Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Self-learning non-linear adaptive heating curve adjustment for intuitive optimization
The aim is to extend an existing linear self-learning algorithm that optimizes the heating curve depending on building physics and external parameters in terms of indoor comfort and energy efficiency. For this purpose, we are working together with one of our industrial partners in the building technology sector in order to be able to test executable prototypes under real conditions in their facilities in addition to the theoretical simulations.
Keywords: Gaussian process modelling and Bayesian Optimization
Optimization
Dynamic heating curve adjustment
Self-learning algorithm
Industry related project
Reduction of energy consumption while maintaining room comfort
Generalization error - generic solution for use in all heating curve controlled heating systems
Machine Learning
Buildings are responsible for about 42% of the final energy consumption and for 26% of the total CO2 emissions in Switzerland, with heating demand being the main driver with about 68% - finding applicable solutions to reduce the energy consumption in this area will have a big effect. The main constraints here are maintaining the comfort of tenants and users. Minimizing the energy consumption of the heating system and meanwhile respecting these constraints makes it a challenging problem. A widespread approach to optimizing heating demand is through adjustments of heating curve. We want to develop a heating curve parameter optimizing controlling system, short adaptive heating tuner, based on Bayesian Optimization, which is scalable, fully automated, and fast convergent to reduce the energy demand while maintaining the room comfort of the inhabitants. The adaptive heating tuner is formulated as an optimization problem, where the decision variables are the parameters that determine the heating curve. The heating curve is uniquely defined by the parameters A, B (see Figure 1, left). The decision variable is therefore denoted by x : x = (A, B). In addition to the decision variables we need to introduce a suitable cost function based on the energy consumption and comfort violations: f : X -> R where X denotes the set of feasible decision variables and R is a real number. Since the cost function of heating curve parameters from energy consumption and comfort violations is not given, the closed form of f is not given. Moreover, the value of f can only be determined by experiment. To evaluate f for a given x, we must first set the heating curve parameters to x, run the heating system for one day, measure the total energy consumption for heating and cooling and the room and outdoor temperatures. Then we can calculate the amount of comfort violations and find the value f(x). Therefore, we use Bayesian optimization techniques to solve the presented optimization problem, see Figure 2 for further details.
In the first version of our algorithm, the parameter space of the heating curve is 4-dimensional, i.e. we can shift the coordinates of the points in the X or Y direction and restrict ourselves to a straight line between the points. This parameterization of the heating curve is referred to as a simple heating curve. In a further step, we are now interested in extending the heating curve to include non-linearities. The simplest extension is the so-called non-linear two-point curve, where a curved curve is introduced between the two points, after which we would add further points to the heating curve (see Figure 1, right). We are interested in the following aspects: to what extent do non-linearities affect our cost function, convergence and how much does this increase the computational effort? In the end, we are interested in finding an economically optimal curve in terms of computational effort and cost minimization. Validation is carried out using simulations based on design-builder models of the real test environments for NEST Unit UMAR and Sprint and the building provided by our industrial partner. Accomplishing the goals of this project requires numerical and experimental steps summarized in the lists given below (tentative and not finalized):
Buildings are responsible for about 42% of the final energy consumption and for 26% of the total CO2 emissions in Switzerland, with heating demand being the main driver with about 68% - finding applicable solutions to reduce the energy consumption in this area will have a big effect. The main constraints here are maintaining the comfort of tenants and users. Minimizing the energy consumption of the heating system and meanwhile respecting these constraints makes it a challenging problem. A widespread approach to optimizing heating demand is through adjustments of heating curve. We want to develop a heating curve parameter optimizing controlling system, short adaptive heating tuner, based on Bayesian Optimization, which is scalable, fully automated, and fast convergent to reduce the energy demand while maintaining the room comfort of the inhabitants. The adaptive heating tuner is formulated as an optimization problem, where the decision variables are the parameters that determine the heating curve. The heating curve is uniquely defined by the parameters A, B (see Figure 1, left). The decision variable is therefore denoted by x : x = (A, B). In addition to the decision variables we need to introduce a suitable cost function based on the energy consumption and comfort violations: f : X -> R where X denotes the set of feasible decision variables and R is a real number. Since the cost function of heating curve parameters from energy consumption and comfort violations is not given, the closed form of f is not given. Moreover, the value of f can only be determined by experiment. To evaluate f for a given x, we must first set the heating curve parameters to x, run the heating system for one day, measure the total energy consumption for heating and cooling and the room and outdoor temperatures. Then we can calculate the amount of comfort violations and find the value f(x). Therefore, we use Bayesian optimization techniques to solve the presented optimization problem, see Figure 2 for further details.
In the first version of our algorithm, the parameter space of the heating curve is 4-dimensional, i.e. we can shift the coordinates of the points in the X or Y direction and restrict ourselves to a straight line between the points. This parameterization of the heating curve is referred to as a simple heating curve. In a further step, we are now interested in extending the heating curve to include non-linearities. The simplest extension is the so-called non-linear two-point curve, where a curved curve is introduced between the two points, after which we would add further points to the heating curve (see Figure 1, right). We are interested in the following aspects: to what extent do non-linearities affect our cost function, convergence and how much does this increase the computational effort? In the end, we are interested in finding an economically optimal curve in terms of computational effort and cost minimization. Validation is carried out using simulations based on design-builder models of the real test environments for NEST Unit UMAR and Sprint and the building provided by our industrial partner. Accomplishing the goals of this project requires numerical and experimental steps summarized in the lists given below (tentative and not finalized):
The main goal of your project will be the implementation of non-linearities and their validation. The first prototypes can be created on the same data as already used for the first version of the algorithm development. In a further step, new data can be added, especially regarding validation.
The following sections outline the questions and deliverables to be addressed during the project. Completion of all tasks is not required for passing; rather, they serve to provide an overview of the project's scope and objectives:
**Research Questions:**
- How do non-linearities, such as the non-linear two-point curve, impact the optimization process of the heating curve parameters in terms of convergence and computational efficiency?
- What is the comparative performance of non-linear heating curve models against traditional linear models in terms of energy savings and comfort maintenance?
- How do different forms of non-linearities (e.g., sigmoidal curves, multi-point curves) influence the behavior of the heating system?
- What are the optimal strategies for incorporating non-linearities into the adaptive heating tuner algorithm while ensuring fast convergence and scalability?
- How do non-linear heating curve models perform under varying external conditions (e.g., different climate zones, building types) and what adjustments are necessary to accommodate these variations effectively?
**Deliverables:**
- A functioning algorithm for adaptive heating curve optimization, capable of incorporating non-linearities and demonstrating improved energy efficiency while maintaining user comfort.
- Technical reports detailing the development process, algorithm design, and validation results.
- Presentation materials for midterm presentation and final presentation
- Open-source software code repository containing the developed algorithm, allowing for further collaboration, refinement, and adoption by the research community.
- Pilot installation showcasing the practical application and benefits of the adaptive heating tuner in real building settings.
**Project timeline:**
_Phase 1: Project Initiation and Planning (Weeks 1-3)_
- Define project objectives, scope, and deliverables.
- Conduct a literature review on heating curve optimization and Bayesian optimization techniques.
- Familiarization with existing algorithms, focusing on linear heating curves.
_Phase 2: Research and Algorithm Development (Weeks 4-8)_
- Develop non-linear extensions to the initial version of the adaptive heating curve optimization algorithm, starting with the non-linear two-point curves.
- Conduct simulations to validate algorithm performance under various scenarios.
- Document the algorithm development process and preliminary findings.
_Phase 3: Midterm Presentation (Week 9)_
- Prepare a presentation summarizing project progress, including research findings, algorithm development, and initial validation results.
- Present the midterm progress.
- Receive feedback and suggestions for refining the algorithm and adjusting project priorities for the remainder of the timeline.
_Phase 4: Non-Linear Algorithm Development (Weeks 10-14)_
- Extend the adaptive heating curve optimization algorithm to incorporate further non-linearities, such as multi-point curves.
- Conduct additional simulations to assess the impact of non-linearities on algorithm performance, convergence, and computational effort.
- Fine-tune algorithm parameters and optimization strategies based on simulation results.
- Document the non-linear algorithm development process and compare performance metrics with linear counterparts.
_Phase 5: Validation and Optimization (Weeks 15-19)_
- Validate the developed algorithm using real-world building data and simulations.
- Optimize algorithm parameters to achieve the desired balance between energy efficiency and user comfort.
- Conduct sensitivity analyses to understand the robustness of the algorithm under different operating conditions.
- Prepare technical reports and documentation summarizing validation results and optimization strategies.
_Phase 6: Finalization, Dissemination and project wrap up (Weeks 20-24)_
- Finalize the adaptive heating curve optimization algorithm based on validation findings.
- Prepare master thesis for dissemination and graduation.
- Organize a final presentation to showcase project outcomes.
- Conduct a review to assess achievements, lessons learned, and areas for future improvement.
- Compile final project deliverables technical reports, and software documentation.
- Archive project data developed algorithm and documentation for future reference and replication.
- Celebrate project success 😊
**Requirement:**
- We are seeking a master's student with experience in machine learning or optimization, who is enthusiastic about contributing to a project with practical applicability.
- Prior knowledge in Gaussian Process (GP) and Bayesian Optimization (BO) is not expected but will be part of your literature studies.
- Basic knowledge in python is mandatory and nice to have are conceptual ideas of digital twin and building energy simulation programs e.g. EnergyPlus.
- Students need to find a supervisor at their home university.
The main goal of your project will be the implementation of non-linearities and their validation. The first prototypes can be created on the same data as already used for the first version of the algorithm development. In a further step, new data can be added, especially regarding validation.
The following sections outline the questions and deliverables to be addressed during the project. Completion of all tasks is not required for passing; rather, they serve to provide an overview of the project's scope and objectives:
**Research Questions:**
- How do non-linearities, such as the non-linear two-point curve, impact the optimization process of the heating curve parameters in terms of convergence and computational efficiency?
- What is the comparative performance of non-linear heating curve models against traditional linear models in terms of energy savings and comfort maintenance?
- How do different forms of non-linearities (e.g., sigmoidal curves, multi-point curves) influence the behavior of the heating system?
- What are the optimal strategies for incorporating non-linearities into the adaptive heating tuner algorithm while ensuring fast convergence and scalability?
- How do non-linear heating curve models perform under varying external conditions (e.g., different climate zones, building types) and what adjustments are necessary to accommodate these variations effectively?
**Deliverables:**
- A functioning algorithm for adaptive heating curve optimization, capable of incorporating non-linearities and demonstrating improved energy efficiency while maintaining user comfort.
- Technical reports detailing the development process, algorithm design, and validation results.
- Presentation materials for midterm presentation and final presentation
- Open-source software code repository containing the developed algorithm, allowing for further collaboration, refinement, and adoption by the research community.
- Pilot installation showcasing the practical application and benefits of the adaptive heating tuner in real building settings.
**Project timeline:**
_Phase 1: Project Initiation and Planning (Weeks 1-3)_
- Define project objectives, scope, and deliverables.
- Conduct a literature review on heating curve optimization and Bayesian optimization techniques.
- Familiarization with existing algorithms, focusing on linear heating curves.
_Phase 2: Research and Algorithm Development (Weeks 4-8)_
- Develop non-linear extensions to the initial version of the adaptive heating curve optimization algorithm, starting with the non-linear two-point curves.
- Conduct simulations to validate algorithm performance under various scenarios.
- Document the algorithm development process and preliminary findings.
_Phase 3: Midterm Presentation (Week 9)_
- Prepare a presentation summarizing project progress, including research findings, algorithm development, and initial validation results.
- Present the midterm progress.
- Receive feedback and suggestions for refining the algorithm and adjusting project priorities for the remainder of the timeline.
_Phase 4: Non-Linear Algorithm Development (Weeks 10-14)_
- Extend the adaptive heating curve optimization algorithm to incorporate further non-linearities, such as multi-point curves.
- Conduct additional simulations to assess the impact of non-linearities on algorithm performance, convergence, and computational effort.
- Fine-tune algorithm parameters and optimization strategies based on simulation results.
- Document the non-linear algorithm development process and compare performance metrics with linear counterparts.
_Phase 5: Validation and Optimization (Weeks 15-19)_
- Validate the developed algorithm using real-world building data and simulations.
- Optimize algorithm parameters to achieve the desired balance between energy efficiency and user comfort.
- Conduct sensitivity analyses to understand the robustness of the algorithm under different operating conditions.
- Prepare technical reports and documentation summarizing validation results and optimization strategies.
_Phase 6: Finalization, Dissemination and project wrap up (Weeks 20-24)_
- Finalize the adaptive heating curve optimization algorithm based on validation findings.
- Prepare master thesis for dissemination and graduation.
- Organize a final presentation to showcase project outcomes.
- Conduct a review to assess achievements, lessons learned, and areas for future improvement.
- Compile final project deliverables technical reports, and software documentation.
- Archive project data developed algorithm and documentation for future reference and replication.
- Celebrate project success 😊
**Requirement:**
- We are seeking a master's student with experience in machine learning or optimization, who is enthusiastic about contributing to a project with practical applicability.
- Prior knowledge in Gaussian Process (GP) and Bayesian Optimization (BO) is not expected but will be part of your literature studies.
- Basic knowledge in python is mandatory and nice to have are conceptual ideas of digital twin and building energy simulation programs e.g. EnergyPlus.
- Students need to find a supervisor at their home university.
If this thesis looks interesting reach out to us and let’s chat about your background and explore how your skills align with our project: michael.locher@empa.ch, alessandro.tell@empa.ch
If this thesis looks interesting reach out to us and let’s chat about your background and explore how your skills align with our project: michael.locher@empa.ch, alessandro.tell@empa.ch
