Neural Network based controller

In summary, the project involves using a neural network to control the temperature of a room within a specific limit. The network has been trained with 99.5% accuracy using prior data and has 5 inputs (x, y, z, A, B) and one output (T). The goal is to use this network in a control architecture. The system is complex, non-linear, and dynamic in nature. Standard inversion of the system has been unsuccessful. The NN will act as a feedback controller for the 5 inputs, with real-time sensors providing input and an actuator controlling the output. It is unclear what type of learning will be used, and the dynamic nature of the environment may make training difficult. Additional clarification may be needed
  • #1
date.chinmay
10
0
I'm working on a project which deals with temperature control of a room. the idea is to control temperature within a limit.

I have prior data which i used to train a neural network which is 99.5% accurate. there are 5 inputs (x , y , z , A , B) and there is one output (T).

Now I want to use this network in a control architecture of some kind. Any suggestions? which should I use?

P.S. The system is highly complex, non linear and dynamic in nature. I have already tried standard inversion of system but it doesn't work as outputs are lesser than inputs.
I'm doing this in Matlab.
 
Engineering news on Phys.org
  • #2
I am not quite following your question. Typically you would train the network using your inputs and the desired output for those inputs. In essence what you would be doing is trained the NN to become a feedback controller for the 5 inputs.

To use the NN you would then put in a real time system with some types of sensors feeding your x,y,z,A,B values and your T value going to an actuator that effects some output.
 
  • #3
Floid said:
I am not quite following your question. Typically you would train the network using your inputs and the desired output for those inputs. In essence what you would be doing is trained the NN to become a feedback controller for the 5 inputs.

To use the NN you would then put in a real time system with some types of sensors feeding your x,y,z,A,B values and your T value going to an actuator that effects some output.

Thanks for replying!
let me come straight..
X-Y-Z are my coordinates... A is temperature controlling device.. and B is the Heat dissipated at that point...

All these affect the output i.e. T at that point.

when i say i am trying to invert it... i mean.. I'm trying to make the ANN act as if.. when it reads the temperature in an actual REAL system.. it should predict the input parameters... so as to push it towards a desired state.. Can I email or private message you for more clarification?
 
  • #4
Sure, you can send me a PM but I am not sure I can be much help. It sounds like you want you NN to provide the feedback for a controller so that based on a temperature T you supply the controller with input (x, y, z, A, B) that go to an actuator that is capable of driving T one way or the other.

Given that, what type of learning do you plan on implementing?

If you are doing online (reinforcement learning) then is your environment dynamic while you are trying to train the NN? If so it is going to be next to impossible to train it. The NN may be giving a good output but the environment is natural changing T in the other direction outside of your control making the learning process think you are giving it bad inputs. Or vice versa... you may be giving bad control outputs but the environment happens to be changing in your favor so the NN thinks it is doing good.
 
  • #5


As a fellow scientist, I find your project on temperature control of a room using a neural network-based controller to be very interesting. It is impressive that you have trained a neural network with 99.5% accuracy using prior data. This shows the potential of using artificial intelligence techniques in control systems.

In terms of suggestions for your control architecture, I would recommend looking into model predictive control (MPC). This approach uses a predictive model of the system to optimize control inputs over a finite time horizon, taking into account constraints and disturbances. It has been shown to be effective in controlling complex, non-linear and dynamic systems like the one you are working on.

Another option could be using a fuzzy logic controller, which can handle non-linear and dynamic systems by using linguistic variables and rules to make decisions. This could be a good fit for your system as it is highly complex and non-linear.

I would also suggest exploring different control algorithms and comparing their performance to see which one works best for your specific system. Additionally, considering implementing a feedback control loop to continuously adjust the control inputs based on the output from the neural network.

In terms of the software, Matlab has built-in tools for implementing control systems, so it would be a good choice for your project. I would also recommend considering using a simulation tool to test and optimize your control system before implementing it in a real-world setting.

Overall, your project has great potential and I am excited to see how you incorporate the neural network into your control architecture and the results of your temperature control system. Best of luck with your project!
 

1. What is a neural network based controller?

A neural network based controller is a type of control system that uses artificial neural networks (ANNs) to make decisions and control a process or system. ANNs are models inspired by the structure and function of the human brain and are used to learn and adapt to complex patterns and relationships in data.

2. How does a neural network based controller work?

A neural network based controller works by taking input data and passing it through multiple layers of interconnected neurons. These neurons use mathematical functions to process the data and produce an output. The network is trained using a large dataset to adjust the weights and biases of the neurons, allowing it to make accurate predictions and decisions.

3. What are the advantages of using a neural network based controller?

There are several advantages to using a neural network based controller. Firstly, it can handle complex and nonlinear relationships in data, making it suitable for a wide range of applications. It can also adapt and learn from new data, making it more robust. Additionally, it can operate in real-time, making it suitable for dynamic systems.

4. What are some applications of neural network based controllers?

Neural network based controllers have a variety of applications in fields such as robotics, finance, healthcare, and engineering. They can be used for tasks such as autonomous control, prediction and forecasting, classification, and optimization. Some specific examples include self-driving cars, stock market prediction, medical diagnosis, and process control.

5. What are the limitations of using a neural network based controller?

While neural network based controllers have many advantages, they also have some limitations. One of the main limitations is the need for large amounts of data for training, which can be time-consuming and expensive. They can also be difficult to interpret and may not always provide explainable results. Additionally, they may suffer from overfitting and require regular retraining to maintain accuracy.

Similar threads

  • Electrical Engineering
Replies
12
Views
2K
  • Programming and Computer Science
Replies
18
Views
1K
  • Computing and Technology
Replies
4
Views
1K
  • Mechanical Engineering
Replies
2
Views
1K
  • Programming and Computer Science
Replies
1
Views
794
  • Set Theory, Logic, Probability, Statistics
Replies
1
Views
956
  • Programming and Computer Science
Replies
9
Views
1K
Replies
6
Views
695
Replies
31
Views
2K
  • Electrical Engineering
Replies
13
Views
4K
Back
Top