Recurrent Trend Predictive Neural Network

This repository contains the implementation of the Recurrent Trend Predictive Neural Network (rTPNN) model as a Keras layer. In addition, it also contains an application of rTPNN for the multi-sensor fire detection in the folder FireDetection_via_rTPNN.

You may find the more detailed explanation of the methodology as well as the results in our publication at https://ieeexplore.ieee.org/document/9451553.

Note that it is an particular implementation of rTPNN, and it may be implemented in different ways.

Inputs for rTPNN Layer

Provide input array "x" as shown in the following figure.

An example usage of rTPNN

import numpy as np
from keras.layers import Input, Dense
from keras import Model
from rTPNN_layer import rTPNN

Random Data

num_samples = 100; num_features = 5

x = np.random.rand(num_samples, 2, num_features)
y = np.random.rand(num_samples)

Create an rTPNN Model

input_layer = Input(input_shape=(2, num_features,))

rtpnn_layer = rTPNN()(input_layer)

fullyconnected_layer = Dense(num_features, activation='relu')(rtpnn_layer)

output_layer = Dense(1, activation='relu')(fullyconnected_layer)

rTPNN_model = Model(inputs=[input_layer], outputs=[output_layer])

rTPNN_model.compile(optimizer='adam', loss='mse')

Train the Model

rTPNN_model.fit(x, y, epochs=10, batch_size=20, verbose=0)

''' batch_size determines the time interval for the update of recurrence. "The last state for each sample at index i in a batch will be used as initial state for the sample of index i in the following batch." [https://keras.io/api/layers/recurrent_layers/simple_rnn/] '''

Make Prediction

prediction = rTPNN_model.predict(x, batch_size=1)

Applications of rTPNN

Fire Detection: https://github.com/mertnakip/Recurrent-Trend-Predictive-Neural-Network/tree/main/FireDetection_via_rTPNN

Energy Management and Forecasting: https://github.com/mertnakip/Recurrent-Trend-Predictive-Neural-Network/tree/rtpnn_sef

Citation Request

The rTPNN as well as its application on multi-sensor fire detection has been published as a journal paper which is entitled as "Recurrent Trend Predictive Neural Network for Multi-Sensor Fire Detection" on IEEE Access. If you use rTPNN or the content of this repository, please cite our following paper (along with the repository citation) as follows:

@ARTICLE{nakip2021rTPNN,  
  author={Nakip, Mert and Güzeliş, Cüneyt and Yildiz, Osman},  
  journal={IEEE Access},  
  title={Recurrent Trend Predictive Neural Network for Multi-Sensor Fire Detection},  
  year={2021},  
  volume={9},  
  number={},  
  pages={84204-84216},  
  doi={10.1109/ACCESS.2021.3087736}  
  }