# -*-coding:utf-8 -*- import lightgbm as lgb import numpy as np import pandas as pd import json import datetime as dt def load_history_data(data_path:str): data = pd.read_csv(data_path) return data def load_config(cfg_path:str): with open(cfg_path, 'r', encoding='utf-8') as fr: config = json.load(fr) return config def load_lgb_model(model_path:str): return lgb.Booster(model_file=model_path) def cal_CO2(coal_cost, ncv): return coal_cost * ncv * 26.37e-3 * 0.94 * 44 / 12 def cal_coal_cost_emission(coal_cost, boiler, emission_factors): factor = emission_factors.get(boiler) if factor is not None: return coal_cost * factor else: return emission_factors.get("default") * coal_cost def cal_PM(c_smoke, boiler, emission_factors): factor = emission_factors.get(boiler) if factor is not None: return c_smoke * factor / 1e6 else: return emission_factors.get("default") * c_smoke / 1e6 def predict(his_data, input_data, model:lgb.Booster, object_cols, emission_factors): feature_names = model.feature_name() date = dt.datetime.strptime(input_data.get('time'), '%Y-%m-%d %H:%M:%S') r_NOx = float(input_data.get('nox')) r_SO2 = float(input_data.get('so2')) r_smoke = float(input_data.get('smoke')) flow = float(input_data.get('flow')) c_NOx, c_SO2, c_smoke = flow * np.asarray([r_NOx, r_SO2, r_smoke]) caloric = float(input_data.get("caloric")) if caloric > 1000: caloric = caloric / 1000 inputs = { "生产设备类型": input_data.get('boiler'), "汽轮机类型": input_data.get('steam'), "冷却方式": input_data.get('cold'), "压力参数": input_data.get('pressure'), "day_of_week": date.weekday(), "month": date.month, "hour": date.hour, "0_r_NOx":np.log1p(float(r_NOx)), "0_r_SO2":np.log1p(float(r_SO2)), "0_r_smoke":np.log1p(float(r_smoke)), "0_c_NOx": np.log1p(float(c_NOx)), "0_c_SO2": np.log1p(float(c_SO2)), "0_c_smoke": np.log1p(float(c_smoke)), "0_flow": np.log1p(float(flow)), "0_O2": np.log1p(float(input_data.get("o2"))), "0_temp": np.log1p(float(input_data.get("temp"))), "额定蒸发量_t/h": np.log1p(float(input_data.get("evaporation"))), "低位发热量": np.log1p(caloric), "单机容量（MW）": np.log1p(float(input_data.get("capacity"))), "lon": np.log1p(float(input_data.get("lon"))), "lat": np.log1p(float(input_data.get("lat"))), } new_df = pd.DataFrame.from_dict(inputs, orient='index').T total_data = pd.concat([his_data, new_df]) new_inputs = pd.get_dummies(total_data, columns=object_cols) new_inputs = new_inputs[feature_names].iloc[-1].values coal_cost = np.expm1(model.predict([new_inputs])[0]) co = cal_coal_cost_emission(coal_cost, input_data.get('boiler'), emission_factors.get('co')) vocs = cal_coal_cost_emission(coal_cost, input_data.get('boiler'), emission_factors.get('vocs')) pm25 = cal_PM(c_smoke, input_data.get('boiler'), emission_factors.get('pm25')) pm10 = cal_PM(c_smoke, input_data.get('boiler'), emission_factors.get('pm10')) co2 = cal_CO2(coal_cost, caloric) return {'coal': coal_cost, 'co':co, 'vocs':vocs, 'pm25':pm25, 'pm10':pm10, 'co2':co2} def predict_df(his_data, input_data:pd.DataFrame, model:lgb.Booster, object_cols:list, emission_factors): feature_names = model.feature_name() input_data_cp = input_data.copy() ignore_cols = ['company', 'unit', 'time'] input_data['time'] = pd.to_datetime(input_data['time']) input_data['day_of_week'] = input_data['time'].apply(lambda x: x.weekday()) input_data['month'] = input_data['time'].apply(lambda x: x.month) input_data['hour'] = input_data['time'].apply(lambda x: x.hour) input_data.drop(columns='time', inplace=True) # for col in object_cols: # input_data[col] = input_data[col].apply(lambda x: str(x).strip()) num_cols = [x for x in input_data.columns if x not in ignore_cols and x not in object_cols] for col in num_cols: input_data[col] = input_data[col].astype(float) input_data['0_c_NOx'] = input_data.apply(lambda x:x['0_r_NOx'] * x['0_flow'], axis=1) input_data['0_c_SO2'] = input_data.apply(lambda x:x['0_r_SO2'] * x['0_flow'], axis=1) input_data['0_c_smoke'] = input_data.apply(lambda x:x['0_r_smoke'] * x['0_flow'], axis=1) input_data['低位发热量'] = input_data['低位发热量'].apply(lambda x: (x/1000 if x > 9999 else x)) num_cols = [x for x in input_data.columns if x not in ignore_cols and x not in object_cols] new_data = input_data.copy() for col in num_cols: new_data[col] = np.log1p(new_data[col]) total_data = pd.concat([his_data, new_data]) new_inputs = pd.get_dummies(total_data, columns=object_cols) input_data_cp['coal'] = np.expm1(model.predict(new_inputs.iloc[-input_data.shape[0]:][feature_names])) input_data_cp['co'] = input_data_cp.apply(lambda x: cal_coal_cost_emission(x['coal'], x['生产设备类型'], emission_factors.get('co')), axis=1) input_data_cp['vocs'] = input_data_cp.apply(lambda x: cal_coal_cost_emission(x['coal'], x['生产设备类型'], emission_factors.get('vocs')), axis=1) input_data_cp['pm25'] = input_data.apply(lambda x: cal_PM(x['0_c_smoke'], x['生产设备类型'], emission_factors.get('pm25')), axis=1) input_data_cp['pm10'] = input_data.apply(lambda x: cal_PM(x['0_c_smoke'], x['生产设备类型'], emission_factors.get('pm10')), axis=1) assert 'coal' in input_data_cp.columns input_data_cp['co2'] = input_data_cp.apply(lambda x: cal_CO2(x['coal'], x['低位发热量']), axis=1) out_cols = ['company', 'unit', 'time', 'coal', 'co', 'vocs', 'pm25', 'pm10'] rst = input_data_cp[out_cols].to_dict(orient='list') return rst if __name__ == '__main__': history = load_history_data('../data/data_sample.csv') object_cols = load_config('../config/object_cols.json') model = load_lgb_model('../model_files/hour_best_model.txt') emission_factors = load_config('../config/emission_factor.json') inputs = history.iloc[:2] for col in inputs.columns[8:]: inputs[col] = np.expm1(inputs[col]) inputs['time'] = dt.datetime.now() inputs['company'] = '浙江秀舟热电' inputs['unit'] = '1' print(inputs.columns) print(predict_df(history, inputs, model, object_cols, emission_factors))