KimferLang/battery_state_prediction

Battery state of charge prediction based on machine learning algorithm for competition

noteId	tags
a7b22930795a11ea88acd9ef9a4d234b

Battery state of charge prediction based on machine learning algorithm

Version 1.0 : Re-write the Electric vehicles and batteries simulation model, build a CNN-LSTM model for SOC prediction, the SOH is also considered.

Cooperator

• Xin Chen: Supervisor
• Yuchen Zhang: SOC Prediction Model
• Siyuan Liang, Peiyuan Sun: Electric Vehicles Model
• Jiali Lu: Investigation

Platform

• Python: 3.x
• Pytorch: 0.4+
• Matlab: R2019a
• Simulink: 10.0

SOC Prediction Model

Improve prediction accuracy and avoid the high errors near zero point by cutting off the parameter transfers between different batches of data

Howto

1. Download smallNewModelData dataset from here
2. Sturcture:

Prediction_Model_v1/
├── smallNewModelData
	├── soc01.csv  
	├── soc01.csv 
	...
├── main.py
    ├── SOC_v1.py
    ├── SOC_test_v1.py
    ├── Data.py
└── net_params.pkl

4. Modify the path to your actual data path
5. Run main.py

Hyperparameters

EPOCHES     = 1500
RATE        = 8e-3
HIDDEN_SIZE = 48
Optimizer   = Adam

The pretiction result

• SOC=83% RMSE=1.76%
• SOC=90% RMSE=1.88%

Sturcture

1. Run main.py to train the model
2. CNN-LSTM model is defined in Soc_v1.py
3. Soc_test_v1.py is for testing
4. Data.py is for data processing, but you don't need to run it
5. Parameter is in net_params.pkl

Electric Vehicles Model

Howto

1. Determine the weight, windward area, internal drive efficiency and the numbers of batteries of the simulation objects
2. Fill them into the corresponding position of the model
3. The speed and slope is also needed

Vehicles Hardware Module

Structure

1. Operating input module
2. Automotive speed control module
3. Torque conversion module
4. Vehicle kinect module
5. Power conversion module
6. Thermodynamics module

Function

• Get access to the simulation of the vehicle's inner situation under different working states which involve the variation of slope and velocity

• Figure out the specific power needed from the battery to support the vehicle's work and the

• State of Charge

• Velocity

• Current

• Terminal voltage

Battery Module

Use the PNGV model to descibe the port of the battery

Parameter Identification

• Lithium battery SOC calculation block diagram based on PNGV model
• PNGV equivalent model
• OCV-SOC

Simulation (Initial SOC = 90%)

• State of Charge
• Temperature
• Terminal Voltage