📊

基于时间、价格和成交量创建 3D 柱状图引入波动率测量·进阶篇

（2/3）· 二维图表漏掉的真实市场结构，靠体积柱与波动率维度在 MT5 里补回来

新手友好第 2/3 篇

很多交易者把两根外形一样的 K 线当成同一种信号，却没注意背后成交量厚度可能差出数倍。把价格和成交量压成平面看，容易在假突破上吃亏。用三维视角拆开柱体，支撑力度一眼就能分出来。

本章目录

砖型块的方向判定与字段组装
砖形块派生指标的归一与统计加工
用四维平稳特征替换简单缩放
把K线拆成价格/成交量/波动三轴特征
特征缩放与三维可视化的落地细节
把量价波动塞进一张子图

◍ 砖型块的方向判定与字段组装

在逐笔合成 Renko 块时，先按价格差除以最小砖高得到 price_bricks，再用成交量砖数与价格砖数是否大于 0 作为触发条件。只要 volume_bricks > 0 或 abs(price_bricks) > 0，就认为本根应生成新砖块，避免零波动时刷出空结构。 direction 的取法很直接：price_bricks 非零时取 np.sign 得到多空方向，若价格没动则默认补 1（多头）。接着比对 prev_direction：同向就 trend_count + 1，反向则重置为 1，这个计数可直接用于识别连砖长度。每块把时间、OHLC、tick_volume、spread、type 以及上面算出的 direction / trend_count / price_change 全写进字典。volume_intensity 用本笔 tick_volume 除以 volume_brick，price_velocity 用 price_diff 除以实际消耗的 volume_bricks（为防零除，volume_bricks 为 0 时分母取 1）。收尾处若 volume_bricks > 0 就把 current_tick_volume 对 volume_brick 取模，price_bricks 非零则把 current_price 累加 min_price_brick * price_bricks，最后把 direction 赋给 prev_direction 供下一根使用。外汇与贵金属波动受杠杆与跳空影响大，连砖统计仅作概率参考，实盘前请在 MT5 用历史数据复算一遍。

MQL5 / C++

price_bricks = class="type">int(price_diff / min_price_brick)

if volume_bricks > class="num">0 or abs(price_bricks) > class="num">0:
    direction = np.sign(price_bricks) if price_bricks != class="num">0 else class="num">1

    if direction == prev_direction:
        trend_count += class="num">1
    else:
        trend_count = class="num">1

    renko_block = {
        &class="macro">#x27;time&class="macro">#x27;: current_time,
        &class="macro">#x27;time_numeric&class="macro">#x27;: class="type">class="kw">float(row[&class="macro">#x27;time_numeric&class="macro">#x27;]),
        &class="macro">#x27;open&class="macro">#x27;: class="type">class="kw">float(row[&class="macro">#x27;open&class="macro">#x27;]),
        &class="macro">#x27;close&class="macro">#x27;: class="type">class="kw">float(row[&class="macro">#x27;close&class="macro">#x27;]),
        &class="macro">#x27;high&class="macro">#x27;: class="type">class="kw">float(row[&class="macro">#x27;high&class="macro">#x27;]),
        &class="macro">#x27;low&class="macro">#x27;: class="type">class="kw">float(row[&class="macro">#x27;low&class="macro">#x27;]),
        &class="macro">#x27;tick_volume&class="macro">#x27;: class="type">class="kw">float(row[&class="macro">#x27;tick_volume&class="macro">#x27;]),
        &class="macro">#x27;direction&class="macro">#x27;: class="type">class="kw">float(direction),
        &class="macro">#x27;spread&class="macro">#x27;: class="type">class="kw">float(current_spread),
        &class="macro">#x27;type&class="macro">#x27;: class="type">class="kw">float(current_type),
        &class="macro">#x27;trend_count&class="macro">#x27;: trend_count,
        &class="macro">#x27;price_change&class="macro">#x27;: price_diff,
        &class="macro">#x27;volume_intensity&class="macro">#x27;: class="type">class="kw">float(row[&class="macro">#x27;tick_volume&class="macro">#x27;]) / volume_brick,
        &class="macro">#x27;price_velocity&class="macro">#x27;: price_diff / (volume_bricks if volume_bricks > class="num">0 else class="num">1)
    }

    if volume_bricks > class="num">0:
        current_tick_volume = current_tick_volume % volume_brick
    if price_bricks != class="num">0:
        current_price += min_price_brick * price_bricks

    prev_direction = direction

「砖形块派生指标的归一与统计加工」

砖形（Renko）块列表生成后若为空，函数直接返回 None，避免在 MT5 外接分析里拿空帧去跑模型。拿到非空列表先转成 DataFrame，对 price_change、volume_intensity、price_velocity、spread 四个派生列做 fit_transform 缩放，把量纲不同的指标压到同一数值区间，否则后续滚动均值会被成交量级带偏。紧接着用缩放后的数据补分析列：close 的 5 与 20 周期滚动均线是趋势骨架，tick_volume 的 5 周期均线和 10 周期 std 分别看量能中枢与波动，trend_strength 用 trend_count 乘 direction 表达带符号的趋势强度。这些 MA 与波动率列再走一遍 scaler，最后用 scipy.stats.zscore 算 close 与 tick_volume 的标准分（nan_policy='omit' 跳过空值），z 分数也归一后一并返回 result_df 与 min_price_brick。外汇与贵金属市场高风险，这类缩放仅服务于特征一致性，不预示任何方向。

MQL5 / C++

    renko_blocks.append(renko_block)
    
    except Exception as e:
        print(f"Error processing data: {e}")
        if len(renko_blocks) == class="num">0:
            class="kw">return None, None
    
    if len(renko_blocks) == class="num">0:
        print("Failed to create any blocks")
        class="kw">return None, None
        
    result_df = pd.DataFrame(renko_blocks)
    
    # Scale derived metrics to same range
    derived_metrics = [&class="macro">#x27;price_change&class="macro">#x27;, &class="macro">#x27;volume_intensity&class="macro">#x27;, &class="macro">#x27;price_velocity&class="macro">#x27;, &class="macro">#x27;spread&class="macro">#x27;]
    result_df[derived_metrics] = scaler.fit_transform(result_df[derived_metrics])
    
    # Add analytical metrics using scaled data
    result_df[&class="macro">#x27;ma_5&class="macro">#x27;] = result_df[&class="macro">#x27;close&class="macro">#x27;].rolling(class="num">5).mean()
    result_df[&class="macro">#x27;ma_20&class="macro">#x27;] = result_df[&class="macro">#x27;close&class="macro">#x27;].rolling(class="num">20).mean()
    result_df[&class="macro">#x27;volume_ma_5&class="macro">#x27;] = result_df[&class="macro">#x27;tick_volume&class="macro">#x27;].rolling(class="num">5).mean()
    result_df[&class="macro">#x27;price_volatility&class="macro">#x27;] = result_df[&class="macro">#x27;price_change&class="macro">#x27;].rolling(class="num">10).std()
    result_df[&class="macro">#x27;volume_volatility&class="macro">#x27;] = result_df[&class="macro">#x27;tick_volume&class="macro">#x27;].rolling(class="num">10).std()
    result_df[&class="macro">#x27;trend_strength&class="macro">#x27;] = result_df[&class="macro">#x27;trend_count&class="macro">#x27;] * result_df[&class="macro">#x27;direction&class="macro">#x27;]
    
    # Scale moving averages and volatility
    ma_columns = [&class="macro">#x27;ma_5&class="macro">#x27;, &class="macro">#x27;ma_20&class="macro">#x27;, &class="macro">#x27;volume_ma_5&class="macro">#x27;, &class="macro">#x27;price_volatility&class="macro">#x27;, &class="macro">#x27;volume_volatility&class="macro">#x27;, &class="macro">#x27;trend_strength&class="macro">#x27;]
    result_df[ma_columns] = scaler.fit_transform(result_df[ma_columns])
    
    # Add statistical metrics and scale them
    result_df[&class="macro">#x27;zscore_price&class="macro">#x27;] = stats.zscore(result_df[&class="macro">#x27;close&class="macro">#x27;], nan_policy=&class="macro">#x27;omit&class="macro">#x27;)
    result_df[&class="macro">#x27;zscore_volume&class="macro">#x27;] = stats.zscore(result_df[&class="macro">#x27;tick_volume&class="macro">#x27;], nan_policy=&class="macro">#x27;omit&class="macro">#x27;)
    zscore_columns = [&class="macro">#x27;zscore_price&class="macro">#x27;, &class="macro">#x27;zscore_volume&class="macro">#x27;]
    result_df[zscore_columns] = scaler.fit_transform(result_df[zscore_columns])
    
    class="kw">return result_df, min_price_brick

用四维平稳特征替换简单缩放

做特征工程时，直接把价格缩到 3-9 区间只是表面处理，序列本身并不平稳。换一套做法：对时间、价格、成交量、波动率四个维度分别做变换，让每个维度在统计意义下保持平稳，同时还能接回原来的 3D 砖形图接口。时间维度用三角变换消掉日周期：sin(2π*hour/24) 与 cos(2π*hour/24) 把 0-23 点映射成连续波动，避免零点跳变。价格维度不碰绝对值，改算典型价 (H+L+C)/3 的回报率和加速度，任意价格水位下分布都可比。成交量分布极不均匀，连续套用 pct_change() 再 diff() 取相对增量，比单纯看量变更抗畸变。波动率走两步：先算回报率的滚动标准差，再取这个标准差的相对变化，得到「波动率的波动率」，对尾部风险更敏感。滑动窗口定在 20 周期，不是拍脑袋——短了统计不显著，长了吃掉局部结构，20 是折中。最终平稳序列才二次缩放到 3-9，纯粹为兼容老实现，不改变平稳性。原函数的 MA、波动率、z-score 全部保留，新函数能直接替掉老函数。外汇与贵金属波动剧烈、杠杆高风险突出，这类平稳特征只降低建模偏误，不预示方向。下面这段 Python 对接 MT5 拉数并构造四维特征，开 MT5 终端跑一遍即可验证接口与输出范围。

MQL5 / C++

def create_true_3d_renko(symbol, timeframe, min_spread_multiplier=class="num">45, volume_brick=class="num">500, lookback=class="num">20000):
    """
    Creates 4D stationary features with same interface as 3D Renko
    """
    rates = mt5.copy_rates_from_pos(symbol, timeframe, class="num">0, lookback)
    if rates is None:
        print(f"Error getting data for {symbol}")
        class="kw">return None, None
        
    df = pd.DataFrame(rates)
    df[&class="macro">#x27;time&class="macro">#x27;] = pd.to_datetime(df[&class="macro">#x27;time&class="macro">#x27;], unit=&class="macro">#x27;s&class="macro">#x27;)
    
    if df.isnull().any().any():
        print("Missing values detected, cleaning...")
        df = df.dropna()
        if len(df) == class="num">0:
            print("No data for analysis after cleaning")
            class="kw">return None, None
    
    symbol_info = mt5.symbol_info(symbol)
    if symbol_info is None:
        print(f"Failed to get symbol info for {symbol}")
        class="kw">return None, None
    
    try:
        min_price_brick = symbol_info.spread * min_spread_multiplier * symbol_info.point
        if min_price_brick <= class="num">0:
            print("Invalid block size")
            class="kw">return None, None
    except AttributeError as e:
        print(f"Error getting symbol parameters: {e}")
        class="kw">return None, None
    
    scaler = MinMaxScaler(feature_range=(class="num">3, class="num">9))
    df_blocks = []
    
    try:
        # Time dimension
        df[&class="macro">#x27;time_sin&class="macro">#x27;] = np.sin(class="num">2 * np.pi * df[&class="macro">#x27;time&class="macro">#x27;].dt.hour / class="num">24)
        df[&class="macro">#x27;time_cos&class="macro">#x27;] = np.cos(class="num">2 * np.pi * df[&class="macro">#x27;time&class="macro">#x27;].dt.hour / class="num">24)
        df[&class="macro">#x27;time_numeric&class="macro">#x27;] = (df[&class="macro">#x27;time&class="macro">#x27;] - df[&class="macro">#x27;time&class="macro">#x27;].min()).dt.total_seconds()

◍ 把K线拆成价格/成交量/波动三轴特征

做行情结构化时，先把单根K线映射成可计算的三个维度：价格、成交量、波动率。典型价取 (高+低+收)/3，再算一阶收益率和二阶加速度，能把肉眼看不出的拐弯提前量化出来。成交量轴用 tick_volume 的 pct_change 做一阶变化，再 diff 一次得加速度；波动轴则对价格收益率取 20 根滚动标准差，再算其环比变化。20 这个窗口在 MT5 的 M1/M5 上回测过，太短噪声大、太长滞后明显。滑窗拼块时，每个 block 取当前柱和前 20 根构成窗口，把末根的价格加速度、量能变化、波动变化塞进 open/high/low/close 字段做特征对齐。direction 用 np.sign 取末根收益符号，标记偏多或偏空倾向。外汇和贵金属杠杆高、滑点跳空频繁，这类特征在重大数据行情里可能失真，上 MT5 用历史中心跑一遍再信。

MQL5 / C++

# Price dimension
df[&class="macro">#x27;typical_price&class="macro">#x27;] = (df[&class="macro">#x27;high&class="macro">#x27;] + df[&class="macro">#x27;low&class="macro">#x27;] + df[&class="macro">#x27;close&class="macro">#x27;]) / class="num">3
df[&class="macro">#x27;price_return&class="macro">#x27;] = df[&class="macro">#x27;typical_price&class="macro">#x27;].pct_change()
df[&class="macro">#x27;price_acceleration&class="macro">#x27;] = df[&class="macro">#x27;price_return&class="macro">#x27;].diff()

# Volume dimension
df[&class="macro">#x27;volume_change&class="macro">#x27;] = df[&class="macro">#x27;tick_volume&class="macro">#x27;].pct_change()
df[&class="macro">#x27;volume_acceleration&class="macro">#x27;] = df[&class="macro">#x27;volume_change&class="macro">#x27;].diff()

# Volatility dimension
df[&class="macro">#x27;volatility&class="macro">#x27;] = df[&class="macro">#x27;price_return&class="macro">#x27;].rolling(class="num">20).std()
df[&class="macro">#x27;volatility_change&class="macro">#x27;] = df[&class="macro">#x27;volatility&class="macro">#x27;].pct_change()

for idx in range(class="num">20, len(df)):
    window = df.iloc[idx-class="num">20:idx+class="num">1]

    block = {
        &class="macro">#x27;time&class="macro">#x27;: df.iloc[idx][&class="macro">#x27;time&class="macro">#x27;],
        &class="macro">#x27;time_numeric&class="macro">#x27;: scaler.fit_transform([[class="type">class="kw">float(df.iloc[idx][&class="macro">#x27;time_numeric&class="macro">#x27;])]]).item(),
        &class="macro">#x27;open&class="macro">#x27;: class="type">class="kw">float(window[&class="macro">#x27;price_return&class="macro">#x27;].iloc[-class="num">1]),
        &class="macro">#x27;high&class="macro">#x27;: class="type">class="kw">float(window[&class="macro">#x27;price_acceleration&class="macro">#x27;].iloc[-class="num">1]),
        &class="macro">#x27;low&class="macro">#x27;: class="type">class="kw">float(window[&class="macro">#x27;volume_change&class="macro">#x27;].iloc[-class="num">1]),
        &class="macro">#x27;close&class="macro">#x27;: class="type">class="kw">float(window[&class="macro">#x27;volatility_change&class="macro">#x27;].iloc[-class="num">1]),
        &class="macro">#x27;tick_volume&class="macro">#x27;: class="type">class="kw">float(window[&class="macro">#x27;volume_acceleration&class="macro">#x27;].iloc[-class="num">1]),
        &class="macro">#x27;direction&class="macro">#x27;: np.sign(window[&class="macro">#x27;price_return&class="macro">#x27;].iloc[-class="num">1]),
        &class="macro">#x27;spread&class="macro">#x27;: class="type">class="kw">float(df.iloc[idx][&class="macro">#x27;time_sin&class="macro">#x27;]),
        &class="macro">#x27;type&class="macro">#x27;: class="type">class="kw">float(df.iloc[idx][&class="macro">#x27;time_cos&class="macro">#x27;]),
        &class="macro">#x27;trend_count&class="macro">#x27;: len(window),
        &class="macro">#x27;price_change&class="macro">#x27;: class="type">class="kw">float(window[&class="macro">#x27;price_return&class="macro">#x27;].mean()),
        &class="macro">#x27;volume_intensity&class="macro">#x27;: class="type">class="kw">float(window[&class="macro">#x27;volume_change&class="macro">#x27;].mean()),
        &class="macro">#x27;price_velocity&class="macro">#x27;: class="type">class="kw">float(window[&class="macro">#x27;price_acceleration&class="macro">#x27;].mean())
    }
    df_blocks.append(block)

except Exception as e:

「特征缩放与三维可视化的落地细节」

把分块数据整理成 DataFrame 后，先排除 time 与 direction 两列，对其余所有特征做标准化。这一步用 scaler.fit_transform 直接覆盖原列，意味着后续模型喂入的是无量纲数据，避免 close 和 tick_volume 量纲差异干扰聚类或分类结果。标准化之后补一组与原函数一致的分析指标：ma_5 用 close 的 5 周期滚动均值，ma_20 取 20 周期；volume_ma_5 是 tick_volume 的 5 周期均值。price_volatility 与 volume_volatility 分别取 price_change、tick_volume 的 10 周期滚动标准差，trend_strength 则由 trend_count 乘 direction 得到，方向加权后趋势强度有正负区分。这批衍生列再走一轮 fit_transform，和 zscore_price、zscore_volume（用 stats.zscore 且 nan_policy='omit' 忽略空值）一起二次缩放。实测中若 df_blocks 为空，函数会打印 Failed to create any blocks 并返回 None, None，调用方必须判空，否则下游绘图会抛异常。可视化侧用 plotly 的 make_subplots 开 1 行 2 列：左边放三维视角，右边贴原始价格图。close、tick_volume、price_volatility、volume_volatility 这四列先各自算 window=100 的滚动均值做平滑，min_periods=1 保证前 99 根也有值，不会开头断档。外汇与贵金属波动剧烈，三维重构仅作形态辅助，实盘仍属高风险，信号失效概率不低。

MQL5 / C++

print(f"Error processing data: {e}")
if len(df_blocks) == class="num">0:
    class="kw">return None, None

if len(df_blocks) == class="num">0:
    print("Failed to create any blocks")
    class="kw">return None, None

result_df = pd.DataFrame(df_blocks)

# Scale all features
features_to_scale = [col for col in result_df.columns if col != &class="macro">#x27;time&class="macro">#x27; and col != &class="macro">#x27;direction&class="macro">#x27;]
result_df[features_to_scale] = scaler.fit_transform(result_df[features_to_scale])

# Add same analytical metrics as in original function
result_df[&class="macro">#x27;ma_5&class="macro">#x27;] = result_df[&class="macro">#x27;close&class="macro">#x27;].rolling(class="num">5).mean()
result_df[&class="macro">#x27;ma_20&class="macro">#x27;] = result_df[&class="macro">#x27;close&class="macro">#x27;].rolling(class="num">20).mean()
result_df[&class="macro">#x27;volume_ma_5&class="macro">#x27;] = result_df[&class="macro">#x27;tick_volume&class="macro">#x27;].rolling(class="num">5).mean()
result_df[&class="macro">#x27;price_volatility&class="macro">#x27;] = result_df[&class="macro">#x27;price_change&class="macro">#x27;].rolling(class="num">10).std()
result_df[&class="macro">#x27;volume_volatility&class="macro">#x27;] = result_df[&class="macro">#x27;tick_volume&class="macro">#x27;].rolling(class="num">10).std()
result_df[&class="macro">#x27;trend_strength&class="macro">#x27;] = result_df[&class="macro">#x27;trend_count&class="macro">#x27;] * result_df[&class="macro">#x27;direction&class="macro">#x27;]

# Scale moving averages and volatility
ma_columns = [&class="macro">#x27;ma_5&class="macro">#x27;, &class="macro">#x27;ma_20&class="macro">#x27;, &class="macro">#x27;volume_ma_5&class="macro">#x27;, &class="macro">#x27;price_volatility&class="macro">#x27;, &class="macro">#x27;volume_volatility&class="macro">#x27;, &class="macro">#x27;trend_strength&class="macro">#x27;]
result_df[ma_columns] = scaler.fit_transform(result_df[ma_columns])

# Add statistical metrics and scale them
result_df[&class="macro">#x27;zscore_price&class="macro">#x27;] = stats.zscore(result_df[&class="macro">#x27;close&class="macro">#x27;], nan_policy=&class="macro">#x27;omit&class="macro">#x27;)
result_df[&class="macro">#x27;zscore_volume&class="macro">#x27;] = stats.zscore(result_df[&class="macro">#x27;tick_volume&class="macro">#x27;], nan_policy=&class="macro">#x27;omit&class="macro">#x27;)
zscore_columns = [&class="macro">#x27;zscore_price&class="macro">#x27;, &class="macro">#x27;zscore_volume&class="macro">#x27;]
result_df[zscore_columns] = scaler.fit_transform(result_df[zscore_columns])

class="kw">return result_df, min_price_brick
class="kw">import plotly.graph_objects as go
from plotly.subplots class="kw">import make_subplots
def create_interactive_3d(df, symbol, save_dir):
    """
    Creates interactive 3D visualization with smoothed data and original price chart
    """
    try:
        save_dir = Path(save_dir)
        
        # Smooth all series with MA(class="num">100)
        df_smooth = df.copy()
        smooth_columns = [&class="macro">#x27;close&class="macro">#x27;, &class="macro">#x27;tick_volume&class="macro">#x27;, &class="macro">#x27;price_volatility&class="macro">#x27;, &class="macro">#x27;volume_volatility&class="macro">#x27;]
        
        for col in smooth_columns:
            df_smooth[f&class="macro">#x27;{col}_smooth&class="macro">#x27;] = df_smooth[col].rolling(window=class="num">100, min_periods=class="num">1).mean()
        
        # Create subplots: 3D view and original chart side by side
        fig = make_subplots(
            rows=class="num">1, cols=class="num">2,

把量价波动塞进一张子图

上面的片段把 3D 散点图和原始 K 线并排画进同一个 Figure，左格看 MA100 平滑后的量价三维结构，右格留原始 OHLC 做对照。horizontal_spacing 设成 0.05，两个子图横向留白只有 5% 宽度，小屏上也挤得下。 3D 散点里 x 是序列索引、y 是 tick_volume_smooth、z 是 close_smooth，marker 用 price_volatility_smooth 映射 Viridis 色阶，size=5、opacity=0.8。colorbar 推到 x=0.45 位置，避免压到左图坐标轴。 hovertemplate 写死了四位精度：Price 留 5 位小数（贵金属 XAUUSD 点数小，必看 0.0000x），Volatility 同样 5 位。右格 Candlestick 直接用未平滑的 df open/high/low/close，和左格平滑线形成肉眼可辨的滞后差。跑这段代码前确认 df_smooth 已含 close_smooth、tick_volume_smooth、price_volatility_smooth 三列，否则 add_trace 会报 KeyError。外汇和贵金属波动剧烈，这种图只辅助看结构，不构成方向判断。

MQL5 / C++

specs=[[{&class="macro">#x27;type&class="macro">#x27;: &class="macro">#x27;scene&class="macro">#x27;}, {&class="macro">#x27;type&class="macro">#x27;: &class="macro">#x27;xy&class="macro">#x27;}]],
subplot_titles=(f&class="macro">#x27;{symbol} 3D View(MA100)&class="macro">#x27;, f&class="macro">#x27;{symbol} Original Price&class="macro">#x27;),
horizontal_spacing=class="num">0.05
)

# Add 3D scatter plot
fig.add_trace(
    go.Scatter3d(
        x=np.arange(len(df_smooth)),
        y=df_smooth[&class="macro">#x27;tick_volume_smooth&class="macro">#x27;],
        z=df_smooth[&class="macro">#x27;close_smooth&class="macro">#x27;],
        mode=&class="macro">#x27;markers&class="macro">#x27;,
        marker=dict(
            size=class="num">5,
            class="type">class="kw">color=df_smooth[&class="macro">#x27;price_volatility_smooth&class="macro">#x27;],
            colorscale=&class="macro">#x27;Viridis&class="macro">#x27;,
            opacity=class="num">0.8,
            showscale=True,
            colorbar=dict(x=class="num">0.45)
        ),
        hovertemplate=
        "Time: %{x}<br>" +
        "Volume: %{y:.2f}<br>" +
        "Price: %{z:.5f}<br>" +
        "Volatility: %{marker.class="type">class="kw">color:.5f}",
        name=&class="macro">#x27;3D View&class="macro">#x27;
    ),
    row=class="num">1, col=class="num">1
)

# Add original price chart
fig.add_trace(
    go.Candlestick(
        x=np.arange(len(df)),
        open=df[&class="macro">#x27;open&class="macro">#x27;],
        high=df[&class="macro">#x27;high&class="macro">#x27;],
        low=df[&class="macro">#x27;low&class="macro">#x27;],
        close=df[&class="macro">#x27;close&class="macro">#x27;],
        name=&class="macro">#x27;OHLC&class="macro">#x27;
    ),
    row=class="num">1, col=class="num">2
)

# Add smoothed price line
fig.add_trace(
    go.Scatter(
        x=np.arange(len(df_smooth)),
        y=df_smooth[&class="macro">#x27;close_smooth&class="macro">#x27;],

交给小布盯盘看盘口

这些诊断小布盯盘的 AIGC 已内置，打开对应品种页即可看到体积柱的实时厚度与波动率预警，你专注决策就行。

←

基于时间、价格和成交量创建 3D 柱状图引入波动率测量（基础篇）

→

基于时间、价格和成交量创建 3D 柱状图引入波动率测量·综合运用

常见问题

普通 K 线只投影价格与时间，3D 柱把成交量作为深度维度，能显示价格区间内成交量分布与买卖不平衡，可能更早暴露走势真实性。

可以，小布盯盘对应品种页已内置基于成交量和时间维度的微观结构视图，省去自己跑 Python 联 MT5 的重复劳动。

柱体从第一笔 tick 就开始累积，体积结构反映当前真实交易动态，对微观波动率的捕捉倾向领先二维投影。

外汇贵金属属高风险品种，3D 度量仅提供概率倾向参考，不代表未来方向，实盘仍需严格风控。

关于多维组合与方程的完整讨论见《基于时间、价格和成交量创建 3D 柱状图引入波动率测量·基础篇》。

基于时间、价格和成交量创建 3D 柱状图引入波动率测量·进阶篇

◍ 砖型块的方向判定与字段组装

「砖形块派生指标的归一与统计加工」

用四维平稳特征替换简单缩放

◍ 把K线拆成价格/成交量/波动三轴特征

「特征缩放与三维可视化的落地细节」

把量价波动塞进一张子图

常见问题

相关阅读