Pulse laser ranging technology based on time of flight (ToF) is widely used to detect non-cooperative targets due to its narrow pulse width and high peak power. In a conventional ranging system based on threshold discrimination, the measurement accuracy is limited by the range walk error of the echo signal, waveform jitter, clock drift, and system delay, resulting in power fluctuation of the echo signal, measurement data exception and error. To solve these problems, in this study, a ranging accuracy compensation based on ToF is proposed using a bi-directional convolutional neural network. The method enables anomaly detection for the pulse width with the bi-directional long short-term memory network using multi-layer constraint information, and adaptive accuracy compensation through one dimensional convolutional neural network regression analysis. In the experimental demonstration, the ranging error is 0.05~0.33 m for a measurement range of 4~110 m. Compared to the polynomial fitting method, the measurement accuracy is improved by 72.5%, which verifies the effectiveness of the proposed method.