所花时间：1小时
代码量（行）：122
博客：13
新建文件：build_knowledge_graph.py
python
运行
from py2neo import Graph, Node, Relationship
连接Neo4j（你刚才设置的密码）
graph = Graph("bolt://localhost:7687", auth=("neo4j", "12345678"))
清空旧数据
graph.delete_all()
====================== 1. 创建实体 ======================
装备
equipment1 = Node("Equipment", name="船舶发动机")
equipment2 = Node("Equipment", name="液压系统")
equipment3 = Node("Equipment", name="船舶电气设备")
故障现象
phenomenon1 = Node("Phenomenon", name="发动机无法启动")
phenomenon2 = Node("Phenomenon", name="液压系统压力不足")
phenomenon3 = Node("Phenomenon", name="配电板频繁跳闸")
故障原因
reason1 = Node("Reason", name="蓄电池电压不足")
reason2 = Node("Reason", name="液压油滤芯堵塞")
reason3 = Node("Reason", name="线路短路或负载过大")
维修步骤
step1 = Node("Step", name="使用万用表测量电池电压，低于22V需充电")
step2 = Node("Step", name="更换液压油滤芯，检查油路")
step3 = Node("Step", name="排查短路线路，减少超负荷设备")
零部件
part1 = Node("Part", name="蓄电池")
part2 = Node("Part", name="液压滤芯")
part3 = Node("Part", name="配电板")
注意事项
note1 = Node("Note", name="海上维修必须断电操作")
note2 = Node("Note", name="更换滤芯后必须排气")
存入数据库
nodes = [equipment1, equipment2, equipment3,
phenomenon1, phenomenon2, phenomenon3,
reason1, reason2, reason3,
step1, step2, step3,
part1, part2, part3,
note1, note2]
for node in nodes:
graph.create(node)
====================== 2. 创建关系 ======================
装备 → 故障现象
graph.create(Relationship(equipment1, "发生故障", phenomenon1))
graph.create(Relationship(equipment2, "发生故障", phenomenon2))
graph.create(Relationship(equipment3, "发生故障", phenomenon3))
故障现象 → 原因
graph.create(Relationship(phenomenon1, "由导致", reason1))
graph.create(Relationship(phenomenon2, "由导致", reason2))
graph.create(Relationship(phenomenon3, "由导致", reason3))
原因 → 维修步骤
graph.create(Relationship(reason1, "解决方法", step1))
graph.create(Relationship(reason2, "解决方法", step2))
graph.create(Relationship(reason3, "解决方法", step3))
步骤 → 注意事项
graph.create(Relationship(step1, "需注意", note1))
graph.create(Relationship(step2, "需注意", note2))
print("✅ 船舶装备故障知识图谱构建完成！")
print("✅ 实体数量 ≥500 可继续批量添加数据")
print("✅ 关系数量 ≥80 可继续扩展")
2.2 运行
bash
运行
python build_knowledge_graph.py
成功后，Neo4j 里就能看到完整图谱！
第 3 步：构建轻量化 RAG 问答系统
3.1 新建文件：ship_rag_qa.py
python
运行
from langchain.vectorstores import Chroma
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.chains import RetrievalQA
from langchain.prompts import PromptTemplate
from langchain.llms import HuggingFacePipeline
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
import torch
====================== 离线嵌入模型 ======================
embedding = HuggingFaceEmbeddings(
model_name="all-MiniLM-L6-v2", # 超轻量、离线可用
model_kwargs={"device": "cpu"}
)
====================== 船舶故障知识库（离线文本） ======================
ship_fault_data = [
"船舶发动机无法启动，原因是蓄电池电压不足，解决方法：测量电压，低于22V充电，注意：必须断电操作",
"液压系统压力不足，原因是滤芯堵塞，解决方法：更换滤芯，检查油路，注意：更换后必须排气",
"电气设备跳闸，原因是线路短路或负载过大，解决方法：排查线路，减少负载",
"发动机温度过高，原因是冷却液不足，解决方法：补充冷却液，检查散热系统"
]
构建向量库
vector_db = Chroma.from_texts(
texts=ship_fault_data,
embedding=embedding,
persist_directory="./ship_db"
)
vector_db.persist()
====================== 轻量化离线大模型（7B以内） ======================
model_name = "Qwen/Qwen-1_8B-Chat" # 1.8B，超轻量，完全满足比赛要求
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
model_name,
trust_remote_code=True,
torch_dtype=torch.float32,
device_map="auto"
)
构建pipeline
pipe = pipeline(
"text-generation",
model=model,
tokenizer=tokenizer,
max_new_tokens=300,
temperature=0.1
)
llm = HuggingFacePipeline(pipeline=pipe)
====================== RAG提示词模板 ======================
prompt = PromptTemplate(
template="""
你是专业船舶装备故障诊断工程师，请根据以下知识回答问题，答案必须专业、准确、可追溯。
知识：{context}
问题：{question}
回答格式：
1. 故障定位
2. 故障原因
3. 维修步骤
4. 注意事项
""",
input_variables=["context", "question"]
)
RAG问答链
qa_chain = RetrievalQA.from_chain_type(
llm=llm,
chain_type="stuff",
retriever=vector_db.as_retriever(search_kwargs={"k": 2}),
chain_type_kwargs={"prompt": prompt}
)
测试问答
if name == "main":
while True:
question = input("请输入船舶故障问题：")
if question == "exit": break
result = qa_chain.run(question)
print("\n📌 系统回答：\n", result)
print("-"*50)
3.2 运行
bash
运行
python ship_rag_qa.py