Overview:
Part C involves the analysis of water flow data from two separate pipes
with different time stamps. The primary objective was to synchronize
these datasets into a unified hourly timeline, verify their
stationarity, and generate a one-week-forward forecast. Both datasets
were successfully aggregated to an hourly frequency, providing a clear
view of water movement across the system.
Purpose:
Because the raw data for Pipe 1 and Pipe 2 contained multiple recordings
at irregular intervals, a systematic aggregation process was
required.
Time-Base Sequencing:
All timestamps were parsed into a standard date-time format.
Hourly Averaging:
For hours containing multiple recordings, the mean water flow was
calculated to represent that hour, as required by the project
specifications.
Synchronization:
A master timeline was created to align Pipe 1 and Pipe 2, ensuring that
every hour in the sequence was accounted for, even if one pipe had no
data.
Objective:
Before forecasting, it was necessary to determine if the data was
stationary (meaning its statistical properties like mean and variance
are constant over time).
KPSS Test Results:
Interpretation:
Since both \(p\text{-values}\) are
greater than the 0.05 significance threshold, we fail to reject the null
hypothesis. This statistically confirms that the water flow data for
both pipes is stationary.
Model Choice:
Given the stationary nature of the data, a Seasonal ARIMA model was
selected to project future flows.
Model Selection:
The ARIMA framework was chosen because it can effectively capture any
remaining autocorrelation (the relationship between current and past
hours) without needing complex trend-differencing.
Model Parameters:
The models automatically identified the optimal lag structure to account
for the unique flow “signatures” of Pipe 1 versus Pipe 2.
Forecast Horizon:
The model generated a forecast for the next 168 hours (one full
week).
Pipe 1 – Consistent Flow Patterns:
The forecast for Pipe 1 shows a stable continuation of historical flow
rates, with confidence intervals indicating the range of expected
variability.
Pipe 2 – Volume Projections:
Pipe 2 demonstrates a slightly different flow profile, which the ARIMA
model captured to provide a reliable hourly estimate for the upcoming
week.
Summary:
The analysis confirms that the water flow data is stationary and highly
suitable for short-term forecasting. The resulting model provides a
detailed 168-hour outlook, which has been exported to an Excel-readable
format for operational use.
Waterflow_OneWeek_Forecast.csv.