117150334A
Authority
CN
China
Prior art keywords
individual
lithium battery
neural network
particle
pigeon
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310713306.0A
Other languages
Chinese (zh)
Inventor
陈立平
宋英杰
丁纪宇
赖振伟
刘创
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hefei University of Technology
Original Assignee
Hefei University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hefei University of Technology filed Critical Hefei University of Technology
Priority to CN202310713306.0A priority Critical patent/CN117150334A/en
Publication of CN117150334A publication Critical patent/CN117150334A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
    • G01R31/385—Arrangements for measuring battery or accumulator variables
    • G01R31/387—Determining ampere-hour charge capacity or SoC
    • G01R31/388—Determining ampere-hour charge capacity or SoC involving voltage measurements
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
    • G01R31/367—Software therefor, e.g. for battery testing using modelling or look-up tables
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
    • G01R31/389—Measuring internal impedance, internal conductance or related variables
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • G06F18/15—Statistical pre-processing, e.g. techniques for normalisation or restoring missing data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • 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
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • 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
    • G06N3/08—Learning methods
    • G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Probability & Statistics with Applications (AREA)
  • Physiology (AREA)
  • Genetics & Genomics (AREA)
  • Feedback Control In General (AREA)

Abstract

The invention discloses a lithium battery multi-condition prediction method and device based on an optimized BiLSTM neural network, wherein the method combines an improved pigeon swarm algorithm with a genetic algorithm, and effectively improves the optimizing capability of the pigeon swarm algorithm; the super-parameters of the BiLSTM are utilized to carry out parameter optimization, so that the artificial setting of the super-parameters of the neural network is avoided, the time is saved, the optimal super-parameters can be tested, an accurate neural network model is built, and the network accurately realizes the prediction of the SOC and SOE of the lithium battery; the square root unscented Kalman filter is used to filter the output, further improving accuracy. The method has the advantages of wide application range, high prediction precision and the like, simultaneously effectively reduces the difficulty of manually adjusting the parameters of the neural network, and effectively improves the accuracy of the model.

Description

Lithium battery multi-condition prediction method and device based on optimized BiLSTM neural network
Technical Field
The invention relates to the technical field of lithium batteries, in particular to a lithium battery multi-condition prediction method and device based on an optimized BiLSTM neural network.
Background
The traditional fuel oil automobile not only produces a large amount of greenhouse gases in running, but also can discharge various fine pollutants, and brings great trouble to human beings and the environment. Electric vehicles with no pollution and high energy efficiency are popular with people. As a power source of an electric vehicle, lithium batteries are widely used due to their long life, large capacity, and the like. In order to ensure the normal operation of lithium batteries, the battery management system needs to accurately monitor numerous parameters. Among them, accurate state-of-charge (SOC) and state-of-energy (SOE) are critical to the reliability of the battery management system. However, lithium batteries have extremely strong non-linear and time-varying characteristics, which makes direct measurement of SOC and SOE difficult.
To date, the research results of students on SOC and SOE estimation methods are mainly divided into two major categories: model-based methods and data-driven methods. Model-based methods require the preferential establishment of a suitable battery model. However, the accuracy of the estimation of the model-based approach depends largely on the quality of the battery model. Unfortunately, it is practically difficult to construct an accurate voltage model due to variations in the internal resistance of the battery. The data-driven approach does not need to take into account complex electrochemical reactions inside the battery nor does it need to identify model parameters, but rather self-learn the intrinsic relationship of input (current, voltage and temperature) and output (like SOC or SOE). However, in general, recurrent neural networks can only analyze and learn information in battery data from forward directions, and part of useful information is lost in forward transmission, so that the prediction effect of the network is affected. Furthermore, after the network type determination, whether or not an appropriate hyper-parameter can be selected will determine the final prediction accuracy of the network. In the above papers, the neural network super-parameter setting is artificial setting, which requires repeated experiments, consumes time and does not necessarily experiment the optimal super-parameter, and the final prediction accuracy is not high.
Disclosure of Invention
In order to solve the technical problems in the background technology, the invention provides a lithium battery multi-condition prediction method based on an optimized BiLSTM neural network.
The invention provides a lithium battery multi-condition prediction method based on an optimized BiLSTM neural network, which is characterized in that an improved pigeon group-genetic algorithm is utilized to identify the super parameters of a BiLSTM (two-way long and short word memory model) neural network model of a lithium battery, then the obtained optimal parameters are fed back into the BiLSTM neural network model of the lithium battery, the charge state of the lithium battery and the residual energy state of the lithium battery are estimated at the same time, and the optimized neural network is output to a square root unscented Kalman filter.
Optionally, the method comprises the following steps:
s1: charging and discharging the lithium battery under different temperatures and driving periods, sampling the lithium battery in the discharging cycle, forming a lithium battery discharging data set from the sampled data, and carrying out normalization treatment;
s2: acquiring identification parameters of a lithium battery BiLSTM neural network: the identification parameters of the lithium battery BiLSTM neural network are as follows: the number ls, the initial learning rate lr, the maximum iteration number ep, the learning rate decline factor lrdf and the number fls of the full-connection layer neurons in the super parameters, namely, in an optimization algorithm, the dimension of a target search space is 5, and the number ls, the initial learning rate lr, the maximum iteration number ep, the learning rate decline factor lrdf and the number fls of the full-connection layer neurons are respectively recorded as an identification parameter 1, an identification parameter 2, an identification parameter 3, an identification parameter 4 and an identification parameter 5;
regarding each identification parameter in the identification parameters as one dimension of a particle individual, wherein the particle individual has two attributes of a particle individual position and a particle individual speed, the particle individual position represents the moving direction, and the particle individual speed represents the moving speed, and the particle individual position is the value of the identification parameter;
step 3, setting parameters of an improved pigeon group-genetic hybrid algorithm:
setting the dimension of the target search space as D, d=5; setting a population consisting of N particles; setting the maximum iteration number M; setting the coding length L of each chromosome; the crossover probability p1 is set.
Setting the position range of the particle individual as { x_min, x_max } and the speed range { v_min, v_max } of the particle individual;
wherein x_min is the minimum value of the individual particle position, x_max is the maximum value of the individual particle position, v_min is the minimum value of the individual particle velocity, and v_max is the maximum value of the individual particle velocity;
any one of the N particle individuals is marked as a particle individual i, i is the serial number of any one particle individual in the population, i=1, 2,..N, and the position vector of the particle individual i is marked as an individual position X i The velocity vector of the individual particle i is denoted as the individual velocity V i The expressions are as follows:
X i ＝{x i1 ，x i2 ，...，x ij ...，x iD }
V i ＝{v i1 ，v i2 ，...v ij ...，v iD }
wherein the individual position X i Solution for a group of identification parameters of lithium battery BiLSTM neural network model, x ij J=1, 2, ·d for the j-th dimension position of the individual particle i; individual speed V i Speed, v, of solution of a set of parameters for lithium battery equivalent circuit model in particle search solution space ij The j-th dimension velocity for individual particles i;
adaptation of individual particles iThe degree value is recorded as an individual fitness value f i The position with the minimum fitness value searched by the whole particle swarm is recorded as a global optimal position X g ；
Step 4, optimizing an iteration flow of the BiLSTM neural network:
step 4.1, initializing the individual position and the individual speed;
step 4.2, updating the individual position and the individual speed;
step 4.3, crossing and mutating, calculating a fitness value, and updating global optimum;
step 4.4, comparing the iteration times: if the current iteration number t > N C1 Turning to step 4.5; otherwise, turning to step 4.1;
step 4.5, finding the center of the pigeon cluster, and updating the position of the pigeon cluster;
step 4.6, judging if the current iteration time t=n C2 Ending the iteration and outputting the current global optimal position X g The group of identification parameters corresponding to the global final position is the optimal group of parameters of the lithium battery BiLSTM neural network;
if t is less than N C2 Returning to the step 4.5 for the next iteration;
step 5, the square root unscented Kalman filter filters the output:
and outputting the predicted result of the optimized BiLSTM neural network to a square root unscented Kalman filter, and filtering the output by the square root unscented Kalman filter.
Optionally, in step S4.1:
initializing the speed and position of pigeon flocks, randomly generating population individuals x= (ls, fls, ep, lr, lrdf), and using the input data x processed in the step 1 n And output data y n Training BiLSTM and calculating fitness value, selecting Mean Square Error (MSE) as loss function of BiLSTM and also as fitness function of optimization algorithm, wherein the fitness function is as follows:
wherein SOC '(t) and SOE' (t) are x n And inputting a prediction result of the BiLSTM network, wherein SOE (t) and SOC (t) are true values.
Optionally, in step S4.2: the updating steps of the individual position and the individual speed are as follows:
the speed and position of the pigeon flock are updated according to the following formula:
V i (t)＝V i (t-1)*e -w +rand*(X g -X i (t-1))
X i (t)＝X i (t-1)+V i (t)
wherein R is min 、R max R is minimum and maximum, 0.2 and 2.2, N respectively c1 The iteration number in the compass operator is represented by w, the inertia weight and X i (t)、V i (t) represents the position and speed of the ith pigeon at the t iteration, rand is a random number of 0-1, t is the number of iterations, X g Is the optimal pigeon position in the pigeon group.
Optionally, in step 4.3, the method specifically includes:
crossing and mutating the individual to strengthen the ability of jumping out of the local optimum, and calculating the fitness value f of the particle individual i in the current iteration i . The fitness value corresponding to the individual position is matched with the full, optimal position X g Corresponding fitness value f g_best The following judgment is made:
comparison f i And f g_best Taking the position with the smallest fitness value as the current global optimal position X g And the current global optimal position X g The corresponding fitness value is marked as the current global optimal fitness value f g_best ；
Optionally, step 4.5, finding the center of the pigeon cluster, and updating the position of the pigeon cluster comprises the following steps:
the pigeon clusters are sequenced according to the fitness value, and the central position of the pigeon clusters is determined according to the following formula:
N C2 ＝Tmax-N C1
where Tmax is the maximum number of iterations. Np (t) is the number of t iteration pigeons, X c ( t) For the central position of the t iteration pigeon group, fitness is the fitness value of the i pigeon, X i (t) represents the position of the ith pigeon, and N is the population number;
and updating the position of the pigeon group according to the following formula, calculating the fitness value and updating the global optimum, wherein the formula is as follows:
X i (t)＝X i (t-1)+rand*(X c (t)-X i (t-1))
specifically, the step S1 specifically includes the steps of:
s10: extracting the following lithium battery sampling parameters related to the lithium battery charge state and the residual energy state from the discharge data set, wherein the parameters comprise sampling end voltage in a discharge cycle, sampling end current in the discharge cycle, sampling lithium battery temperature in the discharge cycle, sampling lithium battery charge state and residual energy in the discharge cycle, and constructing a neural network input matrix x n And output data matrix y n ：
y n ＝[SOC n SOC n ]；
Wherein x is n The input matrix is BiLSTM at the time of n, and the size of the input matrix is m; i n 、V n 、T n Respectively measuring the current, voltage and temperature data of the battery at the moment n, wherein the SOC is the state of charge of the lithium battery, and the SOE is the state of residual energy; y is n The SOC and SOE values output by the layer are output for the n times,the matrix size is 1 multiplied by 2;
s10: normalizing the input matrix data and the output matrix data, wherein the three input variables V, I and T have different dimensionalities, and mapping the original data to intervals [0,1] by adopting a minimum-maximum normalization method in order to eliminate adverse effects; the normalization formula is as follows:
wherein x is the original data; x is x max And x min Respectively the maximum value and the minimum value of the original data; x is x * Is a normalized value.
An optimized BiLSTM neural network-based lithium battery multi-condition prediction device, comprising: at least one processor and memory;
the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored by the memory to perform the method as described above.
In the invention, the lithium battery multi-condition prediction method and device based on the optimized BiLSTM neural network have the following advantages:
1. an improved pigeon colony-genetic algorithm is provided, which can effectively optimize the hyper-parameters of BiLSTM neural network. The method avoids the complex parameter adjusting process of the traditional neural network, is simple and easy to operate, and can effectively improve the network performance.
2. The used optimized BiLSTM neural network estimates two battery states simultaneously, so that one network can estimate two states accurately and simultaneously, and the network efficiency is improved.
3. The prediction result of the optimized BiLSTM neural network is filtered by a square root unscented Kalman filter, so that the estimation accuracy is further improved.
Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
FIG. 1 is an overall block diagram constructed in the lithium battery multi-condition high-precision prediction method based on an optimized BiLSTM neural network;
FIG. 2 is a flowchart of an optimized neural network of the lithium battery multi-condition high-precision prediction method based on the optimized BiLSTM neural network;
FIG. 3 is a graph comparing the predicted results of the state of a BiLSTM neural network lithium battery with the state of an un-optimized and filtered state under the training of the same data set in an embodiment of the present invention;
FIG. 4 is a graph comparing the prediction error results of the embodiment of the present invention with the un-optimized and filtered BiLSTM neural network lithium battery condition under the same data set training.
Detailed Description
Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar symbols indicate like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are exemplary only for explaining the present invention and are not to be construed as limiting the present invention.
Fig. 1 is an overall construction diagram of the present invention. As can be seen from fig. 1, the present invention provides a method for identifying super parameters of a lithium battery BiLSTM neural network model, wherein the identification method utilizes an improved pigeon cluster-genetic algorithm to identify the super parameters of the lithium battery BiLSTM neural network model, then feeds back the obtained optimal parameters to the lithium battery BiLSTM neural network model, and outputs the optimized neural network to a square root unscented kalman filter. The method comprises the following specific steps:
step 1, charging and discharging the lithium battery at different temperatures and driving periods, sampling the lithium battery in the discharging cycle, forming sampling data into a lithium battery discharging data set, and carrying out normalization treatment;
step 1.1 extracts the following lithium battery sampling parameters from the discharge dataset that relate to the state of charge and the remaining energy condition of the lithium battery: sampling end voltage in discharge cycle, sampling end current in discharge cycle, sampling lithium battery temperature in discharge cycle, sampling lithium battery charge state in discharge cycle and residual energyAmount of the components. The parameters construct the neural network input matrix x n And output data matrix y n . The expression is as follows:
y n ＝[SOC n SOC n ]
wherein x is n The input matrix is the BiLSTM of time n, and the size is m. I n 、V n 、T n Battery current, voltage and temperature data measured for time n. SOC is the lithium battery state of charge and SOE is the remaining energy condition. y is n The matrix size of the SOC and SOE values output by the output layer at the time n is 1×2.
And 1.2, normalizing the input matrix and the output matrix data. The three input variables V, I, T have different dimensions, and in order to eliminate adverse effects, the raw data is mapped to the interval [0,1] using a min-max normalization method. The normalization formula is as follows:
wherein x is the original data; x is x max And x min Respectively the maximum value and the minimum value of the original data; x is x * Is a normalized value.
And 2, acquiring identification parameters of the lithium battery BiLSTM neural network. The identification parameters of the lithium battery BiLSTM neural network are as follows: the number ls, the initial learning rate lr, the maximum iteration number ep, the learning rate decline factor lrdf and the number fls of the full-connection layer neurons in the super parameters, namely, in an optimization algorithm, the dimension of a target search space is 5, and the number ls, the initial learning rate lr, the maximum iteration number ep, the learning rate decline factor lrdf and the number fls of the full-connection layer neurons are respectively recorded as an identification parameter 1, an identification parameter 2, an identification parameter 3, an identification parameter 4 and an identification parameter 5;
regarding each identification parameter in the identification parameters as one dimension of a particle individual, wherein the particle individual has two attributes of a particle individual position and a particle individual speed, the particle individual position represents the moving direction, and the particle individual speed represents the moving speed, and the particle individual position is the value of the identification parameter;
and 3, setting parameters of an improved pigeon colony-genetic hybrid algorithm.
Setting the dimension of the target search space as D, d=5; setting a population consisting of N particles; setting the maximum iteration number M; setting the coding length L of each chromosome; the crossover probability p1 is set.
Setting the position range of the particle individual as { x_min, x_max }, and the speed range of the particle individual as { v_min, v_max }, wherein x_min is the minimum value of the position of the particle individual, x_max is the maximum value of the position of the particle individual, v_min is the minimum value of the speed of the particle individual, and v_max is the maximum value of the speed of the particle individual;
any one of the N particle individuals is marked as a particle individual i, i is the serial number of any one particle individual in the population, i=1, 2,..N, and the position vector of the particle individual i is marked as an individual position X i The velocity vector of the individual particle i is denoted as the individual velocity V i The expressions are as follows:
X i ＝{x i1 ，x i2 ，...，x ij ...，x iD }
V i ＝{v i1 ，v i2 ，...v ij …，v iD }
wherein the individual position X i Solution for a group of identification parameters of lithium battery BiLSTM neural network model, x ij J=1, 2, ·d for the j-th dimension position of the individual particle i; individual speed V i Speed, v, of solution of a set of parameters for lithium battery equivalent circuit model in particle search solution space ij The j-th dimension velocity for individual particles i;
the fitness value of the particle individual i is recorded as an individual fitness value f i The position with the minimum fitness value searched by the whole particle swarm is recorded as a global optimal position X g 。
Step 4, optimizing the iterative flow of the BiLSTM neural network, and FIG. 2 is a flow chart of the optimizing process of the invention.
Step 4.1, initializing the individual position and the individual speed.
Initializing the speed and position of pigeon flocks, randomly generating population individuals x= (ls, fls, ep, lr, lrdf), and using the input data x processed in the step 1 n And output data y n Training BiLSTM and calculating fitness value, selecting Mean Square Error (MSE) as loss function of BiLSTM and also as fitness function of optimization algorithm, wherein the fitness function is as follows:
wherein SOC '(t) and SOE' (t) are x n Inputting a prediction result of the BiLSTM network, wherein SOE (t) and SOC (t) are true values;
step 4.2, updating the individual position and the individual speed: the speed and position of the pigeon flock are updated according to the following formula:
V i (t)＝V i (t-1)*e -w +rand*(X g -X i (t-1))
X i (t)＝X i (t-1)+V i (t)
wherein R is min 、R max R is minimum and maximum, 0.2 and 2.2, N respectively c1 Is the number of iterations in the compass operator. w is inertial weight, X i (t)、V i (t) represents the position and speed of the ith pigeon at the t iteration, rand is a random number of 0-1, t is the number of iterations, X g Is the optimal pigeon position in the pigeon group.
And 4.3, crossing and mutating, calculating the fitness value, and updating the global optimum.
And (3) crossing and mutating the individuals to strengthen the ability of jumping out of local optimum. Computing adaptation of individual particles i in the current iterationDegree value f i (f i For the calculation of the fitness function in step 4.1), the fitness value corresponding to the individual position is calculated with the global optimal position X g Corresponding fitness value f g_best The following judgment is made:
comparison f i And f g_best Taking the position with the smallest fitness value as the current global optimal position X g And the current global optimal position X g The corresponding fitness value is marked as the current global optimal fitness value f g_best ；
Step 4.4, comparing the iteration times,
if the current iteration number t > N C1 Turning to step 4.5; otherwise, go to step 4.1.
Step 4.5, finding the center of the pigeon cluster, and updating the position of the pigeon cluster
The pigeon clusters are sequenced according to the fitness value, and the central position of the pigeon clusters is determined according to the following formula:
N C2 ＝Tmax-N C1
where Tmax is the maximum number of iterations. Np (t) is the number of t iteration pigeons, X c (t) is the central position of the t iterative pigeon group, fitness is the fitness value of the i pigeon, X i (t) represents the position of the ith pigeon, and N is the population number.
And updating the position of the pigeon group according to the following formula, calculating the fitness value and updating the global optimum. The formula is as follows:
X i (t)＝X i (t-1)+rand*(X c (t)-X i (t-1))
step 4.6, the following judgment is carried out:
if the current iteration number t=n C2 Ending the iteration and outputting the current global optimal position X g The group of identification parameters corresponding to the global final position is the optimal group of parameters of the lithium battery BiLSTM neural network;
if t is less than N C2 Returning to the step 4.5 for the next iteration.
And 5, filtering the output by a square root unscented Kalman filter.
And outputting the predicted result of the optimized BiLSTM neural network to a square root unscented Kalman filter, and filtering the output by the square root unscented Kalman filter to further improve the estimation accuracy.
To verify the effect of the present invention, simulations were performed.
The lithium battery used in the simulation is a loose INR18650-20R power lithium battery, and the lithium battery working condition data set is composed of the lithium battery UDDS working conditions.
The experimental results of the simulation are shown in fig. 3 and 4. Fig. 3 uses Root Mean Square Error (RMSE) of SOC and SOE as an index for verifying accuracy of the recognition result. The smaller the RMSE, the higher the recognition accuracy. Wherein BiLSTM is a traditional neural network algorithm, and PG-BiLSTM-SAUKF is a result of optimizing BiLSTM post-filtering by a pigeon population-genetic optimization algorithm in the invention. The invention provides the prediction result of the BiLSTM algorithm, and the curves of the SOC and the SOE in the prediction result and the actual measured value of the PG-BiLSTM-SAUKF algorithm, so that the result of the PG-BiLSTM-SAUKF algorithm is obviously closer to the actual value.
FIG. 4 is a graph showing the prediction error of the lithium battery in the embodiment of the present invention, wherein the comparison of the present invention shows the curves of the prediction error result of the BiLSTM algorithm and the prediction error result of the PG-BiLSTM-SAUKF algorithm, and it is obvious that the PG-BiLSTM-SAUKF algorithm of the present invention has smaller error.
An optimized BiLSTM neural network-based lithium battery multi-condition prediction device, comprising: at least one processor and memory;
the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored by the memory to perform the method as described above.
The foregoing is only a preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art, who is within the scope of the present invention, should make equivalent substitutions or modifications according to the technical scheme of the present invention and the inventive concept thereof, and should be covered by the scope of the present invention.

Claims (8)

1. A lithium battery multi-condition prediction method based on an optimized BiLSTM neural network is characterized in that an improved pigeon group-genetic algorithm is utilized to identify super parameters of a lithium battery BiLSTM neural network model, then the obtained optimal parameters are fed back into the lithium battery BiLSTM neural network model, the state of charge of the lithium battery and the residual energy of the lithium battery are estimated at the same time, and the optimized neural network is output to a square root unscented Kalman filter.
2. The lithium battery multi-condition prediction method based on the optimized BiLSTM neural network of claim 1, comprising the steps of:
s1: charging and discharging the lithium battery under different temperatures and driving periods, sampling the lithium battery in the discharging cycle, forming a lithium battery discharging data set from the sampled data, and carrying out normalization treatment;
s2: acquiring identification parameters of a lithium battery BiLSTM neural network: the identification parameters of the lithium battery BiLSTM neural network are as follows: the number ls, the initial learning rate lr, the maximum iteration number ep, the learning rate decline factor lrdf and the number fls of the full-connection layer neurons in the super parameters, namely, in an optimization algorithm, the dimension of a target search space is 5, and the number ls, the initial learning rate lr, the maximum iteration number ep, the learning rate decline factor lrdf and the number fls of the full-connection layer neurons are respectively recorded as an identification parameter 1, an identification parameter 2, an identification parameter 3, an identification parameter 4 and an identification parameter 5;
regarding each identification parameter in the identification parameters as one dimension of a particle individual, wherein the particle individual has two attributes of a particle individual position and a particle individual speed, the particle individual position represents the moving direction, and the particle individual speed represents the moving speed, and the particle individual position is the value of the identification parameter;
step 3, setting parameters of an improved pigeon group-genetic hybrid algorithm:
setting the dimension of the target search space as D, d=5; setting a population consisting of N particles; setting the maximum iteration number M; setting the coding length L of each chromosome; the crossover probability p1 is set.
Setting the position range of the particle individual as { x_min, x_max } and the speed range { v_min, v_max } of the particle individual;
wherein x_min is the minimum value of the individual particle position, x_max is the maximum value of the individual particle position, v_min is the minimum value of the individual particle velocity, and v_max is the maximum value of the individual particle velocity;
any one of the N particle individuals is marked as a particle individual i, i is the serial number of any one particle individual in the population, i=1, 2,..N, and the position vector of the particle individual i is marked as an individual position X i The velocity vector of the individual particle i is denoted as the individual velocity V i The expressions are as follows:
X i ＝{x i1 ，x i2 ，...，x ij ...，x iD }
V i ＝{v i1 ，v i2 ，...v ij ...，v iD }
wherein the individual position X i Solution for a group of identification parameters of lithium battery BiLSTM neural network model, x ij J=1, 2, ·d for the j-th dimension position of the individual particle i; individual speed V i Speed, v, of solution of a set of parameters for lithium battery equivalent circuit model in particle search solution space ij The j-th dimension velocity for individual particles i;
the fitness value of the particle individual i is recorded as an individual fitness value f i The position with the minimum fitness value searched by the whole particle swarm is recorded as a global optimal position X g ；
Step 4, optimizing an iteration flow of the BiLSTM neural network:
step 4.1, initializing the individual position and the individual speed;
step 4.2, updating the individual position and the individual speed;
step 4.3, crossing and mutating, calculating a fitness value, and updating global optimum;
step 4.4, comparing the iteration times: if the current iteration number t > N C1 Turning to step 4.5; otherwise, turning to step 4.1;
step 4.5, finding the center of the pigeon cluster, and updating the position of the pigeon cluster;
step 4.6, judging if the current iteration time t=n C2 Ending the iteration and outputting the current global optimal position X g The group of identification parameters corresponding to the global final position is the optimal group of parameters of the lithium battery BiLSTM neural network;
if t is less than N C2 Returning to the step 4.5 for the next iteration;
step 5, the square root unscented Kalman filter filters the output:
and outputting the predicted result of the optimized BiLSTM neural network to a square root unscented Kalman filter, and filtering the output by the square root unscented Kalman filter.
3. The lithium battery multi-condition prediction method based on the optimized BiLSTM neural network according to claim 2, wherein in step S4.1:
initializing the speed and position of pigeon flocks, randomly generating population individuals x= (ls, fls, ep, lr, lrdf), and using the input data x processed in the step 1 n And output data y n Training BiLSTM and calculating fitness value, selecting Mean Square Error (MSE) as loss function of BiLSTM and also as fitness function of optimization algorithm, wherein the fitness function is as follows:
wherein SOC '(t) and SOE' (t) are x n And inputting a prediction result of the BiLSTM network, wherein SOE (t) and SOC (t) are true values.
4. The lithium battery multi-condition prediction method based on the optimized BiLSTM neural network according to claim 2, wherein in step S4.2: the updating steps of the individual position and the individual speed are as follows:
the speed and position of the pigeon flock are updated according to the following formula:
V i (t)＝V i (t-1)*e -w +rand*(X g -X i (t-1))
X i (t)＝X i (t-1)+V i (t)
wherein R is min 、R max R is minimum and maximum, 0.2 and 2.2, N respectively c1 The iteration number in the compass operator is represented by w, the inertia weight and X i (t)、V i (t) represents the position and speed of the ith pigeon at the t iteration, rand is a random number of 0-1, t is the number of iterations, X g Is the optimal pigeon position in the pigeon group.
5. The lithium battery multi-condition prediction method based on the optimized BiLSTM neural network according to claim 2, wherein in step 4.3, specifically:
crossing and mutating the individual to strengthen the ability of jumping out of the local optimum, and calculating the fitness value f of the particle individual i in the current iteration i . The fitness value corresponding to the individual position is matched with the full, optimal position X g Corresponding fitness value f g_best The following judgment is made:
comparison f i And f g_best Taking the position with the smallest fitness value as the current global optimal position X g And the current global optimal position X g The corresponding fitness value is marked as the current global optimal fitness value f g_best ；
6. The method for predicting the multiple conditions of the lithium battery based on the optimized BiLSTM neural network according to claim 2, wherein the step 4.5 of finding the center of the pigeon cluster and updating the position of the pigeon cluster comprises the steps of:
the pigeon clusters are sequenced according to the fitness value, and the central position of the pigeon clusters is determined according to the following formula:
N C2 ＝Tmax-N C1
where Tmax is the maximum number of iterations. Np (t) is the number of t iteration pigeons, X c (t) is the central position of the t iterative pigeon group, fitness is the fitness value of the i pigeon, X i (t) represents the position of the ith pigeon, and N is the population number;
and updating the position of the pigeon group according to the following formula, calculating the fitness value and updating the global optimum, wherein the formula is as follows:
X i (t)＝X i (t-1)+rand*(X c (t)-X i (t-1))
7. the lithium battery multi-condition prediction method based on the optimized BiLSTM neural network according to claim 2, wherein the step S1 specifically includes the steps of:
s10: extracting the following lithium battery sampling parameters related to the lithium battery charge state and the residual energy state from the discharge data set, wherein the parameters comprise sampling end voltage in a discharge cycle, sampling end current in the discharge cycle, sampling lithium battery temperature in the discharge cycle, sampling lithium battery charge state and residual energy in the discharge cycle, and constructing a neural network input matrix x n And output data matrix y n ：
y n ＝[SOC n SOC n ]；
Wherein x is n The input matrix is BiLSTM at the time of n, and the size of the input matrix is m; i n 、V n 、T n Respectively measuring the current, voltage and temperature data of the battery at the moment n, wherein the SOC is the state of charge of the lithium battery, and the SOE is the state of residual energy; y is n The matrix size of the SOC and SOE values output by the output layer at the moment n is 1 multiplied by 2;
s10: normalizing the input matrix data and the output matrix data, wherein the three input variables V, I and T have different dimensionalities, and mapping the original data to intervals [0,1] by adopting a minimum-maximum normalization method in order to eliminate adverse effects; the normalization formula is as follows:
wherein x is the original data; x is x max And x min Respectively the maximum value and the minimum value of the original data; x is x * Is a normalized value.
8. The utility model provides a lithium cell multi-condition prediction device based on BiLSTM neural network after optimization which characterized in that includes: at least one processor and memory;
the memory stores computer-executable instructions; the at least one processor executing computer-executable instructions stored in the memory to perform the method of any one of claims 1-7.
CN202310713306.0A 2023-06-16 2023-06-16 Lithium battery multi-condition prediction method and device based on optimized BiLSTM neural network Pending CN117150334A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310713306.0A CN117150334A (en) 2023-06-16 2023-06-16 Lithium battery multi-condition prediction method and device based on optimized BiLSTM neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310713306.0A CN117150334A (en) 2023-06-16 2023-06-16 Lithium battery multi-condition prediction method and device based on optimized BiLSTM neural network

Publications (1)

Publication Number Publication Date
CN117150334A true CN117150334A (en) 2023-12-01

Family

ID=88883034

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310713306.0A Pending CN117150334A (en) 2023-06-16 2023-06-16 Lithium battery multi-condition prediction method and device based on optimized BiLSTM neural network

Country Status (1)

Country Link
CN (1) CN117150334A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117709394A (en) * 2024-02-06 2024-03-15 华侨大学 Vehicle track prediction model training method, multi-model migration prediction method and device

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109391515A (en) * 2018-11-07 2019-02-26 武汉烽火技术服务有限公司 Network failure prediction technique and system based on dove group's algorithm optimization support vector machines
CN110232169A (en) * 2019-05-09 2019-09-13 北京航空航天大学 Track denoising method based on two-way length memory models and Kalman filtering in short-term
CN112285568A (en) * 2020-10-21 2021-01-29 合肥工业大学 Estimation method of residual discharge time based on energy state of power lithium battery
CN113723007A (en) * 2021-09-08 2021-11-30 重庆邮电大学 Mechanical equipment residual life prediction method based on DRSN and sparrow search optimization BilSTM
CN114065635A (en) * 2021-11-22 2022-02-18 中国民航大学 Aircraft ground air conditioner energy consumption prediction method based on IALO-AM-BilSTM model and storage medium
CN114545274A (en) * 2022-01-26 2022-05-27 湖州学院 Lithium battery residual life prediction method
CN114966436A (en) * 2022-01-06 2022-08-30 湖北理工学院 Lithium battery state of charge prediction method, device, equipment and readable storage medium
CN115356635A (en) * 2022-07-05 2022-11-18 合肥工业大学 Identification method for lithium battery equivalent circuit model parameters

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109391515A (en) * 2018-11-07 2019-02-26 武汉烽火技术服务有限公司 Network failure prediction technique and system based on dove group's algorithm optimization support vector machines
CN110232169A (en) * 2019-05-09 2019-09-13 北京航空航天大学 Track denoising method based on two-way length memory models and Kalman filtering in short-term
CN112285568A (en) * 2020-10-21 2021-01-29 合肥工业大学 Estimation method of residual discharge time based on energy state of power lithium battery
CN113723007A (en) * 2021-09-08 2021-11-30 重庆邮电大学 Mechanical equipment residual life prediction method based on DRSN and sparrow search optimization BilSTM
CN114065635A (en) * 2021-11-22 2022-02-18 中国民航大学 Aircraft ground air conditioner energy consumption prediction method based on IALO-AM-BilSTM model and storage medium
CN114966436A (en) * 2022-01-06 2022-08-30 湖北理工学院 Lithium battery state of charge prediction method, device, equipment and readable storage medium
CN114545274A (en) * 2022-01-26 2022-05-27 湖州学院 Lithium battery residual life prediction method
CN115356635A (en) * 2022-07-05 2022-11-18 合肥工业大学 Identification method for lithium battery equivalent circuit model parameters

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
MA L, ET AL.: "State of charge and state of energy estimation for lithium-ion batteries based on a long short-term memory neural network", 《JOURNAL OF ENERGY STORAGE》, 30 March 2021 (2021-03-30) *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117709394A (en) * 2024-02-06 2024-03-15 华侨大学 Vehicle track prediction model training method, multi-model migration prediction method and device

Similar Documents

Publication Publication Date Title
Hong et al. Online accurate state of health estimation for battery systems on real-world electric vehicles with variable driving conditions considered
CN111007399B (en) Lithium battery state of charge prediction method based on improved generation countermeasure network
Kim et al. Estimation of Li-ion battery state of health based on multilayer perceptron: As an EV application
CN111458646A (en) Lithium battery SOC estimation method based on PSO-RBF neural network
CN113702843B (en) Lithium battery parameter identification and SOC estimation method based on suburb optimization algorithm
CN113253116A (en) Lithium ion battery state of charge estimation method and storage medium
CN113189490B (en) Lithium battery health state estimation method based on feature screening and Gaussian process regression
Zhang et al. Intelligent state of charge estimation of battery pack based on particle swarm optimization algorithm improved radical basis function neural network
CN112163372B (en) SOC estimation method of power battery
CN113740736A (en) Electric vehicle lithium battery SOH estimation method based on deep network self-adaptation
CN112327168A (en) XGboost-based electric vehicle battery consumption prediction method
Dou et al. Extreme learning machine model for state-of-charge estimation of lithium-ion battery using salp swarm algorithm
CN115856678A (en) Lithium ion battery health state estimation method
CN113687242A (en) Lithium ion battery SOH estimation method for optimizing and improving GRU neural network based on GA algorithm
CN115219906A (en) Multi-model fusion battery state of charge prediction method and system based on GA-PSO optimization
CN117150334A (en) Lithium battery multi-condition prediction method and device based on optimized BiLSTM neural network
CN114726045A (en) Lithium battery SOH estimation method based on IPEA-LSTM model
CN115015760A (en) Lithium battery health state evaluation method based on neural network and migration integrated learning
CN113406503A (en) Lithium battery SOH online estimation method based on deep neural network
CN112782594A (en) Method for estimating SOC (state of charge) of lithium battery by data-driven algorithm considering internal resistance
Zhang et al. A state-of-charge estimation method based on bidirectional lstm networks for lithium-ion batteries
CN110232432B (en) Lithium battery pack SOC prediction method based on artificial life model
Chen et al. Joint estimation of state of charge and state of energy of lithium-ion batteries based on optimized bidirectional gated recurrent neural network
Wen et al. Lithium battery health state assessment based on vehicle-to-grid (V2G) real-world data and natural gradient boosting model
CN112098869B (en) Self-adaptive electric vehicle SOC estimation method based on big data

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