https://www.sciencedirect.com/science/article/pii/S2352484722019540?pes=vor
Cloud cover, wind, humidity, holidays, months, and day
Short-term load forecast
- Hourly, multi hourly, daily, and weekly
- Factors: units entrance and exit, unit’s product limitations and grid constraints, the distribution of optimum load, the allocation of reliable save on grid, the analysis of probable incidents occurrence and also the studies of short circuit and grid stability
The mid-term load forecast
- Monthly or maximum annually forecast
- Managing the consumption peak in special seasons like winter or summer, the optimum management of the product units, scheduling for inspection and repairing, and stable service to the customers
- Considering the energy, the rate of fuel-saving, and contract with associates
- Decision about the time and how to apply the water and thermal power plants, providing the fuel, the exchange rate of electric energy with associates
The long-term forecast
- For a year and several years
- Macro managerial decisions such as determining investments volume in providing the long-term programs in the future. It is mostly used for managing the energy resources and manufacturing or updating of power plants in next years, how to develop the grid, closing or reviewing in contracts. The increase of energy plants number, and distribution equipments in the existed grid should be done at least ten years earlier to be able to meet the requirements concurrent with the growth of the load type.
- Population growth, the number of subscribers, the annual income, the energy average price, the other fuels price, investment, income obtained from export, and value-added.
- Motors, pumps, fans, elevator machines, and transporters are the most electric devices in these groups of consumptions
- Consumption directly related to the development rate of country
- These loads are usually nearly constant value as for the working in the whole day and night in several work shifts, and only holidays can affect their consumption rate.
- Usually the load coefficient (=average load/ peak load) is high
Mid-term forecasting
- Main objective of MTLF is the scheduling of repair and maintenance and economic exploitation of the energy system that is straightly associated to the reliability, helps improving of density management in transmission grids, and as a result improving the total productivity of system and optimizing the energy cost for consumer
- In the past, accurate MTLF has the advantage that businesses can use it to enhance both their distribution and transmission networks. Today, attention in energy transactions and it helps in time interval monthly or annually in energy buying-selling negotiations and developing the contracts of production, transmission, and distribution.
- The weather effect on the prediction load in MTLF has been comprehensively studied in the
Adedeji et al. (2019). The autonomous method has been broadly used in modeling of MTLF, where the load rate in previous times and meteorology data are the effective main variables. - An autonomous model based only on load and weather factors
- AI based approach is empirically shown to be better than statistical approach
- Weather variables usually do not have accurate forecast in time horizon over one week.
- Consumer price index, wage income average, and exchange rate
- Investigating the statistical criteria (such as correlation or analyzing of partial correlation)
- Fourier series, ANN, fuzzy-neural networks
- Industrial growth rate, the consumer price rate, maximum and minimum temperature, the rainfall rate, the humidity rate, and wind speed.
- recursive time series
- Days of the week are broken down into three categories: early weekdays, midweekdays, and holidays
NN Inputs
The inputs of neural network are selected by calculating the correlation coefficient between output (load peak of next day) and the probability inputs (load peak in the last few days).
Fuzzy system: instead of using the variables such as economic indices, temperature, and so on as input, it uses their difference and or their relative difference as input
The neural network input including: the daily peak load on the similar date in previous years, week type, past week type in similar last years.
Past 60 data of load are used as input and by using feature discerning method of the Symbiotic Organism System Optimization (SOSO), the input dimensions are reduced. In other words, the most effective past inputs are identified.
Time intervals 5 to 35 min are used to forecast the energy of solar and wind generation resources. The method used for forecasting in this reference is the use of simple multi-layer ANN in which the Levenberg–Marquardt method is employed for optimization.
Neural network inputs including Gross Domestic Product (GDP), Gross National Income (GNI), Population (POP), electrical energy consumption by industry (EECI).
In the ANN method, also the inputs, peak load in previous month, the average of temperature, and the population growth rate have been considered.
Input of ANN including: the day of week, the holiday or work day, the load in same hour in the last 24 h, load in the past week in the same day and the same hour, the average load in the last 24 h, the average energy price in the last 24 h, the energy price in the same hour in the last 24 h and last week, the average temperature in big cities.
Ant Colony Optimization (ACO)
GDP rate, population rate, the import rate and export rate
GDP, PLD (Peak Load Demand), TED (Total Energy Demand), and population rate
Long-term natural gas consumption forecasting based on analog method and fuzzy decision tree
- https://www.scopus.com/record/display.uri?eid=2-s2.0-85112703419&origin=resultslist&sort=cp-f&cite=2-s2.0-85103702470&src=s&nlo=&nlr=&nls=&imp=t&sid=7042aa02e4b1e15e773eb6b1b3ad7b48&sot=cite&sdt=cl&cluster=scofreetoread%2c%22all%22%2ct&sl=0&relpos=4&citeCnt=9&searchTerm=
- energy intensity (energy consumption per dollar of Gross Domestic Product—GDP) and gas share in energy mix in some countries, usually more developed, are the starting point for forecasts of other countries in the later period.
- typical forecasting error of monthly electricity consumption for one geographical area, measured by the mean absolute percentage error MAPE, is about 2%
- quality of electricity consumption forecasts may vary depending on the group of consumers and the MAPE error variance, for example, from 2% to 10% for short-term forecasts, while the MAPE error for long-term forecasts is from 4% to 32%
- MAPE error in daily gas consumption forecasts is around 10% [3] or slightly less for hourly forecasts on a daily basis—MAPE 6–8% [4]. Forecasts of daily gas consumption on one day in advance, made by the best of the tested algorithms, have the MAPE error of 2% to 5% [5]. Nevertheless, the quality of forecasts significantly differs depending on the time horizon and the algorithms used—the MAPE error can range from 17% to over 100% [6,7]
- classical statistical methods (autoregressive integrated moving average ARIMA and seasonal ARIMA–SARIMA family or exponential smoothing ETS family) to modern computational intelligence tools (long-short term memory LSTM, neutral networks, etc.)
- Unlike classical time series forecasting, the fuzzy approach can deal with vague and incomplete time series data under uncertain circumstances. This problem particularly affects long-term forecasting.
- need 5 years to make long term change to supply chain and infrastructure
- a single country may be included in several clusters in different years. Create a prediction of membership to the energy consumption groups based on the forecasts of socio-economic indicators
- chain increment method: start with a base level of consumption and then adjust it incrementally based on changes in socio-economic, climatic, or energy consumption indicators.
Relative value arbitrage in European commodity markets
- cointegration-based statistical arbitrage strategy on a wide range of European energy sectors
- the magnitude of this intermediation fee
seems to be linked to commodity specific frictions limiting arbitrage possibilities
- convergence traders in
Europe’s commodity markets tend to be non-diversified investors focusing on specific market niches
- daily data of 85 futures contracts on coal, natural gas, crude oil, gasoil, emission allowances, and electricity over the period from
2006 to 2013
- highly profitable and robust to a
conservative setting of transaction costs, yielding excess returns
of about 6–8% annually
- difference between market prices
for refined petroleum products and crude oil (crack spread), intermarket spreads between UK and Dutch natural gas, inter-market
spreads between French and German electricity, or French base-peak
spreads
- inter-commodity, location, or calendar spreads
- oil- and electricity-sector with coal and
emission allowances playing a less important role
- before the recent financial crisis virtually all our trades arise from playing the crack spread, the methodology largely chooses positions in the electricity and gas sector later on. This shift goes hand-in-hand with a major market coupling event between the German and French electricity markets in 2010 which not only increased the dependence between markets but apparently also the trading opportunities.
- Consistent with the
notion that arbitrage can be costly and enforcing it is potentially
more difficult in certain sub-sectors, profits are highest for the case
of electricity and lowest for the coal sector.
- There is overwhelming evidence of long-run relationships between energy commodity prices for both the U.S. (Selertis and Herbert, 1999; Villar and Joutz, 2006; Mohammadi, 2008) and the European energy commodity market (Bencivenga et al., 2010; Joets and Mignon, 2011)
- Johansen’s cointegration procedure
- Trio arbitrage. long-run relationships regularly comprise more than
two securities such as e.g. emission-allowances, gas, and electricity.
- sector specific physical peculiarities in terms
of storage costs seem to have a considerable impact
- Some commodities must be moved from areas with excess
supply to areas with high demand and low stock, thus be spatially
transformed. Others need to be stored in order to make them available in periods of higher demand, thereby in a sense undergoing a
temporal transformation. The costs and associated restrictions with
which these transformations can be undertaken usually determine
the linkages between different commodity prices associated in such a
process.
- Prices for peakload hours of electricity in Germany and France usually closely follow each other since there are various transportation linkages between both markets. In a similar fashion, prices for gas at the TTF- and NBP-hubs never deviate too far from each other due to the Balgzand Bacton Line (BBL), a large transportation pipeline that connects the UK with the European mainland being used to balance excess supply and demand.
- With the introduction of the emissions trading scheme in
Europe, CO2 prices have become an integral component of the costs
for generating power. As a result, the so-called “clean dark- or sparkspread”, that is the spread between power price and adequately
weighted variable costs in terms of fuel and emission prices, gives
a crude measure for the profit-margin of a power-plant owner
- long-term stationary (tri-variate) spreads between power prices, fuel
prices (coal, natural gas, oil), and CO2 prices
- a lot of relationships between different commodities also come along with not so obvious lag structures, which can be a result of market participants’ hedging behavior or special contract clauses in long-term delivery agreements. For example, although crude oil is hardly used to generate electricity in most parts of Europe these days, many market players still purchase gas from Russian gas companies that use oil-linked tariffs. Consequently, oil and power prices might possess linkages as well. Additionally, these tariffs are often based on lagged prices, such that it is rather difficult to tell which prices along the term-structure
- a lot of relationships between different commodities also come along with not so obvious lag structures, which can be a result of market participants’ hedging behavior or special contract clauses in long-term delivery agreements. For example, although crude oil is hardly used to generate electricity in most parts of Europe these days, many market players still purchase gas from Russian gas companies that use oil-linked tariffs. Consequently, oil and power prices might possess linkages as well. Additionally, these tariffs are often based on lagged prices, such that it is rather difficult to tell which prices along the term-structure
Do and Faff (2010, 2012) analyze profits from pairs trading for US equities and observe a decline in profitability over the years. While Gatev et al. (2006) report risk-adjusted annual returns of about 4% for 1988–2002, Do and Faff (2012) obtain mostly insignificant returns for the years 1988–2009 which they partly attribute to a general underestimation of transaction costs by other peers
Weather dependent consumer demand, the role of inventory levels, and storage costs are likely to impact relative value arbitrage strategies directly
The occurrence of both futures from natural gas hubs (TTF and NBP) in our portfolio can most likely be attributed to the BBL, the first natural gas pipeline connecting both markets in 2006.14 Similar arguments apply to the French and German electricity markets, where both national transmission networks - although limited in their transfer capacity - are interconnected, so as to be able to transmit energy from one country to the other
Arbitrage profits are decreasing in access to capital, and fundamental volatility, and increasing in arbitrage trading opportunities.
We calculate the hypothetical cost necessary to produce the equivalent of 1 kWh of electricity from crude oil, gas, and coal. We proxy the storage costs for crude oil by recent quotes from storage futures contracts traded on NYMEX, the storage cost in the case of natural gas from auctions held by ICE Endex, while for coal we take actual storage costs of Richard’s Bay Coal Terminal (RBCT). For electricity, storing is not possible
and thus costs can be regarded as infinite.
Long‑run relationship between the Italian day‑ahead and balancing electricity prices
- Seasonally adjusted DA and balancing services prices are not characterized by unit roots, thus excluding the possible presence of cointegration in the classic sense, that is associated with the long-run equilibrium between non-stationary stochastic processes characterized by unit roots
- All the price series (filtered from the periodic patterns) show evidence of long-range dependence (or long
memory), that is an high and significant correlation between observations distant
in time. Hence, we cannot exclude the possible presence of fractional cointegration
- Even in markets that share the same regulation and common institutional factors, local specifc factors that can be related to the structure of the grid are the key elements that afect price convergence.
No comments:
Post a Comment