Analysis_Multi/app/services/HealthAssessor.py

from functools import lru_cache
from typing import Dict, List

import numpy as np
import pandas as pd
from sklearn.neighbors import BallTree

from app.logger import logger


class HealthAssessor:
    def __init__(self):
        self.subsystem_config = {
            # 发电机
            'generator': {
                # 双馈
                'dfig': {
                    'fixed': ['generator_winding1_temperature', 'generator_winding2_temperature',
                              'generator_winding3_temperature', 'generatordrive_end_bearing_temperature',
                              'generatornon_drive_end_bearing_temperature'],
                },
                # 直驱
                'direct': {
                    'fixed': ['generator_winding1_temperature', 'generator_winding2_temperature',
                              'generator_winding3_temperature', 'main_bearing_temperature'],
                }
            },
            # 机舱系统
            'nacelle': {
                'fixed': ['front_back_vibration_of_the_cabin', 'side_to_side_vibration_of_the_cabin',
                          'cabin_position', 'cabin_temperature'],
            },
            # 电网环境
            'grid': {
                'fixed': ['reactive_power', 'active_power', 'grid_a_phase_current',
                          'grid_b_phase_current', 'grid_c_phase_current'],
            },
            # 传动系统
            'drive_train': {
                'fixed': ['main_bearing_temperature'],
                'keywords': [
                    {'include': ['gearbox', 'temperature'], 'exclude': [], 'min_count': 2},
                ]
            }
        }

        # 嵌入源代码的MSET实现
        self.mset = self._create_mset_core()

    def _create_mset_core(self):
        """创建MSET核心计算模块"""
        class MSETCore:
            def __init__(self):
                self.matrixD = None
                self.normalDataBallTree = None
                self.healthyResidual = None

            def calcSimilarity(self, x, y):
                """优化后的相似度计算"""
                diff = np.array(x) - np.array(y)
                return 1/(1 + np.sqrt(np.sum(diff**2)))

            def genDLMatrix(self, trainDataset, dataSize4D=60, dataSize4L=5):
                """优化矩阵生成过程"""
                m, n = trainDataset.shape

                # 快速选择极值点
                min_indices = np.argmin(trainDataset, axis=0)
                max_indices = np.argmax(trainDataset, axis=0)
                unique_indices = np.unique(np.concatenate([min_indices, max_indices]))
                self.matrixD = trainDataset[unique_indices].copy()

                # 快速填充剩余点
                remaining_indices = np.setdiff1d(np.arange(m), unique_indices)
                np.random.shuffle(remaining_indices)
                needed = max(0, dataSize4D - len(unique_indices))
                if needed > 0:
                    self.matrixD = np.vstack([self.matrixD, trainDataset[remaining_indices[:needed]]])

                # 使用与源代码一致的BallTree参数
                self.normalDataBallTree = BallTree(
                    self.matrixD,
                    leaf_size=4,
                    metric=lambda i,j: 1-self.calcSimilarity(i,j)  # 自定义相似度
                )

                # 使用所有数据计算残差
                self.healthyResidual = self.calcResidualByLocallyWeightedLR(trainDataset)
                return 0

            def calcResidualByLocallyWeightedLR(self, newStates):
                """优化残差计算"""
                if len(newStates.shape) == 1:
                    newStates = newStates.reshape(-1, 1)

                dist, iList = self.normalDataBallTree.query(
                    newStates,
                    k=min(10, len(self.matrixD)),
                    return_distance=True
                )
                weights = 1/(dist + 1e-5)
                weights /= weights.sum(axis=1)[:, np.newaxis]

                est_X = np.sum(weights[:, :, np.newaxis] * self.matrixD[iList[0]], axis=1)
                return est_X - newStates

            def calcSPRT(self, newsStates, feature_weight, alpha=0.1, beta=0.1, decisionGroup=1):
                """优化SPRT计算"""
                stateResidual = self.calcResidualByLocallyWeightedLR(newsStates)
                weightedStateResidual = np.dot(stateResidual, feature_weight)
                weightedHealthyResidual = np.dot(self.healthyResidual, feature_weight)

                mu0 = np.mean(weightedHealthyResidual)
                sigma0 = np.std(weightedHealthyResidual)

                # 向量化计算
                n = len(newsStates)
                if n < decisionGroup:
                    return [50]  # 中性值

                rolling_mean = np.convolve(weightedStateResidual, np.ones(decisionGroup)/decisionGroup, 'valid')
                si = (rolling_mean - mu0) * (rolling_mean + mu0 - 2*mu0) / (2*sigma0**2)

                lowThres = np.log(beta/(1-alpha))
                highThres = np.log((1-beta)/alpha)

                si = np.clip(si, lowThres, highThres)
                si = np.where(si > 0, si/highThres, si/lowThres)
                flag = 100 - si*100

                # 填充不足的部分
                if len(flag) < n:
                    flag = np.pad(flag, (0, n-len(flag)), mode='edge')

                return flag.tolist()

            def CRITIC_prepare(self, data, flag=1):
                """标准化处理"""
                data = data.astype(float)
                numeric_cols = data.select_dtypes(include=[np.number]).columns
                #需要确认哪些指标是正向标准化 哪些是负向标准化
                negative_cols = [col for col in numeric_cols
                                 if any(kw in col for kw in ['temperature'])]
                positive_cols = list(set(numeric_cols) - set(negative_cols))

                # 负向标准化
                if negative_cols:
                    max_val = data[negative_cols].max()
                    min_val = data[negative_cols].min()
                    data[negative_cols] = (max_val - data[negative_cols]) / (max_val - min_val).replace(0, 1e-5)

                # 正向标准化
                if positive_cols:
                    max_val = data[positive_cols].max()
                    min_val = data[positive_cols].min()
                    data[positive_cols] = (data[positive_cols] - min_val) / (max_val - min_val).replace(0, 1e-5)

                return data

            def CRITIC(self, data):
                """CRITIC权重计算"""
                data_norm = self.CRITIC_prepare(data.copy())
                std = data_norm.std(ddof=0).clip(lower=0.01)
                corr = np.abs(np.corrcoef(data_norm.T))
                np.fill_diagonal(corr, 0)
                conflict = np.sum(1 - corr, axis=1)
                info = std * conflict
                weights = info / info.sum()
                return pd.Series(weights, index=data.columns)

            def ahp(self, matrix):
                """AHP权重计算"""
                eigenvalue, eigenvector = np.linalg.eig(matrix)
                max_idx = np.argmax(eigenvalue)
                weight = eigenvector[:, max_idx].real
                return weight / weight.sum(), eigenvalue[max_idx].real

        return MSETCore()


    def assess_turbine(self, engine_code, data, mill_type, wind_turbine_name):
        """评估单个风机
        """
        results = {
            "engine_code": engine_code,
            "wind_turbine_name": wind_turbine_name,
            "mill_type": mill_type,
            "total_health_score": None,
            "subsystems": {},
            "assessed_subsystems": []
        }

        # 各子系统评估
        subsystems_to_assess = [
            ('generator', self.subsystem_config['generator'][mill_type], 1),
            ('nacelle', self.subsystem_config['nacelle'],1),
            ('grid', self.subsystem_config['grid'], 1),
            ('drive_train', self.subsystem_config['drive_train'] if mill_type == 'dfig' else None,1)
        ]

        for subsystem, config, min_features in subsystems_to_assess:
            if config is None:
                continue

            features = self._get_subsystem_features(config, data)

            # 功能1：无论特征数量是否足够都输出结果
            if len(features) >= min_features:
                assessment = self._assess_subsystem(data[features])
            else:
                assessment = {
                    'health_score': -1,  # 特征不足时输出'-'
                    'weights': {},
                    'message': f'Insufficient features (required {min_features}, got {len(features)})'
                }

            # 功能3：删除features内容
            if 'features' in assessment:
                del assessment['features']

            results["subsystems"][subsystem] = assessment

        # 计算整机健康度（使用新字段名）
        if results["subsystems"]:
            # 只计算健康值为数字的子系统
            valid_subsystems = [
                k for k, v in results["subsystems"].items()
                if isinstance(v['health_score'], (int, float)) and v['health_score'] >= 0
            ]

            if valid_subsystems:
                weights = self._get_subsystem_weights(valid_subsystems)
                health_scores = [results["subsystems"][sys]['health_score'] for sys in valid_subsystems]
                results["total_health_score"] = float(np.dot(health_scores, weights))
                results["assessed_subsystems"] = valid_subsystems

        return results




    def _get_all_possible_features(self,assessor, mill_type, available_columns):
        """
        获取所有可能的特征列（基于实际存在的列）

        参数:
            assessor: HealthAssessor实例
            mill_type: 机型类型
            available_columns: 数据库实际存在的列名列表
        """
        features = []
        available_columns_lower = [col.lower() for col in available_columns]  # 不区分大小写匹配

        for subsys_name, subsys_config in assessor.subsystem_config.items():
            # 处理子系统配置
            if subsys_name == 'generator':
                config = subsys_config.get(mill_type, {})
            elif subsys_name == 'drive_train' and mill_type != 'dfig':
                continue
            else:
                config = subsys_config

            # 处理固定特征
            if 'fixed' in config:
                for f in config['fixed']:
                    if f in available_columns:
                        features.append(f)

            # 处理关键词特征
            if 'keywords' in config:
                for rule in config['keywords']:
                    matched = []
                    include_kws = [kw.lower() for kw in rule['include']]
                    exclude_kws = [ex.lower() for ex in rule.get('exclude', [])]

                    for col in available_columns:
                        col_lower = col.lower()
                        # 检查包含关键词
                        include_ok = all(kw in col_lower for kw in include_kws)
                        # 检查排除关键词
                        exclude_ok = not any(ex in col_lower for ex in exclude_kws)

                        if include_ok and exclude_ok:
                            matched.append(col)

                    if len(matched) >= rule.get('min_count', 1):
                        features.extend(matched)

        return list(set(features))  # 去重

    def _get_subsystem_features(self, config: Dict, data: pd.DataFrame) -> List[str]:
        """最终版特征获取方法"""
        available_features = []

        # 固定特征检查（要求至少10%非空）
        if 'fixed' in config:
            for f in config['fixed']:
                if f in data.columns and data[f].notna().mean() > 0.1:
                    available_features.append(f)
        logger.info(f"匹配到的固定特征: {available_features}")
        # 关键词特征检查
        if 'keywords' in config:
            for rule in config['keywords']:
                matched = [
                    col for col in data.columns
                    if all(kw.lower() in col.lower() for kw in rule['include'])
                       and not any(ex.lower() in col.lower() for ex in rule.get('exclude', []))
                       and data[col].notna().mean() > 0.1  # 数据有效性检查
                ]
                if len(matched) >= rule.get('min_count', 1):
                    available_features.extend(matched)
        logger.info(f"匹配到的关键词特征: {available_features}")
        return list(set(available_features))

    def _assess_subsystem(self, data: pd.DataFrame) -> Dict:
        """评估子系统（与源代码逻辑完全一致）"""
        # 数据清洗
        clean_data = data.dropna()
        if len(clean_data) < 20:  # 数据量不足
            return {'health_score': -1, 'weights': {}, 'features': list(data.columns), 'message': 'Insufficient data'}

        try:
            # 标准化
            normalized_data = self.mset.CRITIC_prepare(clean_data)

            # 计算权重
            weights = self.mset.CRITIC(normalized_data)

            # MSET评估
            health_score = self._run_mset_assessment(normalized_data.values, weights.values)

            return {
                'health_score': float(health_score),
                'weights': weights.to_dict(),
                'features': list(data.columns)
            }
        except Exception as e:
            return {'health_score': -1, 'weights': {}, 'features': list(data.columns), 'message': str(e)}

    @lru_cache(maxsize=10)
    def _get_mset_model(self, train_data: tuple):
        """缓存MSET模型"""
        # 注意：由于lru_cache需要可哈希参数，这里使用元组
        arr = np.array(train_data)
        model = self._create_mset_core()
        model.genDLMatrix(arr)
        return model

    def _run_mset_assessment(self, data: np.ndarray, weights: np.ndarray) -> float:
        """执行MSET评估"""
        # 分割训练集和测试集
        split_idx = len(data) // 2
        train_data = data[:split_idx]
        test_data = data[split_idx:]

        # 使用缓存模型
        model = self._get_mset_model(tuple(map(tuple, train_data)))  # 转换为可哈希的元组

        # 计算SPRT标志
        flags = model.calcSPRT(test_data, weights)
        return np.mean(flags)

    def _get_subsystem_weights(self, subsystems: List[str]) -> np.ndarray:
        """生成等权重的子系统权重向量"""
        n = len(subsystems)
        if n == 0:
            return np.array([])

        # 直接返回等权重向量
        return np.ones(n) / n