RBF自适应神经网络，一般选用高斯函数，他的输入一般是实际速度与期望速度差，看您说参数选取一定范围内就可以

from sklearn.cluster import KMeans
from sklearn.metrics import pairwise_distances_argmin_min
import numpy as np

# 准备训练数据
X_train = np.random.rand(100, 1) * 10
y_train = np.sin(X_train)

# 使用K-means选择中心点
n_centers = 10
kmeans = KMeans(n_clusters=n_centers, random_state=0).fit(X_train)
centers = kmeans.cluster_centers_

# 计算带宽
distances = pairwise_distances_argmin_min(X_train, centers)
bandwidth = np.mean(distances[1]) * 2

# 创建RBF网络
def gauss(x, c, bw):
    return np.exp(-0.5 * ((x - c) / bw) ** 2)

def rbf_network(X_train, y_train, centers, bandwidth):
    n_centers = centers.shape[0]
    W = np.zeros((n_centers, 1))

    for i in range(n_centers):
        phi = gauss(X_train, centers[i], bandwidth)
        W[i] = np.dot(phi.T, y_train) / np.dot(phi.T, phi)

    def predict(X):
        n_samples = X.shape[0]
        Y = np.zeros((n_samples, 1))

        for i in range(n_centers):
            phi = gauss(X, centers[i], bandwidth)
            Y += np.dot(phi, W[i])

        return Y

    return predict

# 训练RBF网络
predict_func = rbf_network(X_train, y_train, centers, bandwidth)

# 测试RBF网络
X_test = np.linspace(0, 10, 100).reshape(-1, 1)
y_test = np.sin(X_test)
y_pred = predict_func(X_test)

# 计算误差
mse = np.mean((y_test - y_pred) ** 2)
print(f"Mean Squared Error: {mse}")