DECLARATION OF ORIGINALITY

Supervisor) of the Department __Computer and Communication Technology, Faculty _______Information and Communication Technology._. The state space model and neural network model can be identified as the prediction model.

INTRODUCTION

• Problem Statement and Motivation
• Project Objectives
• Project Scope
• Contributions
• Background Information
• Report Organization

Furthermore, the energy consumption of the HVAC system is highly dependent on the ambient temperature and the cabin temperature of the EV. The results will be presented in this chapter based on the implementation of the controllers.

Table 1.1: Energy consumption of some auxiliary system [5]

LITERATURE REVIEW

Model Predictive Control (MPC)

The best range of inputs that has the minimum cost function will be selected by MBK. More discussion will be included in the next section as the project focuses on MPC.

Rule-based bang-bang controller

9 If the cabin temperature is above the upper limit of the target temperature, the temperature will drop to the lower limit of the target temperature as soon as possible, so that the compressor will operate at maximum cooling capacity. If the cabin temperature has not reached the lower limit of the target temperature, the cooling capacity is controlled by the function as shown in Eq. Therefore, the cabin temperature can be controlled within the lower limit and upper limit of the target temperature [8].

Especially in this case with the set point for the upper limit and the lower temperature limit.

Fuzzy Logic Control (FLC)

The ratio between the cooling capacity and power consumption of the AC system is called the coefficient of performance (COP) of the AC system. Where P is the power consumption and Qcool is the cooling capacity of the AC system. The sensible heat brought by the passenger of the vehicle will also affect the thermal load of the electric vehicle and is related to gender, age and work intensity, etc.

Where ρair and Vair are the density and volume of the compressor and Qcooling is the cooling capacity of the compressor. The first state variable of the SMPC is the cabin air temperature and the second state variable and control variable of the SMPC is the cooling capacity of the AC system, so that the compressor can be prevented from changing the operating conditions too often. The optimal control decisions are calculated by the SMPC controller to minimize the cost function.

The first set of optimal control signals, feedback of the system state and repetition of the control process are performed.

Table 2.1: Input and Output of the Fuzzy Logic Controller [15].

SYSTEM METHODOLOGY

System Model

• Simplified First-Principle Model
• CoolSim Model

Where Ks and Kglass are the heat transfer coefficient of sheet metal and glass, respectively, Fs and Fglass are the heat transfer surface of sheet metal and glass, respectively, while Tsurf_s and Tsur_glass are the surface temperature of sheet metal and glass, respectively. Where ρair is the air density of the cabin, Vair is the volume of the cabin, Cρair is the heat capacity of the cabin and Qcool is the cooling capacity of the compressor. CoolSim, an open-source modeling platform available at the National Renewable Energy Lab, was used to create a high-fidelity simulation model of the passenger car air conditioning system (NREL).

In this model, there are four major sub-components consisting of a boundary block, refrigerant circuit block, compressor block, and cabin space block [19]. The sub-model of interest in this project can be described by the following non-linear model. Due to its complexity and non-linearity, the neural network model obtained by system identification will be used to predict future temperature output.

The outputs of the CoolSim model, stimulated with random input signals, are then sampled at 0.2 Hz to create data for determining the unknown parameters [19].

Table 3.1: Parameters of the CoolSim Model [19]

SYSTEM DESIGN

System Identification

• State-Space Model
• Feed-Forward Model – Neural Network Model

The purpose of the neural network is to predict the future production of the plant. The input layer will take the previous inputs and plant outputs as inputs for the model. Then the inputs will be passed on to the hidden layer of the neural network.

The size of the hidden layer depends on the number of neurons in the hidden layer. The larger the size of the hidden layer (number of neurons), the stronger the power of the network. Weights play an important role in the neural network model as they are the parameters of the equation.

It was used to improve the weights of the neural network so that the neural network model can have a good performance.

Figure 4.2 shows the block diagram of state space model equation.

Implementation of MPC

There are various training algorithms such as gradient descent, Newton's method, Levenberg-Marquardt algorithm, etc. Levenberg-Marquardt is an algorithm used to cope with loss functions that are expressed as a sum of squared errors such as mean squared error and uses the Jacobian matrix and the gradient vector instead of calculating the actual Hessian matrix. The state diagram below depicts the process of training a neural network using the Levenberg-Marquardt algorithm.

Therefore, the second model, which is CoolSim model, will use the neural network model for system identification. 26 Dynamic model is crucial for MPC to model the states of the system in a particular horizon with various input options. In our case, state space model and neural network model are selected as the dynamic models for this system.

Then the model is the dynamic model of the controller which is state space model and neural network model.

Simulation Result

First, plant means a plant model, in this case it will be an HVAC plant model, which are a simplified first principle model and a CoolSim model. The actual response of the plant is then sent back to the model block for future prediction.

SYSTEM IMPLEMENTATION

• Hardware Setup
• Software Setup
• Implementation of State Space MPC
• Implementation of Neural Network MPC
• Simulink Design of Models
• Parameter Configuration

The MPC toolbox is used to model the state space and was implemented on the command line instead of Simulink. Since the CoolSim model is a nonlinear model, it cannot be recognized in the state space approach. After defining the plant model, an MPC must be created to feed the plant model inside the controller.

Since the neural network model predictive control (NNMPC) toolbox is only supported in Simulink, the models must therefore be designed in Simulink as the plant model instead of command line. The output of the reference will then be fed into the optimizer block with the output of the NN model. The control signal will then be fed into the NN model as well as the plant model's actual output to predict the future output of the plant model.

At the same time, the control signal is also fed into the plant model to generate the actual system output.

Figure 5.1: Flow Chart of State Space MPC

SYSTEM EVALUATION AND DISCUSSION

Result of Simulations

Same with Figure 6.2, which is the input and output of the Simplified First-Principle Model with changing trajectory. For Figure 6.3, the system took less time to reach the static reference path compared to Figure 6.1. For Figure 6.4, based on the plant's output, it can be said that the output is almost fulfilled throughout the changing path.

In fact, both parameters in Table 5.2 and Table 5.3 also show good results, but in comparison between them, the implementation of parameters in Table 5.3 may be better than Table 5.2. Figure 6.3 and 6.4 show the simulation results of NNMPC with changing and static trajectories respectively. From Figure 6.5, it can be observed that the cabin temperature is also almost reached the reference trajectory successfully even though the trajectory changes continuously even though the outputs were initially unstable.

In Figure 6-6, which has a static trajectory, the output also remains at 23 degrees Celsius after a few seconds.

Figure 6.3: Input and Output of Linear MPC with static trajectory (Parameters obtained from Table 5.3)

Project Challenges

In the end, both NNMPC and Linear MPC show good results, as cabin temperatures are successfully optimized by meeting the system's reference trajectory. As there are other courses to be taken this semester, very good time management is required to complete this project and the other courses.' In other words, it could be a very difficult process to complete this project within the deadline. laps.

If time management is poor, the project will definitely not be completed on time. Finally, the process of adjusting the parameters of the controller and the device model may also have failed. If the outputs of the model or system continue to fail, meaning that parameters need to be adjusted again and again until the model becomes successful.

In summary, the three challenges we encountered in this project were defining the factory models of the system, time management for the project, and fine-tuning the factory model and controller parameters.

CONCLUSION AND RECOMMENDATION

Conclusion

Recommendation

Benachour, "Impact of the air conditioning system on the power consumption of an electric vehicle powered by lithium-ion battery," Modeling and Simulation in Engineering, vol. Kim, “Modeling and Control of Temperature and Humidity in Building Heating, Ventilation and Air Conditioning System Using Model Predictive Control,” Energies, vol. Trächtler, “A new approach using model predictive control to improve the range of electric vehicles,” Procedia Technology, vol.

Sun, “Stochastic model predictive control of electric vehicle air conditioning system: sensitivity study, comparison, and improvement,” IEEE Transactions on Industrial Informatics, vol. 43 https://www.mathworks.com/help/deeplearning/ug/design-neural-network-predictive- controller-in-simulink.html. Lack of time to do the project, since many of the midterms and deadlines in this week.

Program / Course BACHELOR OF COMPUTER SCIENCE (HONOURS) Final Year Title Project Model Predictive Control of Air Conditioning System for. Note The Supervisor/Candidate(s) must/are required to provide the Faculty/Institute with a full copy of the complete set of originality report. Based on the above results, I declare that I am satisfied with the authenticity of the Final Year Project Report submitted by my student(s) as mentioned above.