def revise(message):
# 订正解析结果，并插值到1km分辨率
mdpath, gcpath, savepath, *_ = Writefile.readxml(glovar.trafficpath, 1)
net = torch.load(mdpath)
dem = pd.read_csv(gcpath, index_col=0).values
arrays = np.array(
[np.nan_to_num([data, dem]) for data in message[:, :801, :1381]])
inputs = torch.from_numpy(arrays)
# torch.no_grad()
outputs = [net(it[np.newaxis, :]).detach().numpy() for it in inputs]
outputs = np.nan_to_num(outputs)
outputs[outputs < 0] = 0
print(outputs.shape)
output = np.squeeze(outputs)
lat = np.linspace(31.4, 39.4, 801)
lon = np.linspace(89.3, 103.1, 1381)
raingb = np.array([
np.nan_to_num(
interp.interpolateGridData(op, lat, lon, glovar.lat, glovar.lon))
for op in output
])
Writefile.write_to_nc(savepath, raingb, glovar.lat, glovar.lon, 'Rain',
glovar.fnames, glovar.filetime)
return outputs
def write(path, data, name, lat=None, lon=None, type=0):
filetime = ecmwf.ecreptime()
fh = range(3, 169, 3)
fnames = ['_%03d' % i for i in fh]
if type == 0:
Writefile.write_to_nc(path, data, lat, lon, name, fnames, filetime)
else:
Writefile.write_to_csv(path, data, name, fnames, filetime)
def write(self, data, lat=None, lon=None, type=0):
filetime = ecmwf.ecreptime()
fh = range(3, 169, 3)
fnames = ['_%03d' % i for i in fh]
name = 'Snow'
if type == 0:
Writefile.write_to_nc(self.savepath, data, lat, lon, name, fnames,
filetime)
else:
Writefile.write_to_csv(self.indexpath, data, 'SnowIndex', fnames,
filetime)
def reverse(saltedata, dataset, snowdepth):
"""
# 加载模型生成积雪深度模型结果
:param saltedata: 卫星数据
:param dataset: ec气象要素数据
:return:
"""
tmp = [data.reshape(-1, 1) for data in dataset] # 转换基础要素
ele = np.concatenate(tmp, axis=1)
ele.resize(56, 901 * 1401, 4) # 转换形状，将上一时刻积雪输入
temp = np.nan_to_num(ele)
snowdepth = snowdepth.reshape(-1, 1) # 积雪深度数据，仅包含前一时刻
m1, m2, savepath, roadpath, indexpath, _ = Writefile.readxml(
glovar.trafficpath, 0)
# m2 = r'/home/cqkj/LZD/Product/Product/Source/snow.pickle'
if saltedata is not None:
with open(m1, 'rb') as f:
model1 = pickle.load(f)
#########################################
saltedata.resize(901 * 1401, 1)
typecode = 1
else:
with open(m2, 'rb') as f:
model2 = pickle.load(f)
typecode = 2
alldata = []
################################################
for i in range(56):
# temp = [data.reshape(-1, 1) for data in dataset[i]] # 仅包含基础要素
# newdataset = np.concatenate([temp, snowdepth, saltedata], axis=1)
if typecode == 1:
newdataset = np.concatenate([temp[i], snowdepth, saltedata],
axis=1)
prediction = np.array(model1.predict(newdataset)) # 每轮结果
if typecode == 2:
#print(presnow.shape)
# 此处预报结果不可用图像呈现出分块
newdataset = np.concatenate([temp[i], snowdepth], axis=1)
prediction = np.array(model2.predict(
np.nan_to_num(newdataset))) # 每轮结果
# predictions = np.nan_to_num(model2.predict(np.nan_to_num(newdataset)))
predictions = np.nan_to_num(prediction)
# predictions[predictions < 0] = 0
print(predictions.shape)
snowdepth = predictions[:, np.newaxis] # 结果作为下一次预测的输入
predictions.resize(len(glovar.lat), len(glovar.lon))
sdgrid = np.nan_to_num(predictions)
sdgrid[sdgrid < 0] = 0
alldata.append(sdgrid)
sp = r'/home/cqkj/QHTraffic/Data//'
Writefile.write_to_nc(sp, np.array(alldata), glovar.lat, glovar.lon,
'SnowDepth', glovar.fnames, glovar.filetime)
return np.array(alldata) # 返回 [56, 901, 1401]网格数据
def clcRoadic(snowdepth, skint, rain):
# 计算道路结冰厚度数据
now = datetime.datetime.now().strftime('%Y%m%d%H')
# #########
with open(r'道路结冰模型', 'rb') as f:
model = pickle.load(f)
Roadic = model.predict(
[snowdepth.reshape[-1, 1], skint.reshape[-1, 1], rain.reshape[-1, 1]])
Roadic.resize(1, glovar.lat, glovar.lon)
path = r'填入网格文件的保存路径'
Writefile.write_to_nc(path, Roadic, glovar.lat, glovar.lon, 'iceDepth', '',
now)
return Roadic
def revise(message):
# 订正解析结果，并插值到1km分辨率
mdpath, gcpath, savepath, *_ = Writefile.readxml(glovar.trafficpath, 1)
net = torch.load(mdpath)
dem = pd.read_csv(gcpath, index_col=0).values
arrays = np.nan_to_num([np.array([data, dem]) for data in message])
inputs = torch.from_numpy(arrays)
# torch.no_grad()
outputs = [net(it[np.newaxis, :]).detach().numpy() for it in inputs]
outputs = np.nan_to_num(outputs)
outputs[outputs < 0] = 0
print(outputs.shape)
output = np.squeeze(outputs)
Writefile.write_to_nc(savepath, output, glovar.lat, glovar.lon, 'rain',
glovar.fnames, glovar.filetime)
return outputs
def revise(path, message):
# 订正解析结果，并插值到1km分辨率
mdpath, _, gcpath, savepath, indexpath, *_ = Writefile.readxml(path, 2)
data = [message[::2, :, :][:56], message[1::2, :, :][:56]]
# data = [data[::2, :, :][:56], data[1::2, :, :][:56]] # 分别取出来U风V风
net = torch.load(mdpath)
net.eval()
dem = pd.read_csv(gcpath, index_col=0).values
arrays = np.nan_to_num(
[np.array([i, j, dem]) for i, j in zip(data[0], data[1])])
inputs = torch.from_numpy(arrays)
torch.no_grad()
outputs = [net(it[np.newaxis, :]).detach().numpy() for it in inputs]
output = np.squeeze(np.nan_to_num(outputs))
datau, datav = output[:, 0], output[:, 1]
Writefile.write_to_nc(savepath, datau, glovar.lat, glovar.lon, 'U',
glovar.fnames, glovar.filetime)
Writefile.write_to_nc(savepath, datav, glovar.lat, glovar.lon, 'V',
glovar.fnames, glovar.filetime)
return indexpath, datau, datav
def revise(path, message):
# 订正解析结果，并插值到1km分辨率
mdpath, _, gcpath, savepath, indexpath, *_ = Writefile.readxml(path, 2)
data = np.array(message)
data = [
np.nan_to_num(data[::2, :, :][:56]),
np.nan_to_num(data[1::2, :, :][:56])
]
net = torch.load(mdpath)
net.eval()
dem = pd.read_csv(gcpath, index_col=0).values
arrays = np.array([
np.array([i, j, dem])
for i, j in zip(data[0][:, :801, :1381], data[1][:, :801, :1381])
])
inputs = torch.from_numpy(arrays)
torch.no_grad()
outputs = [
np.nan_to_num(net(it[np.newaxis, :]).detach().numpy()) for it in inputs
]
datau, datav = np.squeeze(outputs)[:, 0, ...], np.squeeze(outputs)[:, 1,
...]
# 统一格式
lat = np.linspace(31.4, 39.4, 801)
lon = np.linspace(89.3, 103.1, 1381)
uwind = [
np.nan_to_num(
interp.interpolateGridData(u, lat, lon, glovar.lat, glovar.lon))
for u in datau
]
vwind = [
np.nan_to_num(
interp.interpolateGridData(v, lat, lon, glovar.lat, glovar.lon))
for v in datav
]
Writefile.write_to_nc(savepath, np.array(uwind), glovar.lat, glovar.lon,
'U', glovar.fnames, glovar.filetime)
Writefile.write_to_nc(savepath, np.array(vwind), glovar.lat, glovar.lon,
'V', glovar.fnames, glovar.filetime)
return indexpath, datau, datav
def reverse(dataset):
# 此函数用来生成地表最高、 最低温度产品，对现有模型进行整体改动
maxpath, minpath, *_ = Writefile.readxml(glovar.trafficpath, 3)
with open(maxpath, 'rb') as f:
maxmodel = pickle.load(f)
with open(minpath, 'rb') as f:
minmodel = pickle.load(f)
# dataset.resize(56, 801 * 1381, 4)
################################################
temp = [data.reshape(-1, 1) for data in dataset] # 数据量可能过大
allele = np.concatenate(temp, axis=1) # 训练用元素
# 先采用一个进行测试
maxvalue = maxmodel.predict(allele).reshape(56, 901, 1401)
minvalue = minmodel.predict(allele).reshape(56, 901, 1401)
# 路径
savepath = r'/home/cqkj/QHtraffic/Data/skint//'
Writefile.write_to_nc(savepath, maxvalue, glovar.lat, glovar.lon,
'maxskint', glovar.fnames, glovar.filetime)
Writefile.write_to_nc(savepath, minvalue, glovar.lat, glovar.lon,
'minskint', glovar.fnames, glovar.filetime)
return maxvalue, minvalue # 返回地表最高、最低温度
