PowerForecast User Guide

Welcome to PowerForecast

This guide explains how to use PowerForecast to create accurate time series forecasts for your business. Whether you're forecasting demand, revenue, workload, or any other metric, this guide will help you get the best results.

Before You Begin

What You Need

1. Your historical data — Time series data in CSV or JSON format

2. Business context — What you want to forecast and why

3. Planning horizon — How far ahead you need to predict

Understanding Time Series Forecasting

Time series forecasting predicts future values based on historical patterns. PowerForecast uses advanced statistical models to:

Identify trends (increasing, decreasing, stable)

Detect seasonality (recurring patterns)

Handle uncertainty (confidence intervals)

Adapt to your data characteristics

Example: If you have monthly sales data for the past 2 years, PowerForecast can predict sales for the next 12 months, considering seasonal patterns (like holiday spikes) and trends.

The 5-Step Wizard

PowerForecast guides you through 5 steps to create a forecast:

Step	Name	What You Do
1	Business Goal	Select what you want to forecast
2	Time Configuration	Set frequency, horizon, confidence
3	Data Source	Upload your historical data
4	Model Selection	Choose Auto or Manual model selection
5	Review & Run	Validate and execute the forecast

Step 1: Business Goal

What to Select

Choose the business target that best matches what you want to predict:

Target	Best For
Demand	Product demand, sales volume, order quantities
Workload	Staff requirements, task volume, capacity needs
Revenue	Sales revenue, income projections
Claims	Insurance claims, requests, incidents
Transactions	Transaction volume, payments
Custom	Any other metric you define

Why it matters: The business target helps PowerForecast understand context and select appropriate models.

Granularity (Optional)

Granularity segments your forecast:

**No segmentation** — Aggregate forecast across all data

**Per Product** — Separate forecast for each product

**Per Region** — Forecast by geographic area

**Per Customer Segment** — Forecast by customer type

**Per Branch/Location** — Forecast by location

**Per Category** — Forecast by category

Example: If you upload sales data for 10 products and select "Per Product," PowerForecast creates 10 separate forecasts.

Tips

Choose the target that matches your business question

Granularity is optional — you can aggregate results later

Custom targets work just like predefined ones

Step 2: Time Configuration

Forecast Frequency

How often your data points occur:

Frequency	When to Use
Hourly	High-frequency monitoring (energy, traffic)
Daily	Daily operations (daily sales, daily demand)
Weekly	Weekly planning (weekly reports, capacity)
Monthly	Monthly planning (monthly revenue, budgets)
Quarterly	Quarterly reviews (quarterly sales)
Yearly	Long-term planning (annual forecasts)

Important: Match frequency to your data's natural period. If you have daily data, use Daily frequency.

Forecast Horizon

How many periods ahead to forecast:

**1-6 periods** — Short-term planning

**6-12 periods** — Medium-term planning (recommended)

**12-24 periods** — Long-term planning

**24+ periods** — Strategic planning

Note: Horizon is measured in your selected frequency:

Monthly frequency, horizon 12 = 12 months ahead

Daily frequency, horizon 30 = 30 days ahead

Guidelines:

Don't forecast too far ahead (forecast accuracy decreases with distance)

Match horizon to your planning needs

More historical data allows longer horizons

Confidence Level

Probability that actual values fall within prediction intervals:

Level	Use For
80%	Less conservative, narrower intervals
90%	Standard business planning
95%	Default — balanced confidence
99%	High-stakes decisions, risk-averse

Understanding Confidence Intervals:

A 95% confidence interval means:

There's a 95% chance the actual value will be between the lower and upper bounds

Higher confidence = wider intervals (more uncertainty)

Lower confidence = narrower intervals (less uncertainty)

Example: Forecast: 100 units, 95% interval: [90, 110]

95% chance actual value will be between 90 and 110

Scenario (Optional)

Select a scenario type:

Scenario	Description	When to Use
Baseline	Most likely outcome	Standard forecasting
Optimistic	Best-case scenario	Growth planning
Pessimistic	Worst-case scenario	Risk planning
Stress	Extreme conditions	Stress testing

Note: Scenarios affect how models interpret trends. The same data may produce different forecasts under different scenarios.

Step 3: Data Source

Data Format

PowerForecast accepts two formats:

CSV Format

```csv

unique_id,ds,y

product_1,2024-01-01,100

product_1,2024-02-01,120

product_2,2024-01-01,50

product_2,2024-02-01,55

```

Required Columns:

`unique_id`: Series identifier (product name, location, etc.)

`ds`: Date in ISO format (YYYY-MM-DD)

`y`: Numeric value to forecast

JSON Format

```json

[

{

"unique_id": "product_1",

"ds": ["2024-01-01", "2024-02-01"],

"y": [100, 120]

}

]

```

Required Fields:

`unique_id`: String identifier

`ds`: Array of date strings (ISO format)

`y`: Array of numeric values

Data Requirements

Minimum Data Points:

At least 2× the seasonal period

Example: Monthly data with yearly seasonality needs at least 24 months

Data Quality:

Dates should be in chronological order

No duplicate dates for the same series

Values should be numeric (no text)

Best Practices:

More data = better forecasts

Consistent spacing (no large gaps)

Clean data (handle outliers before uploading)

Data Quality Options

Option	Description
Handle Missing Values	Automatically interpolate missing data points
Handle Outliers	Detect and handle outliers using IQR method

Missing Values:

PowerForecast interpolates gaps using surrounding data

Critical for maintaining time series continuity

Outliers:

Detected using Interquartile Range (IQR) method

Adjusted to prevent skewing forecasts

Useful for handling data entry errors

Step 4: Model Selection

Automatic Mode 🤖

What it does:

Analyzes your data characteristics

Performs statistical profiling

Automatically selects best model(s)

Considers multiple models and chooses the most suitable

When to use:

✅ You're not sure which model to use

✅ You want the best model for your data

✅ You're new to forecasting

✅ Your data patterns may change

Recommended for: Most users, especially beginners

Manual Mode 🎯

What it does:

You choose specific models

PowerForecast runs all selected models

You can compare results

When to use:

✅ You know which models work well

✅ You want to compare multiple models

✅ You have specific requirements

Available Models

Model	Best For	Characteristics
AutoARIMA	Automatic ARIMA selection	Handles trends and seasonality, robust
ETS	Exponential Smoothing	Handles various patterns, seasonal data
SeasonalNaive	Simple seasonal patterns	Baseline model, comparison

AutoARIMA:

Automatically finds optimal ARIMA parameters

Handles trend and seasonality

Good for most business time series

**Recommended for beginners**

ETS (Exponential Smoothing):

Flexible smoothing methods

Handles additive and multiplicative seasonality

Good for data with clear patterns

Fast and interpretable

SeasonalNaive:

Simple baseline model

Uses last season's value as forecast

Good for comparison

Fast but less sophisticated

Step 5: Review & Run

Specification Summary

Review all settings before running:

✅ Business target

✅ Frequency, horizon, confidence

✅ Data summary (series count, data points, date range)

✅ Model selection

✅ Data quality options

Running the Forecast

1. Review the specification summary

2. Click "Run Forecast"

3. Wait for execution (progress indicator shows status)

4. Results page appears when complete

Execution Time:

Small datasets (< 10 series): 5-15 seconds

Medium datasets (10-100 series): 15-60 seconds

Large datasets (100+ series): 1-5 minutes

What happens:

1. Data profiling (statistical analysis)

2. Model selection (if Auto mode)

3. Model training

4. Forecast generation

5. Quality diagnostics

6. Results compilation

Understanding Results

Forecast Summary

Field	Description
Run ID	Unique identifier for this forecast
Status	COMPLETED, RUNNING, or FAILED
Business Target	What was forecasted
Series Count	Number of time series
Models Used	Which models were applied
Execution Time	How long it took

Forecast Chart

Visualization showing:

**Historical Data** (blue) — Your actual past values

**Forecast** (green) — Predicted future values

**Confidence Intervals** (shaded) — Upper and lower bounds

Chart Controls:

**Aggregation**: View raw, daily, weekly, or monthly

**Zoom**: Click and drag to zoom

**Export**: Download as image

Profiling Summary

Statistical analysis:

Metric	Description
Mean	Average value
Trend	Upward, downward, or stable
Seasonality	Detected seasonal patterns
Stationarity	Whether data is stationary
Data Quality	Overall assessment

Diagnostics

Quality assessment:

Aspect	Description
Overall Quality	Good, Fair, or Poor
Recommendations	Suggestions for improvement
Model Performance	How well models fit

Working with Results

Exporting Results

Format	Use For
Specification	Re-running forecasts later
Results	Sharing with stakeholders
Diagnostics	Quality analysis
Profiling	Statistical details

Interpreting Forecasts

Forecast Values:

Point forecast (single number) — Most likely value

Confidence intervals — Range of likely values

Example:

Forecast: 1,000 units

95% interval: [900, 1,100]

Meaning: 95% chance actual will be between 900 and 1,100

Confidence Intervals:

Wider intervals = more uncertainty

Narrower intervals = less uncertainty

Use intervals for risk planning

Best Practices

Data Preparation

1. Clean your data — Remove errors, handle missing values

2. Ensure consistency — Regular spacing, no gaps

3. Include enough history — At least 2× seasonal period

4. Check data quality — Review before uploading

Model Selection

1. Start with Automatic — Let PowerForecast choose

2. Compare models — Use Manual mode to compare

3. Review diagnostics — Check model performance

4. Iterate — Try different models if needed

Forecast Interpretation

1. Consider confidence intervals — Don't just look at point forecasts

2. Review diagnostics — Check forecast quality

3. Understand limitations — Forecasts are predictions, not guarantees

4. Update regularly — Refresh forecasts as new data arrives

Limitations and Rules

What PowerForecast Can Do

✅ Forecast single or multiple time series

✅ Handle trends and seasonality

✅ Provide confidence intervals

✅ Automatically select best models

✅ Handle missing values and outliers

✅ Process large datasets efficiently

What PowerForecast Cannot Do

❌ Forecast without historical data — Need past data to predict future

❌ Handle non-time-series data — Data must have time component

❌ Guarantee accuracy — Forecasts are probabilistic, not certain

❌ Handle structural breaks — Sudden changes in patterns may not be captured

❌ Forecast infinite horizons — Accuracy decreases with distance

Data Requirements

Minimum:

At least 2× seasonal period of data

Consistent time spacing

Numeric values

Recommended:

3-5× seasonal period for better accuracy

Clean, consistent data

No large gaps or structural breaks

Troubleshooting

No Feasible Forecast

Symptoms: Forecast fails or returns errors

Causes:

1. Insufficient data (not enough historical points)

2. Data quality issues (missing values, outliers)

3. Invalid data format

4. API services offline

Solutions:

Provide at least 2× seasonal period of data

Check data quality and clean if needed

Verify data format matches requirements

Check services are running (ports 8020, 8021)

Poor Forecast Quality

Symptoms: Wide confidence intervals, poor diagnostics

Causes:

1. Not enough historical data

2. High volatility in data

3. Structural breaks or changes

4. Wrong model selection

Solutions:

Collect more historical data

Review diagnostics for recommendations

Try different models (use Manual mode)

Consider data transformations

Forecast Takes Too Long

Symptoms: Execution takes very long

Causes:

1. Large dataset (many series or long history)

2. Complex models

3. System resources

Solutions:

Normal for large datasets — wait for completion

Consider reducing dataset size for testing

Check system resources

Unexpected Forecasts

Symptoms: Forecasts don't match expectations

Causes:

1. Data quality issues

2. Model mismatch

3. Missing context

Solutions:

Review data quality and diagnostics

Try different models

Check if data captures all relevant factors

Getting Help

In-App Help

Click the **?** icon in any step for context help

Review validation messages for specific errors

Check diagnostics for recommendations

Documentation

[Concepts Guide](CONCEPT_GUIDE.md) — Understanding forecasting concepts

[Quick Start](QUICK_START.md) — Step-by-step tutorials

[Wizard User Guide](WIZARD_USER_GUIDE.md) — Screen-by-screen walkthrough

Support

Email: support@planningpowertools.com

Quick Reference

Keyboard Shortcuts

Shortcut	Action
`Ctrl+S`	Save specification
`Escape`	Close modal
`Enter`	Confirm / Next

Data Format Quick Reference

CSV:

```csv

unique_id,ds,y

series_1,2024-01-01,100

```

JSON:

```json

[{"unique_id": "series_1", "ds": ["2024-01-01"], "y": [100]}]

```

API Endpoints

Forecast Engine (8020):

`POST /api/powerforecast/v1/forecast` — Run forecast

`GET /api/powerforecast/v1/models` — List models

SQLite API (8021):

`GET /api/forecast/specifications` — List specs

`GET /api/forecast/runs/{runId}` — Get run details

`GET /api/forecast/results/{runId}` — Get results

Remember: Forecasting is both art and science. Start with Automatic mode, review diagnostics, and iterate based on results. PowerForecast handles the complexity so you can focus on business decisions.