PCA

greedy search for signal processing

日前调度计划优化，根据负荷、风电、光伏、燃气轮机、蓄电池出力计划，安排日前调度出力，以成本最优为目标函数进行调度(Recently scheduling optimization, depending on the load, wind power, photovoltaic, gas turbines, battery output plan arranged before scheduling output, optimal scheduling cost as the objective function)

把辨识分成两步进行：第一步：利用相关分析法获得对象的非参数模型（脉冲响应或相关函数）；第二步：利用最小二乘法、辅助变量法或增广最小二乘法等，进一步求的对象的参数模型。如果模型噪声与输入无关，则Cor-ls相关最小二乘法（二步法）可以得到较好的辨识结果。Cor-ls相关最小二乘法（二步法）实质上是先对数据进行一次相关分析，滤除了有色噪声的影响，再利用最小二乘法必然就会改善辨识结果。能适应较宽广的噪声范围，计算量不大，初始值对辨识结果影响较小。但要求输入信号与噪声不相关(The identification is divided into two steps: the first step: using the correlation analysis method for object nonparametric model (impulse response or related function) The second step: using the least squares method, auxiliary variable method or augmented least squares method and so on, further for the parameters of the object model. If the model input noise and independent, the Cor- ls related least squares (two footwork) can get good recognition results. Cor- ls related least squares (two footwork) is essentially the data to a correlation analysis, filter in addition to the influence of colored noise, using least squares inevitable will improve the identification result. To adapt to a wide range of noise and calculation is not big, the initial value to identification results less effect. But for the input signal and noise are not related)

在matlab下编程使用jacket工具包进行GPU加速图像平滑算法，对900×900的图像平滑可达8倍多的加速比。需要jacket工具包及cuda toolkit。(In matlab programming toolkit used jacket GPU-accelerated image smoothing algorithm, the 900,900 image smooth acceleration up to 8 times more than that. Need jacket toolkit and cuda toolkit.)

该代码能够实现改进的线性调频Z变换的功能，取代FFT变换能够起到细化频谱的作用。(The code can achieve improved chirp Z transform function, replace the FFT spectrum can play a role in refinement.)

matlab实例，很实用，很简单，尤其对新手有很大帮助(matlab examples, very practical, very simple, very helpful especially for novice)

some code in mimo ofdm

matlab code for high pass filter

结构可靠度分析模型拟合样本点数目影响下的功能函数画图，分别采用ANN和SVM模型(Structural reliability analysis model fitting function drawing under the influence of the number of sample points, respectively, ANN and SVM models)

利用SOM神经网络设计矢量量化的码书，结果优于传统的LBG方法，失真度较小。(SOM neural network design use vector quantization codebook, the result is better than traditional methods LBG, distortion is small.)