Peramalan Pelayanan Service Mobil (After-Sale) Menggunakan Backpropagation Neural Network (BPNN)

The automotive industry in Indonesia, primarily cars, is getting more and more varied. Along with increasing the number of vehicles, Brand Holder Sole Agents (ATPM) compete to provide after-sale services (mobile service). However, the company has difficulty knowing the rate of growth in the number of mobile services handled, thus causing losses that impact sources of income. Therefore, we need a standard method in determining the forecasting of the number of car services in the following year. This study implements the Backpropagation Neural Network (BPNN) method in forecasting car service services (after-sale) and Mean Square Error (MSE) for the process of testing the accuracy of the forecasting results formed. The data used in this study is car service data (after-sale) for the last five years. The results show that the best architecture for forecasting after-sales services using BPNN is the 5-10-5-1 architectural model with a learning rate of 0.2 and the learning function of trainlm and MSE of 0.00045581. This proves that the BPNN method can predict mobile service (after-sale) services with good forecasting accuracy values.

An efficient method for calculating system non-probabilistic reliability index

Collecting enough samples is difficult in real applications. Several interval-based non-probabilistic reliability methods have been reported. The key of these methods is to estimate system non-probabilistic reliability index. In this paper, a new method is proposed to calculate system non-probabilistic reliability index. Kriging model is used to replace time-consuming simulations, and the efficient global optimization is used to determine the new training samples. A refinement learning function is proposed to determine the best component (or performance function) during the iterative process. The proposed refinement learning function has considered two important factors: (1) the contributions of components to system nonprobabilistic reliability index, and (2) the accuracy of the Kriging model at current iteration. Two stopping criteria are given to terminate the algorithm. The system non-probabilistic index is finally calculated based on the Kriging model and Monte Carlo simulation. Two numerical examples are given to show the applicability of the proposed method.

Basic unit; as a common module of neural networks

In this paper, the logic is developed assuming that all parts of the brain are composed of a combination of modules that basically have the same structure. The feeding behavior of searching for food while avoiding the dangers of animals in the early stages of evolution is regarded as the basis of time series data processing. The module that performs the processing is presented by a neural network equipped with a learning function based on Hebb's rule, and is called a basic unit. The basic units are arranged in layers, and the information between the layers is bidirectional. This new neural network is an extension of the traditional neural network that has evolved from pattern recognition. The biggest feature is that in the processing of time series data, the activated part changes according to the context structure inherent in the data, and can be mathematically expressed the method of predicting events from the context of learned behavior and utilizing it in best action.

Fast Linear Interpolation

We present fast implementations of linear interpolation operators for piecewise linear functions and multi-dimensional look-up tables. These operators are common for efficient transformations in image processing and are the core operations needed for lattice models like deep lattice networks, a popular machine learning function class for interpretable, shape-constrained machine learning. We present new strategies for an efficient compiler-based solution using MLIR to accelerate linear interpolation. For real-world machine-learned multi-layer lattice models that use multidimensional linear interpolation, we show these strategies run 5-10× faster on a standard CPU compared to an optimized C++ interpreter implementation.

A novel adaptive support vector machine method for reliability analysis

For the structural reliability analysis, although many methods have been proposed, they still suffer from substantial computational cost or slow convergence rate for complex structures, the limit state function of which are highly non-linear, high dimensional, or implicit. A novel adaptive surrogate model method is proposed by combining support vector machine (SVM) and Monte Carlo simulation (MCS) to improve the computational efficiency of estimating structural failure probability in this paper. In the proposed method, a new adaptive learning method is established based on the kernel function of the SVM, and a new stop criterion is constructed by measuring the relative position between sample points and the margin of SVM. Then, MCS is employed to estimate failure probability based on the convergent SVM model instead of the actual limit state function. Due to the introduction of adaptive learning function, the effectiveness of the proposed method is significantly higher than those that employed random training set to construct the SVM model only once. Compared with the existing adaptive SVM combined with MCS, the proposed method avoids information loss caused by inconsistent distance scales and the normalization of the learning function, and the proposed convergence criterion is also more concise than that employed in the existing method. The examples in the paper show that the proposed method is more efficient and has broader applicability than other similar surrogate methods.