Intelligent mass air flow (MAF) prediction system with neural network

A Method of Estimating or Predicting MAF Sensor Information, comprising the steps of:

• providing an engine with an engine controller, and an engine intake Mass Air Flow (MAF) sensor;
• training multiple candidate Artificial Neural Network (ANN) architectures using a training set of data;
• selecting an ANN architecture from the multiple candidate ANN architectures by way of an automated ANN architecture selection algorithm using a testing set of data;
• configuring an intelligent engine intake MAF prediction or estimation system to provide an engine intake MAF output variable using the selected ANN architecture;
• deploying the intelligent engine intake MAF prediction or estimation system comprising the selected ANN architecture into the engine controller;
• the training set of data and the testing set of data comprising at least one input variable from at least one of an engine sensor and an actuator that affects or is affected by engine intake MAF;
• and
• the training set of data and the testing set of data are acquired by testing a target engine and pre-processing a body of data acquired therefrom, pre-processing the body of data comprising splitting the body of data into the training set of data and the testing set of data;
• pre-processing the body of data further comprising normalizing the data;
• triggering the automated ANN architecture selection algorithm after pre-processing the body of data;
• generating all possible combinations of ANNs from a number of hidden layers and hidden neurons using the automated ANN architecture selection algorithm;
• training each of the possible combinations of ANNs by way of a Levenberg-Marquardt algorithm using the automated ANN architecture selection algorithm;
• and
• selecting the optimal ANN architecture based on Root Mean Squared Error (RMSE) analysis using the automated ANN architecture selection algorithm and the testing set of data.