Genesis of Anaconda technology, an Artificial Intelligence CHPDH system: not exactly a Star Wars episode

Because of the upcoming Star Wars episode, we all have had a chance in these days to re-watch the famous saga and certainly in our case we have used this opportunity to see if in Star Wars there were systems similar to a modern Artificial Intelligence CHPDH system. The analysis has shown that in the famous saga there are quite a few of them, but most of them are lacking some key features such as the ability to predict and to act as humans would do.

Before introducing all the features of Anaconda technology, a modern Artificial Intelligence technology able to produce electricity, heat, cool and smartly and efficiently distribute them, it is relevant to have an understanding of how any CHPDH system works. CHPDH stands for Combined Heat and Power District Heating.

Based on the simultaneous heat demand, individually originated by the different end-users (i.e. block of flats, schools, facilities etc.), a CHP system, or Combined Heat and Power system, will be exposed to an aggregate heat demand expressed by the sum of the individual heat demand of the end-users. The aggregate heat demand represents a function relating the quantity of heat that is demanded by the community (i.e. the sum of the served end-users reached by the district heating network) and its time (i.e. at what hour is that heat demanded) as schematically represented in the chart below.

Chart 1 Example of daily aggregate heat demand

CHP system tends to produce electricity and heat or cool in a mode that results in a constant, continuous and at full power generation and therefore is ill-suited for satisfying such an irregular profile without wasting the majority of the produced heat/cool.

In this fictional case, the CHP system is designed to supply continuously 18 MWh (being it the peak of the demand) but except for the hour 7 there is no need for such a high heat production, resulting in a complete waste of thermal resources and inefficient heat production as pictured in Chart 2.

Chart 2 Example of inefficient heat production based on continuous production of a CHP system

Certainly, one might consider producing less thermal energy with the CHP system but this would leave some of the heat demand unsatisfied.

The best way to start addressing this issue is to add to the CHP system one or more boilers, for example running on natural gas, that could be more versatile and flexible in the production of heat than CHP. Nowadays, condensing boilers also have thermal efficiency in excess of 80%. The result of the insertion of natural boilers is represented in the following chart.

Chart 3 Typical CHPDH system made up of a CHP system and condensing boilers

In this case it is possible to cover the aggregate heat demand with the boilers until hour 6, then with the CHP system and the boilers between the hours 7 and 24 (some of the heat is now produced by the CHP system and not by the boilers). The CHP system would run for only 18 hours a day. In order to better understand Chart 3, one must imagine that the CHP production and boilers integration are added up so that at hour 7 one must add 6 MWh produced by the CHP system and 12 MWh produced by the boilers for a total of 18 MWh that is equal to aggregate heat demand.

Even if the typical CHPDH system here presented is more efficient than previously shown configurations, we can still observe that between hours 16 and 19, when the boilers are not producing, there is still an overproduction due to the CHP system that translates into inefficient production.

Theoretically, it could be possible to reduce the CHP heat production to avoid that overproduction but this means that at the beginning of the project one should perfectly know the size of the CHP system as usually these systems run at full power to avoid losing efficiency and more importantly to avoid a premature aging of the system. Practically, this is never done and this is the reason that produces a less efficient system.

Finally, in explaining how the daily aggregate heat demand is served, it has been assumed that the CHP system knows in advance the aggregate heat demand of the community in a particular day, but since this depends on what each individual end-user decides to do with its own heating system and on what is the weather actually like and what is in reality perceived, the actual aggregate heat demand is a rather changing function that becomes rather difficult to be satisfied unless one develops a technology that could forecast the aggregate the heat demand and quickly adapt the heat production to it.

From this point the story of Cogenpower and of its Anaconda technology starts.

Cogenpower has been able to develop the Anaconda technology: a low-carbon set of technologies that automatically determines in advance the aggregate heat demand and is able to serve it without wasted heat.

Aggregate heat demand forecast

In the previous section, the basic assumption that we have used has been the fact that somehow the aggregate heat demand is known in advance so that an operator could decide the production plan of the CHP system and of the condensing boilers.

Unfortunately, this is not possible or easily done as the aggregate heat demand depends on what each individual end-user decides to do with its own heating system and on what is the weather actually like and what is in reality perceived.

A possible solution could come using a mathematical model for each end-user. Such a model could include the following parameters:

1.      Size of the end-user, i.e. the maximum heat it can absorb in the unit of time

2.      Time of the day when it needs the heat and how much heat it needs

It is clear that the first parameter indicates how big the absorption of heat can become for example in a very cold winter day but what is critical in the model is the second set of parameters. This set can be decided in advance from the end-user but typically it is based on other factors, first among other the external weather.

As a matter of facts, modern technologies designed to provide the best comfort for end-users use not only the current weather conditions (for instance the outdoor temperature) but also the history of that parameter in the last 48 hours (this avoids to be fouled if after days of cold and rain, one has one hour of nice weather). This inertia in the reaction of these technologies is difficult to model.

Artificial Intelligence Anaconda technology uses a better solution.

In fact, it employs the ability to use sophisticated algorithms that, based on a continuous flow of data from the end-users to the central command and control unit, located in the CHPDH automated power plants, can predict, calibrate and re-predict continuously the evolution of the single end-user and then of the aggregate heat demand.

In order to achieve this, Anaconda should be able to obtain that information on a continuous basis and real-time. This is achieved equipping the DH network with a fibre optic infrastructure (FOI) that can read continuously the heat meters installed at each end-user’s facility and fetch back that information to a dynamic algorithm that interprets that reaction considering the actual weather conditions.

This FOI can rightly be equated to a nervous system and the algorithms that one can use rely on mathematical procedures such as genetic algorithmsneural networks and agent-based systems. All these sophisticated processes allow the command and control system to know in advance the aggregate heat demand and to decide the more efficient and cost-effective production plan. Before coming back to this point let us add one technology that is necessary to create the Artificial Intelligence Anaconda system or the automated CHPDH plant developed by Cogenpower.

Heat storage facility

Let us go back to the previous problem in which we have assumed that somehow we know that the aggregate heat demand will be in the next 24 hours similar to the one represented below.

We have already seen at the beginning that the typical way to serve this profile is to consider a CHP system, whose size can usually cover a portion of the aggregate heat demand, and then to add one or more condensing boilers to cover the rest of the demand. This configuration sometimes may result not perfectly efficient because of the inflexibility of the CHP system during some hours.

The results changes dramatically if we can add a heat storage facility able to store the heat when is not demanded and release it when demand peaks.

The following chart explains the daily dynamics of such configuration that now includes an integrated heat storage facility.

Chart 4 CHPDH with heat storage facility allows storing heat when is not requested and use it when demanded

In this configuration it is possible to see that during the night, hours 1-6, the CHP system runs and produces heat in excess of the aggregate heat demand but that heat is now recovered in the heat storage facility until hour 5 after which a portion of that heat starts to be used to satisfy the increasing aggregate heat demand that reaches its peak between hour 7 and 8, while, because of the specific case, it is required to use some of the heat produced by boilers. To be noted is also that thanks to the heat storage facility the inefficient heat production occurring between the hours 16 and 19 is now fully recovered in the heat storage facility.

The following table summarises the previous results in such a way that it should be now evident the importance to have the ability to integrate one or more heat storage facilities into a CHP system connected to a district heating network, supported by condensing boilers.

Table Comparison of different CHPDH architectures

The table clearly shows that the CHPDH system with a heat storage facility is essentially 100% efficient in the fact that it is able to recover all the heat that is produced, provided that the size of the CHP system and the heat storage facility are designed accordingly. Certainly, we are considering the system as an ideal system and therefore we are assuming that the district heating network is without losses while in reality there are some heat losses along the pipeline even if they are immaterial during the winter months and become worth noting also during the summer time as they are a material portion of the distributed heat.

Another important aspect that should be considered in the analysis is that the amount of CHP heat production is in the last casemuch higher as the unit produces continuously for 24 hours, making all the production more efficient and with lower carbon content.

In conclusion, Cogenpower core Artificial Intelligence technology, Anaconda, is represented by its ability to master CHPDH with heat storage facility. This is what Anaconda is made up of together with its technological ability to forecast the heat demand and something else as it will be apparent in the next section.

Considering in fact the economic value of the output produced and sold – electricity and heat/cool – and knowing that cost of fuel is usually linked to heat tariffs in such a way that the profits of the system are maintained over the period, one might wonder what happens to electricity prices as the size of these CHPDH systems are small compared to large utility power plants that determine the electricity prices in the market.

Electricity prices forecast

CHP system produces electricity together with the heat that is served to a community distributed by a DH network. The electricity is sold to a price that is typically decided on a power exchange by the total electricity demand and total electricity supply on an hourly basis.

The large power plants have a significant and crucial role in deciding how much to produce and at what level of prices sell it. The dynamics of electricity prices is further complicated by the impact of renewable producers that are allowed to supply all their green energy independently from any price constraint as they receive feed-in-tariffs.

In order to be sure to produce according to the most efficient configuration but also according to the lowest possible cost, seeing the electricity price received for the electricity produced by the CHP system and sold to the grid as a negative cost, one is forced to have a tool to predict in which hour it is best to produce.

It is important to notice that the electricity market is called day-ahead market because the supply and demand offers must be presented the day before.

Usually, during the winter months with aggregate heat demand profiles similar to the one previously used (remember that the CHP system works continuously for 24 hours) the need to forecast the electricity prices for the next day is not at all important.

During the rest of the year and more precisely between April and September, the aggregate heat demand is typically much lower than during the winter months and this offers an opportunity to use the CHP system less than 24 hours, but when?

It becomes therefore quite relevant to decide on which hours the CHP system produces because it is desirable that in those hours we have the highest possible prices. The chart below shows a typical electricity price pattern on a working day.

Chart 5 Electricity prices

It is quite evident looking at the chart that it is a good idea to produce electricity around the two peaks in price located at the hours 9 and 21. Also, depending on how many hours the CHP system is allowed to produce to avoid wasting heat (and so depending on the aggregate heat demand and on the status of the heat storage facility) one can allocate production around these peaks. For instance, if the production hours were just 3, one could decide for the hours 8, 9 and 10. If the production hours were 11, instead, one could decide for the period ranging from hours 7 to 12 and from hours 19 to 23.

The difficulty is that the electricity market is a day-ahead market that is one must know what to produce and how much to sell it at the day before.

Therefore, it is mandatory to have forecast ability if one wants to maximise the economic value of electricity.

After thousands of simulations, Cogenpower has built a specific Seasonal Auto Regressive Integrated Moving Average (SARIMA) model to be used for forecasting short-term movements in the electricity market, which was then integrated into its SCADA system in Anaconda. The mathematical model that resulted from this approach has been proven to estimate the electricity price with an accuracy of 8.8% MAPE (Moving Average Percentage Error) and 6.089 MAE (Mean Absolute Error). Furthermore, it has also been proven to reach accuracy of 7.9% MAPE and 5.4 MAE on a 1 day forecast.

We show an example of a typical 7-day simulation that shows how chose the prediction is close to actual numbers.

Cogenpower has developed an advanced methodology to predict electricity prices and this methodology has been coded in the Artificial Intelligence technology Anaconda, which can predict aggregate heat demand, electricity prices, store the heat in excess and recover it, achieving level of efficiency and performance quite unique.

In the next article we will delve more into the fascinating Artificial Intelligence Anaconda technology of Cogenpower.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: