PineScript技术实现：按密度/邻近度分割枢轴高点数组聚类

问题背景 #

我正在开发交易策略，目标是绘制支撑阻力（S&R）通道。目前已实现通过枢轴高点（PH）绘制多个支撑阻力线并添加至图表，现有PH数组：
[1.09118, 1.09192, 1.10201, 1.076, 1.07898, 1.07614, 1.06937, 1.06714, 1.06866, 1.07396, 1.07396, 1.07448, 1.07814, 1.09038]

需要用简便方法将任意此类数组分割为对应不同聚类的子数组，无需Jenks自然断点法这类复杂实现。例如输入数组[1,2,3,5,10,11,13,65,70]，期望输出[[1,2,3,5],[10,11,13],[65,70]]。

实现思路 #

采用阈值分割法，这是最轻量化的聚类方案，核心逻辑：

• 先对PH数组排序，消除时间维度的干扰
• 设定一个聚类阈值（可作为参数调整，适配不同品种波动）
• 遍历排序后的数组，若当前元素与当前聚类最后一个元素的差值超过阈值，则新建聚类；否则加入当前聚类

修改后的完整PineScript代码 #

//@version=5
strategy("Support and Resistance with PH Clustering", overlay=true)
// ============================================================================================================ //
// =============================================TIME RANGE===================================================== //
// ============================================================================================================ //

// 日期范围筛选设置
i_dateFilter    = input(true,  "═════ 日期范围筛选 ═════")
i_fromYear      = input.int(2023,   "起始年份",    minval = 1900)
i_fromMonth     = input.int(1,      "起始月份",   minval = 1, maxval = 12)
i_fromDay       = input.int(22,     "起始日期",     minval = 1, maxval = 31)
i_toYear        = input.int(2023,   "结束年份",      minval = 1900)
i_toMonth       = input.int(3,      "结束月份",     minval = 1, maxval = 12)
i_toDay         = input.int(23,      "结束日期",       minval = 1, maxval = 31)

// 生成时间范围条件
start     = timestamp(i_fromYear, i_fromMonth, i_fromDay, 00, 00)
finish    = timestamp(i_toYear, i_toMonth, i_toDay, 23, 59)
window()  => (time >= start and time <= finish) or not i_dateFilter ? true : false

// 显示日期范围背景
showDate = input(defval = true, title = "显示日期范围背景") 
bgcolor(color = showDate and window() ? color.new(color.gray, 90) : na)

// ============================================================================================================ //
// =============================================枢轴高点聚类==================================================== //
// ============================================================================================================ //

// 枢轴高点参数
leftBars = input(5, "左侧确认K线数")
rightBars = input(5, "右侧确认K线数")
ph = ta.pivothigh(close, leftBars, rightBars)
plot(ph, style=plot.style_cross, linewidth=5, color= color.blue, offset=-rightBars)

// 存储所有枢轴高点的数组
var float[] allPh = array.new_float(0)

// 聚类参数
clusterThreshold = input.float(0.003, "聚类阈值", tooltip="根据品种波动调整，EURUSD建议0.002-0.005")
var array<float[]> phClusters = array.new_float_array(0)

// 收集枢轴高点并实时聚类
if ph and window()
    line.new(bar_index - rightBars, ph, last_bar_index, ph, color = color.blue)
    array.push(allPh, ph)
    // 每次新增PH后重新聚类
    array.clear(phClusters)
    if array.size(allPh) > 0
        // 1. 对PH数组升序排序
        sortedPh = array.sort(allPh, order.ascending)
        // 2. 初始化第一个聚类
        currentCluster = array.new_float(1, array.get(sortedPh, 0))
        array.push(phClusters, currentCluster)
        // 3. 遍历剩余元素完成聚类
        for i from 1 to array.size(sortedPh)-1
            currentVal = array.get(sortedPh, i)
            lastCluster = array.get(phClusters, array.size(phClusters)-1)
            lastValInCluster = array.get(lastCluster, array.size(lastCluster)-1)
            if currentVal - lastValInCluster > clusterThreshold
                // 差值超过阈值，新建聚类
                newCluster = array.new_float(1, currentVal)
                array.push(phClusters, newCluster)
            else
                // 加入当前聚类
                array.push(lastCluster, currentVal)

// 输出聚类结果到图表（用于调试）
if barstate.islast
    label.new(bar_index, high, str.tostring(phClusters), color=color.white, textcolor=color.black, style=label.style_label_down)

效果说明 #

• 对于示例数组[1,2,3,5,10,11,13,65,70]，将clusterThreshold设为2，即可得到期望的[[1,2,3,5],[10,11,13],[65,70]]聚类结果
• 针对给定的EURUSD PH数组，设置阈值为0.003时，会将数值接近的枢轴高点归为同一聚类，对应同一阻力区间，便于后续绘制支撑阻力通道

内容的提问来源于stack exchange，提问作者PepeW