**Input**: Training samples , the Gaussian kernel function is selected as the basic |

kernel function , the nonzero constant , the number of hidden nodes and |

the number of data blocks in a data set. |

**Output**: The predictive value of the polished rod load and suspension point displacement |

Randomly generate the input layer parameters ; |

Construct the data-dependent kernel function by modifying the initial basic kernel function according |

to (29), whose parameters are optimized by Algorithm 1; |

**for** each **do** |

Update the output weight by (27); |

**end for** |

Compute the predictive value according to (20). |