Time Series Data Noise Handling Questions

There are manufacturing time series data as shown in the picture.

The average of the two variables is about 100.

However, the noise value is 6500 and 65000, which is too different from other values.

I think those are data collection errors or misrecorded noise.

I want to process those values and analyze the time series, is there any way?

I want to use expert domain knowledge to scope and remove it, but it is difficult to meet an expert.