In reality, readings of sensors on highways are usually missing at various unexpected moments due to some sensor or communication errors. These missing values do not only influence the real-time traffic monitoring but also prevent further traffic data mining. In this paper, we propose a multi-view learning method to estimate the missing values for traffic-related time series data. The model combines data-driven algorithms (long-short term memory and support vector regression) and collaborative filtering techniques. It can consider the local and global variation in temporal and spatial views to capture more information from the existing data. The estimations of missing values from four views are aggregated to obtain a final value with a kernel function. Data from a highway network are used to evaluate the performance of the proposed model in terms of accuracy, precision, and agreement. The results indicate that our proposed model outperforms other baselines, especially for block missing pattern with a high missing ratio. Furthermore, the sensitivity of the parameters is analyzed. We can conclude that combining different views can improve the performance of the imputation.

Missing data, especially a block of missing data, inevitably occur in structural health monitoring systems. Because of their severe negative effects, many methods that use measured data to infer missing data have been proposed in previous research to solve the problem. However, capturing complex correlations from raw measured signal data remains a challenge. In this study, empirical mode decomposition is combined with a long short-term memory deep learning network for the recovery of the measured signal data. The proposed hybrid method converts the missing data imputation task as a time series prediction task, which is then solved by a “divide and conquer” strategy. The core concept of this strategy is the prediction of the subsequences of the raw measured signal data, which are decomposed by empirical mode decomposition rather than directly predicted, as the decomposition can assist in the modeling of the irregular periodic changes of the measured signal data. In addition, the long short-term memory network in the hybrid model can remember more long-range correlations of subsequences than can the traditional artificial neural network. Three widely used prediction models, namely, the autoregressive integrated moving average, support vector regression, and artificial neural network models, are also implemented as benchmark models. Raw acceleration data collected from a cable-stayed bridge are used to evaluate the performance of the proposed method for missing measured signal data imputation. The recovery results of the measured signal data demonstrate that the proposed hybrid method exhibits excellent performance from two perspectives. First, the decomposition by empirical mode decomposition can improve the accuracy of the core long short-term memory prediction model. Second, the long short-term memory model outperforms other benchmark models because it can fit more microscopic changes of measured values. The experiments conducted in this study also suggest that the change patterns of raw measured signal data are complex, and it is therefore important to extract the features before modeling.

The introduction of data-driven inertial navigation provides new opportunities that the pedestrian dead reckoning could not well provide for constraining inertial system error drift on smartphones, and has been considered as another promising approach to meet the requirement of location-based services. However, indoor localization systems based on a single technology still have their limitations, such as the drift of inertial navigation and the received signal strength fluctuation of Bluetooth, making them unable to provide reliable positioning. To exploit the complementary strengths of each technology, this paper proposes a feasible fusion framework by utilizing a particle filter to integrate data-driven inertial navigation with localization based on Bluetooth Low Energy (BLE). For data-driven inertial navigation, under the premise of using the deep neural network with great potential in model-free generalization to regress pedestrian motion characteristics, we effectively combined the method of using gravity to stabilize inertial measurement units data to make the network more robust. Experimental results show that in the test of different smartphone usages, the proposed data-driven inertial navigation and BLE-based localization technology have good results in modeling user’s movement and positioning respectively. And due to this, the proposed fusion algorithm has almost unaffected by the usages of smartphones. Compared with BLE-based localization that achieved a good mean positional error (MPE) of 1.76m, for the four usages of texting, swinging, calling and pocket, the proposed fusion algorithm reduced the MPE by 32.35%, 20.51%, 20.74%, and 45.37%, respectively, and can further improve localization accuracy on the basis of existing fusion method.