怎样在PHP项目中实现用户自定义审批AI？

wen java案例 2026-07-17 207

本文目录导读：

整体架构设计
核心组件实现
工作流与审批集成
数据库设计
AI模型训练与持续学习
前端实现示例（Vue.js）
部署与优化建议
高级功能

在PHP项目中实现用户自定义审批AI是一个复杂的系统工程,需要结合规则引擎、工作流和机器学习，以下是一个分步骤的实践方案：

整体架构设计

用户界面 (自定义审批规则) → API层 → 规则引擎 → AI推理 → 审批决策
                                              ↓
                                    工作流引擎 → 执行审批动作

核心组件实现

1 自定义规则系统

// 规则定义类
class ApprovalRule {
    public $id;
    public $name;
    public $conditions = []; // 条件组
    public $actions = [];    // 动作
    public $priority = 0;   // 优先级
    public $status = 'active';
}
// 条件类
class RuleCondition {
    public $field;      // 字段名
    public $operator;   // 运算符 (eq, gt, lt, contains, in等)
    public $value;      // 值
    public $logic;      // and/or
}

2 规则引擎实现

class RuleEngine {
    private $rules = [];
    private $aiService;
    public function __construct(AIService $aiService) {
        $this->aiService = $aiService;
    }
    public function evaluate($data, $context = []) {
        $matchedRules = [];
        // 1. 获取所有匹配的硬规则
        foreach ($this->rules as $rule) {
            if ($this->matchRule($rule, $data)) {
                $matchedRules[] = $rule;
            }
        }
        // 2. 如果有AI增强，让AI评估规则优先级
        if (!empty($matchedRules)) {
            $matchedRules = $this->aiService->rankRules($matchedRules, $data, $context);
        }
        // 3. 返回最高优先级的动作
        return $this->getHighestPriorityAction($matchedRules);
    }
    private function matchRule($rule, $data) {
        foreach ($rule->conditions as $condition) {
            $result = $this->evaluateCondition($condition, $data);
            if ($condition->logic === 'and' && !$result) {
                return false;
            }
            if ($condition->logic === 'or' && $result) {
                // 继续评估
            }
        }
        return true;
    }
}

3 用户自定义规则界面

// 控制器
class ApprovalRuleController {
    public function createRule(Request $request) {
        $validator = Validator::make($request->all(), [
            'name' => 'required|string|max:255',
            'conditions' => 'required|array',
            'actions' => 'required|array',
            'ai_boost' => 'boolean'
        ]);
        if ($validator->fails()) {
            return response()->json(['errors' => $validator->errors()], 400);
        }
        $rule = new ApprovalRule();
        $rule->name = $request->name;
        $rule->conditions = $this->parseConditions($request->conditions);
        $rule->actions = $this->parseActions($request->actions);
        $rule->ai_boost = $request->ai_boost ?? false;
        $rule->save();
        return response()->json(['rule' => $rule], 201);
    }
}

4 AI推理服务

class AIService {
    private $model;
    private $trainingData = [];
    public function __construct() {
        // 加载预训练模型或初始化
        $this->loadModel();
    }
    // AI增强的规则排序
    public function rankRules($rules, $data, $context) {
        // 1. 提取特征
        $features = $this->extractFeatures($data, $context);
        // 2. 使用AI模型预测每个规则的适用性分数
        $scores = [];
        foreach ($rules as $rule) {
            $scores[$rule->id] = $this->predict($rule, $features);
        }
        // 3. 按分数排序
        usort($rules, function($a, $b) use ($scores) {
            return $scores[$b->id] <=> $scores[$a->id];
        });
        return $rules;
    }
    // 智能建议新规则
    public function suggestRules($historicalData) {
        // 使用聚类或频繁项集挖掘发现常见模式
        $patterns = $this->mineFrequentPatterns($historicalData);
        return array_map(function($pattern) {
            return $this->patternToRule($pattern);
        }, $patterns);
    }
    private function loadModel() {
        // 加载PHP-ML或其他ML库的模型
        // 或者调用外部AI API
    }
    private function predict($rule, $features) {
        // 使用训练好的模型进行预测
        // 返回0-1之间的分数
        return 0.85; // 示例
    }
}

工作流与审批集成

class ApprovalWorkflow {
    private $ruleEngine;
    private $workflowEngine;
    public function process($entity, $action) {
        // 1. 构建审批上下文
        $context = $this->buildContext($entity, $action);
        // 2. 评估规则
        $decision = $this->ruleEngine->evaluate($entity->toArray(), $context);
        // 3. 根据决策执行流程
        switch ($decision->action) {
            case 'auto_approve':
                $this->autoApprove($entity);
                break;
            case 'auto_reject':
                $this->autoReject($entity);
                break;
            case 'route_to':
                $this->routeToApprover($entity, $decision->approver);
                break;
            case 'escalate':
                $this->escalate($entity);
                break;
        }
    }
}

数据库设计

-- 规则表
CREATE TABLE approval_rules (
    id INT PRIMARY KEY AUTO_INCREMENT,
    name VARCHAR(255),
    type ENUM('hard_rule', 'ai_suggested', 'user_defined'),
    conditions JSON,
    actions JSON,
    priority INT DEFAULT 0,
    ai_boost BOOLEAN DEFAULT FALSE,
    created_by INT,
    status ENUM('active', 'inactive', 'draft'),
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);
-- 规则评估历史
CREATE TABLE rule_evaluation_history (
    id INT PRIMARY KEY AUTO_INCREMENT,
    rule_id INT,
    entity_type VARCHAR(100),
    entity_id INT,
    context JSON,
    decision JSON,
    ai_score DECIMAL(5,2),
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
    FOREIGN KEY (rule_id) REFERENCES approval_rules(id)
);
-- 训练数据表
CREATE TABLE training_data (
    id INT PRIMARY KEY AUTO_INCREMENT,
    features JSON,
    label ENUM('approved', 'rejected', 'escalated'),
    actual_result ENUM('approved', 'rejected', 'escalated'),
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

AI模型训练与持续学习

class AITrainer {
    private $ml;
    public function __construct() {
        // 使用php-ml库或调用外部API
        $this->ml = new Phpml\Classification\SVC();
    }
    public function train(array $samples, array $labels) {
        // 训练模型
        $this->ml->train($samples, $labels);
        // 保存模型
        $this->saveModel();
    }
    public function retrainWithFeedback() {
        // 获取用户反馈数据
        $feedback = $this->getUserFeedback();
        // 混合训练
        $this->train($feedback['samples'], $feedback['labels']);
    }
    private function getUserFeedback() {
        // 从数据库中获取用户对AI建议的反馈
        return TrainingData::where('feedback', '!=', null)
                          ->pluck('features', 'label');
    }
}

前端实现示例（Vue.js）

<template>
  <div class="rule-builder">
    <div class="rule-header">
      <input v-model="rule.name" placeholder="规则名称">
      <label>
        <input type="checkbox" v-model="rule.ai_boost">
        启用AI增强
      </label>
    </div>
    <div class="conditions-builder">
      <div v-for="(group, index) in rule.conditions" :key="index">
        <select v-model="group.logic">
          <option value="and">且</option>
          <option value="or">或</option>
        </select>
        <select v-model="group.field">
          <option value="amount">金额</option>
          <option value="department">部门</option>
          <option value="project_type">项目类型</option>
        </select>
        <select v-model="group.operator">
          <option value="eq">等于</option>
          <option value="gt">大于</option>
          <option value="lt">小于</option>
        </select>
        <input v-model="group.value" placeholder="值">
        <button @click="removeCondition(index)">×</button>
      </div>
      <button @click="addCondition">+ 添加条件</button>
    </div>
    <div class="actions">
      <select v-model="rule.action">
        <option value="auto_approve">自动批准</option>
        <option value="auto_reject">自动拒绝</option>
        <option value="route_to">转交审批</option>
      </select>
      <div v-if="rule.action === 'route_to'">
        <input v-model="rule.approver" placeholder="审批人ID或角色">
      </div>
    </div>
    <div class="ai-suggestions" v-if="showAiSuggestions">
      <h3>AI建议的规则</h3>
      <div v-for="suggestion in aiSuggestions" :key="suggestion.id">
        <p>{{ suggestion.description }}</p>
        <button @click="applySuggestion(suggestion)">应用</button>
      </div>
    </div>
    <button @click="saveRule">保存规则</button>
  </div>
</template>

部署与优化建议

1 性能优化

使用Redis缓存常用规则评估结果
规则索引和预编译
批量评估多条记录
异步AI推理

2 可扩展性

支持插件系统扩展规则类型
微服务架构分离规则引擎和AI服务
API版本控制

3 监控与日志

记录每次规则评估和AI决策
A/B测试不同AI模型
用户反馈收集机制

高级功能

// 时间序列预测
class TimeAwareAI extends AIService {
    public function predictForTimePeriod($rule, $features, $timeContext) {
        // 考虑时间因素（月底审批量大、节假日等）
        $timeFeatures = $this->extractTimeFeatures($timeContext);
        return $this->predict($rule, array_merge($features, $timeFeatures));
    }
}
// 多层级审批
class MultiLevelApproval {
    public function determineApprovalChain($amount, $department) {
        // 动态决定审批层级
        $chain = [];
        if ($amount > 100000) {
            $chain[] = 'CTO';
            $chain[] = 'CEO';
        } elseif ($amount > 50000) {
            $chain[] = 'DepartmentHead';
            $chain[] = 'VP';
        }
        return $chain;
    }
}

这个方案提供了从基础规则引擎到AI增强的完整实现,你可以根据项目复杂度选择实现全部或部分功能，对于中小型项目，可以先实现规则引擎部分，然后逐步集成AI能力。

本文地址： https://www.bmjup.com/post/23419.html