assist/src/agent/intelligent_agent.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
智能Agent核心 - 集成大模型和智能决策
高效实现Agent的智能处理能力
"""

import logging
import asyncio
import json
from typing import Dict, Any, List, Optional, Tuple
from datetime import datetime
from dataclasses import dataclass
from enum import Enum

logger = logging.getLogger(__name__)

class ActionType(Enum):
    """动作类型枚举"""
    ALERT_RESPONSE = "alert_response"
    KNOWLEDGE_UPDATE = "knowledge_update"
    WORKORDER_CREATE = "workorder_create"
    SYSTEM_OPTIMIZE = "system_optimize"
    USER_NOTIFY = "user_notify"

class ConfidenceLevel(Enum):
    """置信度等级"""
    HIGH = "high"      # 高置信度 (>0.8)
    MEDIUM = "medium"  # 中等置信度 (0.5-0.8)
    LOW = "low"        # 低置信度 (<0.5)

@dataclass
class AgentAction:
    """Agent动作"""
    action_type: ActionType
    description: str
    priority: int  # 1-5, 5最高
    confidence: float  # 0-1
    parameters: Dict[str, Any]
    estimated_time: int  # 预计执行时间(秒)

@dataclass
class AlertContext:
    """预警上下文"""
    alert_id: str
    alert_type: str
    severity: str
    description: str
    affected_systems: List[str]
    metrics: Dict[str, Any]

@dataclass
class KnowledgeContext:
    """知识库上下文"""
    question: str
    answer: str
    confidence: float
    source: str
    category: str

class IntelligentAgent:
    """智能Agent核心"""
    
    def __init__(self, llm_client=None):
        self.llm_client = llm_client
        self.action_history = []
        self.learning_data = {}
        self.confidence_thresholds = {
            'high': 0.8,
            'medium': 0.5,
            'low': 0.3
        }
        
    async def process_alert(self, alert_context: AlertContext) -> List[AgentAction]:
        """处理预警信息，生成智能动作"""
        try:
            # 构建预警分析提示
            prompt = self._build_alert_analysis_prompt(alert_context)
            
            # 调用大模型分析
            analysis = await self._call_llm(prompt)
            
            # 解析动作
            actions = self._parse_alert_actions(analysis, alert_context)
            
            # 按优先级排序
            actions.sort(key=lambda x: x.priority, reverse=True)
            
            return actions
            
        except Exception as e:
            logger.error(f"处理预警失败: {e}")
            return [self._create_default_alert_action(alert_context)]
    
    async def process_knowledge_confidence(self, knowledge_context: KnowledgeContext) -> List[AgentAction]:
        """处理知识库置信度问题"""
        try:
            if knowledge_context.confidence >= self.confidence_thresholds['high']:
                return []  # 高置信度，无需处理
            
            # 构建知识增强提示
            prompt = self._build_knowledge_enhancement_prompt(knowledge_context)
            
            # 调用大模型增强知识
            enhancement = await self._call_llm(prompt)
            
            # 生成增强动作
            actions = self._parse_knowledge_actions(enhancement, knowledge_context)
            
            return actions
            
        except Exception as e:
            logger.error(f"处理知识库置信度失败: {e}")
            return [self._create_default_knowledge_action(knowledge_context)]
    
    async def execute_action(self, action: AgentAction) -> Dict[str, Any]:
        """执行Agent动作"""
        try:
            logger.info(f"执行Agent动作: {action.action_type.value} - {action.description}")
            
            if action.action_type == ActionType.ALERT_RESPONSE:
                return await self._execute_alert_response(action)
            elif action.action_type == ActionType.KNOWLEDGE_UPDATE:
                return await self._execute_knowledge_update(action)
            elif action.action_type == ActionType.WORKORDER_CREATE:
                return await self._execute_workorder_create(action)
            elif action.action_type == ActionType.SYSTEM_OPTIMIZE:
                return await self._execute_system_optimize(action)
            elif action.action_type == ActionType.USER_NOTIFY:
                return await self._execute_user_notify(action)
            else:
                return {"success": False, "error": "未知动作类型"}
                
        except Exception as e:
            logger.error(f"执行动作失败: {e}")
            return {"success": False, "error": str(e)}
    
    def _build_alert_analysis_prompt(self, alert_context: AlertContext) -> str:
        """构建预警分析提示"""
        return f"""
作为TSP智能助手，请分析以下预警信息并提供处理建议：

预警信息：
- 类型: {alert_context.alert_type}
- 严重程度: {alert_context.severity}
- 描述: {alert_context.description}
- 影响系统: {', '.join(alert_context.affected_systems)}
- 指标数据: {json.dumps(alert_context.metrics, ensure_ascii=False)}

请提供以下格式的JSON响应：
{{
    "analysis": "预警原因分析",
    "immediate_actions": [
        {{
            "action": "立即执行的动作",
            "priority": 5,
            "confidence": 0.9,
            "parameters": {{"key": "value"}}
        }}
    ],
    "follow_up_actions": [
        {{
            "action": "后续跟进动作",
            "priority": 3,
            "confidence": 0.7,
            "parameters": {{"key": "value"}}
        }}
    ],
    "prevention_measures": [
        "预防措施1",
        "预防措施2"
    ]
}}
"""
    
    def _build_knowledge_enhancement_prompt(self, knowledge_context: KnowledgeContext) -> str:
        """构建知识增强提示"""
        return f"""
作为TSP智能助手，请分析以下知识库条目并提供增强建议：

知识条目：
- 问题: {knowledge_context.question}
- 答案: {knowledge_context.answer}
- 置信度: {knowledge_context.confidence}
- 来源: {knowledge_context.source}
- 分类: {knowledge_context.category}

请提供以下格式的JSON响应：
{{
    "confidence_analysis": "置信度分析",
    "enhancement_suggestions": [
        "增强建议1",
        "增强建议2"
    ],
    "actions": [
        {{
            "action": "知识更新动作",
            "priority": 4,
            "confidence": 0.8,
            "parameters": {{"enhanced_answer": "增强后的答案"}}
        }}
    ],
    "learning_opportunities": [
        "学习机会1",
        "学习机会2"
    ]
}}
"""
    
    async def _call_llm(self, prompt: str) -> Dict[str, Any]:
        """调用大模型"""
        try:
            if self.llm_client:
                # 使用真实的大模型客户端
                response = await self.llm_client.generate(prompt)
                return json.loads(response)
            else:
                # 模拟大模型响应
                return self._simulate_llm_response(prompt)
        except Exception as e:
            logger.error(f"调用大模型失败: {e}")
            return self._simulate_llm_response(prompt)
    
    def _simulate_llm_response(self, prompt: str) -> Dict[str, Any]:
        """模拟大模型响应 - 千问模型风格"""
        if "预警信息" in prompt:
            return {
                "analysis": "【千问分析】系统性能下降，需要立即处理。根据历史数据分析，这可能是由于资源不足或配置问题导致的。",
                "immediate_actions": [
                    {
                        "action": "重启相关服务",
                        "priority": 5,
                        "confidence": 0.9,
                        "parameters": {"service": "main_service", "reason": "服务响应超时"}
                    }
                ],
                "follow_up_actions": [
                    {
                        "action": "检查系统日志",
                        "priority": 3,
                        "confidence": 0.7,
                        "parameters": {"log_level": "error", "time_range": "last_hour"}
                    }
                ],
                "prevention_measures": [
                    "增加监控频率，提前发现问题",
                    "优化系统配置，提升性能",
                    "建立预警机制，减少故障影响"
                ]
            }
        else:
            return {
                "confidence_analysis": "【千问分析】当前答案置信度较低，需要更多上下文信息。建议结合用户反馈和历史工单数据来提升答案质量。",
                "enhancement_suggestions": [
                    "添加更多实际案例和操作步骤",
                    "提供详细的故障排除指南",
                    "结合系统架构图进行说明"
                ],
                "actions": [
                    {
                        "action": "更新知识库条目",
                        "priority": 4,
                        "confidence": 0.8,
                        "parameters": {"enhanced_answer": "基于千问模型分析的增强答案"}
                    }
                ],
                "learning_opportunities": [
                    "收集用户反馈，持续优化答案",
                    "分析相似问题，建立知识关联",
                    "利用千问模型的学习能力，提升知识质量"
                ]
            }
    
    def _parse_alert_actions(self, analysis: Dict[str, Any], alert_context: AlertContext) -> List[AgentAction]:
        """解析预警动作"""
        actions = []
        
        # 立即动作
        for action_data in analysis.get("immediate_actions", []):
            action = AgentAction(
                action_type=ActionType.ALERT_RESPONSE,
                description=action_data["action"],
                priority=action_data["priority"],
                confidence=action_data["confidence"],
                parameters=action_data["parameters"],
                estimated_time=30
            )
            actions.append(action)
        
        # 后续动作
        for action_data in analysis.get("follow_up_actions", []):
            action = AgentAction(
                action_type=ActionType.SYSTEM_OPTIMIZE,
                description=action_data["action"],
                priority=action_data["priority"],
                confidence=action_data["confidence"],
                parameters=action_data["parameters"],
                estimated_time=300
            )
            actions.append(action)
        
        return actions
    
    def _parse_knowledge_actions(self, enhancement: Dict[str, Any], knowledge_context: KnowledgeContext) -> List[AgentAction]:
        """解析知识库动作"""
        actions = []
        
        for action_data in enhancement.get("actions", []):
            action = AgentAction(
                action_type=ActionType.KNOWLEDGE_UPDATE,
                description=action_data["action"],
                priority=action_data["priority"],
                confidence=action_data["confidence"],
                parameters=action_data["parameters"],
                estimated_time=60
            )
            actions.append(action)
        
        return actions
    
    def _create_default_alert_action(self, alert_context: AlertContext) -> AgentAction:
        """创建默认预警动作"""
        return AgentAction(
            action_type=ActionType.USER_NOTIFY,
            description=f"通知管理员处理{alert_context.alert_type}预警",
            priority=3,
            confidence=0.5,
            parameters={"alert_id": alert_context.alert_id},
            estimated_time=10
        )
    
    def _create_default_knowledge_action(self, knowledge_context: KnowledgeContext) -> AgentAction:
        """创建默认知识库动作"""
        return AgentAction(
            action_type=ActionType.KNOWLEDGE_UPDATE,
            description="标记低置信度知识条目，等待人工审核",
            priority=2,
            confidence=0.3,
            parameters={"question": knowledge_context.question},
            estimated_time=5
        )
    
    async def _execute_alert_response(self, action: AgentAction) -> Dict[str, Any]:
        """执行预警响应动作"""
        # 这里实现具体的预警响应逻辑
        logger.info(f"执行预警响应: {action.description}")
        return {"success": True, "message": "预警响应已执行"}
    
    async def _execute_knowledge_update(self, action: AgentAction) -> Dict[str, Any]:
        """执行知识库更新动作"""
        # 这里实现具体的知识库更新逻辑
        logger.info(f"执行知识库更新: {action.description}")
        return {"success": True, "message": "知识库已更新"}
    
    async def _execute_workorder_create(self, action: AgentAction) -> Dict[str, Any]:
        """执行工单创建动作"""
        # 这里实现具体的工单创建逻辑
        logger.info(f"执行工单创建: {action.description}")
        return {"success": True, "message": "工单已创建"}
    
    async def _execute_system_optimize(self, action: AgentAction) -> Dict[str, Any]:
        """执行系统优化动作"""
        # 这里实现具体的系统优化逻辑
        logger.info(f"执行系统优化: {action.description}")
        return {"success": True, "message": "系统优化已执行"}
    
    async def _execute_user_notify(self, action: AgentAction) -> Dict[str, Any]:
        """执行用户通知动作"""
        # 这里实现具体的用户通知逻辑
        logger.info(f"执行用户通知: {action.description}")
        return {"success": True, "message": "用户已通知"}