Seed yield is influenced by the combined effects of genes, including additive and non-additive interactions. Therefore, accurately predicting seed yield holds significant importance in rapeseed breeding. Nonetheless, limited information exists regarding yield estimation for canola using neural networks. This study employs multi-layer perceptron (MLP) neural network, radial basis function neural network and support vector machine, to forecast rapeseed yield. The models are trained using phenological, morphological, yield and yield-related data, as well as molecular marker information from 8 genotypes and 56 hybrids. Comparative analysis of the models reveals that the MLP model effectively forecasts hybrid yield with root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (R²) values of 226, 183, and 92%, respectively. Among the 40 primers examined, the ISJ10 primer demonstrates superior discriminatory power compared to others. The use of molecular and phenotypic data as inputs in the model highlights the MLP model’s superiority, presenting lower RMSE and MAE values, along with a higher R², compared to direct crosses in predicting the performance of reciprocal crosses. The proposed neural network model enables performance estimation of hybrids prior to crossing parent studied, thereby enabling spring rapeseed breeders to focus on the most promising hybrids.

种子产量受到基因综合效应的影响，包括加性和非加性相互作用。因此，准确预测种子产量对于油菜育种具有重要意义。尽管如此，关于使用神经网络估算油菜籽产量的信息仍然有限。本研究采用多层感知器（MLP）神经网络、径向基函数神经网络和支持向量机来预测油菜产量。这些模型使用物候、形态、产量和产量相关数据以及来自 8 个基因型和 56 个杂交品种的分子标记信息进行训练。模型比较分析表明，MLP模型可以有效预测混合产量，均方根误差（RMSE）、平均绝对误差（MAE）和决定系数（R ²）值分别为226%、183%和92%。在检查的 40 种引物中，ISJ10 引物与其他引物相比表现出优越的区分能力。使用分子和表型数据作为模型的输入突显了 MLP 模型的优越性，与直接杂交相比，在预测互交性能方面呈现出较低的 RMSE 和 MAE 值以及较高的 R ^{2 。}所提出的神经网络模型能够在研究杂交亲本之前对杂交品种进行性能评估，从而使春油菜育种者能够专注于最有前途的杂交品种。