To address the issue of the disconnection between planning and operation of transmission and distribution networks in the process of distribution network planning (DNP), a DNP method is proposed based on integrated planning and operation with the transmission network. The method aims to minimize the sum of investment and operation costs for both the distribution and transmission networks, while considering local units, distributed renewable energy, and network constraints. The optimization model is solved using a heterogeneous decomposition algorithm(HGD), which alternates between the optimization of the transmission and distribution regions, and an auxiliary function is introduced to ensure the optimality of the whole network by injecting power and other parameters. The topology information of the planning solution is extracted, and a convolutional neural network is used to build a surrogate model for the complex coordination between transmission and distribution operations, enabling a fast and accurate search for the optimal DNP solution. The proposed algorithm has been tested with numerical examples, demonstrating high accuracy, good convergence, and practical value for engineering applications.