Energy consumption data (in kWh) was obtained from university meter logs, covering both total annual and monthly use for each year from 2018 to 2024. Sub-metered data from ten major buildings and facilities were also collected to analyse building-level consumption patterns. The ten sub-metered buildings were selected because they represent the most energy-intensive and functionally diverse spaces on campus. These include student hostels, teaching and laboratory blocks, administrative facilities, and essential service installations such as water pumps. Together, they capture the main categories of functional space at JKUAT academic, residential, service, and research ensuring that the inventory reflects both core and support operations. This provides a reliable basis for analysing institutional consumption patterns.

Student population data included annual enrolment totals and intake statistics disaggregated by semester (January, May, September), obtained from university admission and registry records.

Climatic data (temperature, humidity, rainfall, wind speed) were sourced from the Kenya Meteorological Department, specifically for the Juja area, where JKUAT is located. Monthly averages were used to align with energy consumption trends.

All data were organized into Microsoft Excel, with separate sheets for Annual and monthly energy consumption, Enrolment and intake trends, Monthly average climate variables.

Missing data were interpolated using yearly averages when applicable. Outliers were examined and retained only if consistent with known events (e.g., COVID-19 disruptions in 2020).

For missing monthly entries, mean substitution based on annual averages was applied rather than linear interpolation, to maintain consistency with academic calendar cycles. While this approach reduces bias in long-term trend analysis, it may slightly smooth out short-term fluctuations, potentially underestimating true peak-load variations. This limitation was considered when interpreting results, particularly for monthly peak demand assessments.

Table 1 represents the descriptive statistics including means, standard deviations, ranges, and coefficients of variation computed for energy consumption and all influencing variables. This provided a foundational understanding of the variability and patterns in the dataset.

Time series Figures were developed to visualize:

i. Annual trends in energy use vs. enrolment,

ii. Monthly consumption patterns aligned with academic calendars,

iii. The relationship between energy use and each climate variable.

Trend lines were fitted using Excel’s built-in regression tool. Both linear and non-linear trendlines were tested. For each, the equation and “r” value were determined to get the strength of fit.

To further quantify relationships between energy use and its drivers, simple linear regression models were created for each variable:

Here, y is the estimated monthly energy consumption, and each X represents one influencing variable. These regressions were not validated as predictive models but served to establish empirical links during inventory development.

The analysis was based on a sample size of n = 84 monthly observations (2018-2024). To ensure model validity, multicollinearity among climate variables was checked using Variance Inflation Factor (VIF) values, all of which were below common cut-offs (VIF < 5), indicating no severe multicollinearity.

To capture the combined effects of enrolment and climate drivers, a multivariable ordinary least squares (OLS) regression was estimated:

$y={\beta }_0+{\beta }_1{X}_1+{\beta }_2{X}_2+{\beta }_3{X}_3+{\beta }_4{X}_4+{\beta }_5{X}_5+\epsilon$ (6)

where, y is monthly energy consumption (kWh) and each X represents an influencing factor for example $X_1$ is student enrolment, $X_2$ is average temperature (˚C), $X_3$ is average wind speed (km/h), $X_4$ is average humidity (%) and $X_5$ is average rainfall (mm). Predictors were mean-centered, so that the intercept ${\beta }_0$ represents the baseline monthly energy use at average conditions. This preliminary model was used not for forecasting but to showcase the inventory’s ability to support model formulation in subsequent phases.

Pearson correlation coefficients were computed to determine the strength and direction of linear associations between energy consumption and independent variables. The results informed the choice of regression models and helped confirm observed trends.

The following equations were applied in the analysis:

1. Mean/average: $\bar{x}=\frac{1}{n}{\displaystyle {\sum }_{i=1}^n{x}_i}$ (7)

2. Standard deviation: $s=\sqrt{\frac{1}{n-1}}{\displaystyle {\sum }_{i=1}^n{\left(x_i-\bar{x}\right)}^2}$ (8)

3. Coefficient of variation: $CV=\frac{S}{\bar{X}}\times 100\%$ (9)

4. Pearson correlation coefficient: $r=\frac{{\displaystyle \sum \left(x_i-\bar{x}\right)\left(y_i-\bar{y}\right)}}{\sqrt{{\displaystyle \sum {\left(x_i-\bar{x}\right)}^2}{\displaystyle \sum {\left(y_i-\bar{y}\right)}^2}}}$ (10)

Table 2 shows the historical data of: annual of energy consumption, Annual total enrolment and annual total intakes at JKUAT for the past 7 years. It shows that in the past three years the highest annual intake was in 2021 (9421 students) and lowest in 2020 (7295 students). The rise in the 2021 intake was to account for the students who failed to enroll during the COVID-19 pandemic which had affected the 2020 intake. The total enrolment was the highest in 2023 (37,595 students) and lowest in 2020 (28,359 students). It also showed that total annual energy consumption in the institution ranged from 2,330,370 KWh in 2020 to 3,279692 KWh in 2018 which also had the second highest total enrollment.

Figure 1 shows a positive correlation between total annual energy consumption and student enrollment between 2018 and 2024. The highest energy usage (3.28 GWh) occurred in 2018, which also saw high enrollment. The lowest energy consumption (2.33 GWh) was recorded in 2020, aligning with a drop in student presence due to COVID-19 disruptions.

Figure 1 revealed a direct positive correlation between student enrollment and energy consumption, confirming that academic activity is a primary driver of energy demand hence Student presence on campus is the most dominant factor influencing energy usage. More students mean more demand for classrooms, labs, lighting, computing, and support services.

Table 2 shows the intakes for the past years. This breaks the numbers to three annual intakes that JKUAT always receives. Both Table 3 and Figure 4 illustrate that September remains the dominant intake month every year, contributing over 80% of total annual admissions in most cases. For example, in 2024, 8,558 out of 9,254 new students were admitted in September alone.

Since the majority of students are admitted in September, this month leads to higher utility loads. New admissions also increase the use of ICT labs, orientation activities, dormitories, and cafeterias hence strongly influencing the university’s energy profile.

Figure 2 a trend in the total intakes and the total enrolment, showing that intake affects the enrolment as the drop in intakes leads to a drop in enrolment.

Figure 3 shows the trend of total energy consumption for JKUAT from 2018-2024. It clearly shows a drop in 2020 due to COVID-19 pandemic before the total enrolment normalized. The average total consumption being between 2,500,000 KWh to 3,000,000 KWh.

Figure 5 reveals distinct monthly energy data indicating that July consistently recorded the highest energy use, while May showed the lowest. The peak monthly consumption occurred in July 2018, while the lowest was in May 2020. Figure 5 trends aligned more closely with academic term cycles than with environmental factors, reaffirming the role of institutional operations in driving electricity demand.

It should be noted that because missing values were treated using mean substitution, some short-term variability may have been smoothed. While this does not affect the long-term trends presented here, it could result in a slight underestimation of true peak loads.

Table 4 illustrates how energy consumption analysis of sub-metered data for 2024 revealed that the Main Meter alone accounted for 49% of total consumption.

Buildings associated with student use, including main, Hall 6, the Science Complex, and COHRED, collectively consumed approximately 95% of campus electricity. This is shown in Table 5 with a small percentage of about 5% not associated with students enrolment.

The ten sub-metered sites were intentionally selected because they capture the major functional categories of campus space: academic blocks (Science Complex, COHRED, IEET, COHES), residential facilities (Hall 6), administrative/service installations (Main, Sajorec), and essential utilities (Booster Pump, Ndarugu Pump). Together, these meters account for approximately 95% of total electricity consumption, providing a comprehensive picture of both student-driven and support operations.

Table 5 confirms that there is a high energy use strongly linked to student-focused buildings of about 95%, indicating the combined effect of lighting, laboratory use, ICT equipment, and increased occupancy, therefore future energy-saving interventions must focus on student-accessed spaces.

While JKUAT has minimal reliance on HVAC systems, climate still exhibited modest effects. A negative correlation was observed between rainfall and energy consumption, particularly during April and May when heavy rains coincided with decreased campus activity. Rainfall tends to reduce physical movement and activity on campus therefore students and staff are less likely to attend classes or use facilities extensively during heavy rain leading to cancellation or postponement of the activities like lab activities due to lower attendance hence reduced overall occupancy. Moreso JKUAT with agricultural programs may rely on irrigation systems, pumps, and greenhouses, which operate less during rainy months. Since these operations are powered electrically, rainfall reduces this demand significantly.

Figure 6 shows the relationship between total monthly energy consumption and average rainfall and the relationship seen is slightly negative. Temperature showed a slight positive correlation with energy use, especially during warm months like March and October in Figure 7 due to cooling requirement and more irrigation need for the farms and the green houses.

Humidity and windspeed had minimal direct influence but appeared to align with periods of lower academic activity as shown in Figure 8 and Figure 9 respectively. The relation is negative for rainfall and humidity because an increase in rainfall and humidity reduces the need to pump water in the farms as JKUAT is an agricultural university and also reduces the Laboratory activities. It also reduces the need for the use of fans for cooling.

Figure 11 and Figure 14 show total daily consumption of the two-building metering that is COHES and IEET respectively. The results show that more energy was consumed by the building during weekdays as compared to weekends. For example, the weekdays ranged from 42.3 kwh to 46.6 kwh and weekends between 24 kwh to 25 kwh for COHES metering and weekdays for IEET between 164.5 kwh to 171.8 kwh and weekends between 143.4 kwh to 156.1 kwh. This is because during the weekends the school activities like lecture, lab practical was not done or just a few. On Sunday even the offices were not operating.

Figure 10 and Figure 13 show the energy consumption at night is nearly constant daily for both building metering but varies during daytime depending on the day for example week days have a higher consumption as compared to weekends. Figure 13 further explains that IEET metering contains greenhouses which need to be run by night just as day time hence more energy is consumed during the night compared to day time.

Figure 12 and Figure 15 show an hourly trend in energy consumption during daytime and night and it clearly shows that the energy consumed every hour at night is nearly constant as compared to daytime hours where energy varies depending on time that is lunch hours energy use reduces, departure time the energy use reduces and before working hours the energy use is low.

To support predictive model development, the combined influence of enrolment and climate factors was quantified using the multivariable OLS regression model described in Section 2.6. With predictors centered on their means, the intercept ${\beta }_0$ represents the baseline monthly energy consumption under average enrolment and climate conditions.

The estimated coefficients provide marginal effects of each variable while holding others constant. For instance, the enrolment coefficient reflects the additional kWh associated with each extra student, independent of climate variation. This specification avoids the omitted-variable bias inherent in univariate regressions and ensures a statistically coherent predictive equation.

Note: The intercept represents the baseline energy consumption independent of the selected predictors.

This model provides a useful foundation for forecasting monthly energy needs based on anticipated student numbers and seasonal climatic variations, paving the way for the next phase of data-driven energy management and planning. While the model captures broad patterns, future work should validate predictive capacity using out-of-sample testing and additional performance indicators for example the Root Mean Square Error and the mean Absolute Error.

Table 7 illustrates Pearson correlation coefficient between energy consumption and five influencing variables. Energy use shows a strong positive correlation with student enrolment (r = 0.6337) confirming that total number of students is a major driver of electricity demand. A slight positive correlation was found with average temperature and a very week correlation with windspeed.

Conversely, rainfall showed a negative small correlation with energy use hence supporting the observed trend that heavy rain reduces campus activities and energy consumption. These results statistically confirmed the trends identified visually in Figures and seasonal patterns.

To further support model validity, multicollinearity among the climate variables was tested using Variance Inflation Factor (VIF) analysis. All VIF values were below 5, confirming that the variables could be used together in regression without posing collinearity concerns.