Optimization of available energy in microgrids
By predicting the battery charging status and extra energy and optimizing the power distribution, the power distribution problem in the microgrid is solved, the availability and service quality of the power grid are improved, and the dependence on high-cost diesel generators is reduced.
Patent Information
- Application Number
- CN202111150663.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2020-10-05
- Filing Date
- 2021-09-29
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2041-09-29
AI Technical Summary
The prior art fails to effectively optimize the distribution of electricity in microgrids, especially when taking into account the actual consumption of subscribers and battery characteristics, resulting in difficulty in optimizing grid availability and production costs.
A method is proposed to adjust the amount of electricity distribution by predicting the charging state and extra energy of the battery, considering weather forecast, battery capacity and subscriber consumption habits, optimize the distribution of electricity, ensure that the charging state of the battery is within the safety threshold, avoid power outages, and improve the Availability of the power grid.
The optimization of the power distribution of micro grids has been achieved, the availability and service quality of the power grid has been improved, the dependence on high-cost diesel generators has been reduced, and the stable supply of electricity has been ensured.
Smart Images

Figure CN114389311B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of distribution of electrical energy originating from solar energy to a subscriber grid. Background Art
[0002] In some countries with high levels of sunshine and where it is not possible to use the traditional electricity grid over the entire territory, microgrids or miniature power grids are increasingly developing, which are power grids supplied by autonomous production stations that convert solar energy into electricity. The microgrids are therefore isolated from the traditional electricity grid.
[0003] Microgrids typically include dozens to thousands of subscribers belonging to a region, village or town.
[0004] A production station typically includes a plurality of photovoltaic solar panels, batteries for storing the electricity generated by the sun, and various electrical devices designed to convert and distribute the electricity.
[0005] Of course, it is essential to optimize the distribution of electrical energy in microgrids to increase the availability of electrical energy and minimize the duration of power outages for grid subscribers.
[0006] To do this, it seems necessary to take into account not only the actual consumption of the subscriber, but also the characteristics of the batteries used.
[0007] Most solutions from the prior art linked to grid load shedding and optimizing energy production and distribution can be divided into two main categories.
[0008] The first category includes solutions (which may be algorithms or architectures) that make it possible to monitor the electrical network, in particular the voltage and frequency on the network, making it possible also to predict the evolution of these variables and to make load shedding decisions in order to protect the integrity of the production sources and guarantee the availability of energy for other loads using decision trees.
[0009] The second category includes solutions (in this case also algorithms or architectures) applicable to power grids with multiple energy production and storage systems, which aim to keep energy available while minimizing production costs.
[0010] In the first category, for example, solutions are known that consist in using a model for predicting the voltage on a line based on the state of the commands on this line in order to make load shedding decisions. This solution predicts the voltage law in a multi-source and multi-consumer grid based on the state of the inputs and outputs. Therefore, this solution does not solve the problem of optimizing the energy distribution in a microgrid.
[0011] Methods are also known for applying load shedding based on the voltage measured at the output of an electrical energy production station. This solution is based on measuring the output voltage to apply load shedding, especially in the case of an unbalanced grid. Therefore, said solution also does not solve the problem of optimizing the energy distribution in a microgrid.
[0012] In the second category, methods are known for optimizing the energy supply on the grid when there are multiple sources. The optimization consists in increasing the availability of energy while using the cheapest energy at time t. In this case, the variable sought to be optimized is the energy production cost. Therefore, this solution does not solve the problem involved here.
[0013] In summary, the solutions from the prior art related to the problems of load shedding and optimizing energy distribution do not address the context of microgrids and generally do not take into account the actual consumption of the subscribers or the characteristics of the batteries used. Optimization is generally associated with production costs, while load shedding is associated with erroneous behavior of the loads. Summary of the invention
[0014] The invention aims at optimizing the distribution of electrical energy in a microgrid supplied by solar energy.
[0015] To achieve this purpose, a method for allocating electrical energy produced by a production plant comprising at least one solar panel and at least one battery to a power grid of at least one subscriber is proposed, said method comprising, for a given day comprising day and night, an optimization step implemented before said given day:
[0016] - define the amount of energy for said given day;
[0017] - predicting a final state of charge equal to the state of charge of the battery at the end of the night on the given date and an assumed additional energy, the final state of charge being the energy not produced by the solar panels during the day on the given date but which could be produced if the subscriber's consumption and / or the energy demand of the battery increased;
[0018] - adjusting said amount of energy based on said final state of charge and said assumed additional energy;
[0019] After said optimization step, said allocation method further comprises a step of allocating said amount of energy to subscribers of said electrical network during said given date.
[0020] The allocation method thus makes it possible to optimize the amount of energy allocated to the subscribers of the grid. This optimization is particularly effective because it simultaneously takes into account weather forecasts, battery capacity, variations in the battery state of charge, and the consumption habits of the grid subscribers.
[0021] There is also proposed a distribution method as described above, wherein the amount of energy comprises an amount of daytime energy intended to be distributed during the daytime of the given date and an amount of nighttime energy intended to be distributed during the nighttime of the given date.
[0022] A distribution method as described above is also proposed, wherein adjusting the amount of energy comprises the following steps: checking whether the assumed additional energy belongs to a predetermined energy interval, and if this is not the case, increasing or decreasing the amount of daytime energy, and then predicting the final state of charge and the assumed additional energy again, and repeating these steps until the assumed additional energy belongs to the predetermined energy interval.
[0023] A distribution method as described above is also proposed, wherein adjusting the amount of energy comprises the following steps: checking whether the final state of charge belongs to a predetermined charging interval, and if this is not the case, increasing or decreasing the amount of night energy, and then predicting the final state of charge and the assumed additional energy again, and repeating these steps until the final state of charge belongs to the predetermined charging interval.
[0024] A method of allocation as described above is also proposed, wherein the prediction of said final state of charge and of said assumed additional energy implements a set of prediction algorithms that collaborate with one another, said set of prediction algorithms using as input an initial state of charge equal to the state of charge of the battery at the beginning of daylight on said given date, a weather forecast and the consumption of said subscriber during the day before said given date.
[0025] It is also proposed an allocation method as described above, wherein the final state of charge and the assumed additional energy are predicted before the given date for the given date and for one or more days after the given date.
[0026] A method of allocation as described above is also proposed, wherein the set of forecasting algorithms comprises an algorithm for forecasting the energy demand of the battery, an algorithm for forecasting the solar energy potential, an algorithm for forecasting the consumption of the subscribers and an algorithm for forecasting the evolution of the state of charge of the battery.
[0027] It is also proposed an allocation method as described above, wherein said algorithm for predicting the energy demand of said battery uses said initial state of charge as input.
[0028] It is also proposed an allocation method as described above, wherein the algorithm for predicting solar energy potential uses a weather forecast as input.
[0029] It is also proposed an allocation method as described above, wherein the algorithm for predicting the solar energy potential implements a supervised learning algorithm for generating an estimate of the irradiance, using which estimate to calculate the solar energy potential.
[0030] It is also proposed an allocation method as described above, wherein said algorithm for predicting said subscriber's consumption uses as input said subscriber's consumption during a day preceding said given date.
[0031] A method of allocation as described above is also proposed, wherein the algorithm for predicting the evolution of the state of charge of the battery uses as input the energy demand of the battery, the solar energy potential and the consumption of the subscribers during the daytime on the given date in order to predict the evolution of the state of charge of the battery during charging of the battery.
[0032] A method of allocation as described above is also proposed, wherein the algorithm for predicting the evolution of the state of charge of the battery uses as input the consumption of the subscriber during the night of the given date, the state of charge at the end of the day of the given date and the assumed additional energy in order to predict the evolution of the state of charge of the battery during the discharge of the battery.
[0033] A processing module is also proposed, in which the allocation method described above is implemented.
[0034] A computer program comprising instructions causing a processing module as described above to execute the steps of the dispensing method as described above is also proposed.
[0035] There is also proposed a computer-readable recording medium on which the computer program as described above is recorded.
[0036] The present invention will be better understood from the following description of a particular non-limiting embodiment of the invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Reference will be made to the accompanying drawings, in which:
[0038] [ Figure 1 ] Figure 1 A rural electrification architecture is shown in which the present invention is implemented;
[0039] [ Figure 2 ] Figure 2 is a graph comprising a solar energy potential curve and a curve of power provided by solar panels, defining between these two curves a region representing the assumed additional energy;
[0040] [ Figure 3 ] Figure 3The steps of the main algorithm implemented in the allocation method according to the present invention are shown;
[0041] [ Figure 4 ] Figure 4 shows the way in which the prediction algorithms implemented in the allocation method according to the invention cooperate;
[0042] [ Figure 5 ] Figure 5 Predictions for the final state of charge and assumed additional energy over several days are shown;
[0043] [ Figure 6 ] Figure 6 The steps of an algorithm for predicting solar energy potential are shown;
[0044] [ Figure 7 ] Figure 7 The steps of an algorithm for predicting the energy demand of a battery are shown. DETAILED DESCRIPTION
[0045] The distribution method according to the invention is implemented here in an “end-to-end” rural electrification solution for supplying electricity to subscribers 1 of a microgrid 2 .
[0046] The rural electrification architecture consists of a monitoring / configuration part 3 and a production / distribution part 4.
[0047] The monitoring / configuration part 3 includes a client data center 5 (including a server), an operation center 6 (including a network client for connecting to the client data center 5 ), and a processing unit 7 .
[0048] The processing unit 7 comprises a weather forecast collecting module 8 and a processing module 10 in which the present invention is implemented.
[0049] The processing unit 7 may be any component or electronic system for processing data, and may include, for example, a processor, a microcontroller, or a programmable logic circuit such as an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit).
[0050] The processing unit 7 also includes a communication device for communicating with the client data center 5 .
[0051] The production / distribution section 4 comprises a production station 11 .
[0052] The production station 11 includes a photovoltaic solar panel 12 (herein referred to as a "solar panel") and an inverter connected to the solar panel 12, and also includes a battery 14 and an inverter connected to the battery 14. Of course, there may be a single solar panel 12 and a single battery 14.
[0053] The production station 11 also comprises a diesel generator 15, the operation of which requires high costs. The invention makes it possible to reduce the use of the diesel generator 15 by optimizing as much as possible the use of the energy coming from the solar panels 12 and the battery 14. It should be noted that the presence of a diesel generator is optional and some production stations do not have one.
[0054] The production / distribution part 4 also comprises a production controller 17, a distribution controller 18, a hub 19 and a meter 20 making it possible to measure the electric energy consumed by the subscribers 1 of the microgrid 2. The meter 20 communicates with the hub 19 via a communication according to the RS-485 standard. The hub 19 communicates with the distribution controller 18 by power line communication (G3-PLC protocol).
[0055] The architecture of the monitoring / configuration part 3 and the production / distribution part 4 can of course be different from that described here. For example, with respect to the production / distribution part 4, it can include a plurality of hubs distributed over the microgrid, each hub having a plurality of meters. The monitoring / configuration part 3 and the production / distribution part 4 communicate with each other by implementing the TCP / IP protocol on the telecommunication network 21 (WAN network).
[0056] The monitoring / configuration part 3 sends technical configuration data, such as the authorized amount of energy, to the production / distribution part 4 .
[0057] The production / distribution part 4 collects production and distribution data as well as subscriber data and sends the data to the monitoring / configuration part 3 .
[0058] The processing module 10 in which the invention is implemented is therefore designed for communicating with the distribution controller 18 via the client data center 5 and the telecommunication network 21 .
[0059] The allocation method according to the invention uses the collected data and defines the amount of energy to be allocated to the subscriber 1 each day.
[0060] Here, a "day" is considered to include a given complete day and a given complete night, and thus starts with sunrise, ie, for example, at 5 or 6 am, and ends 24 hours later.
[0061] The amount of energy is technical configuration data sent to the distribution controller 18 of each microgrid 2 , which manages the distribution of the appropriate amount of energy to the subscribers 1 of the microgrid 2 every day.
[0062] The allocation method according to the invention makes it possible to predict the maximum authorized amounts of energy in order to avoid blackouts (i.e., power outages affecting a large number of subscribers). The allocation method according to the invention is able to estimate these amounts of energy for the following day and for several days. The further out in time the forecast is, the lower the confidence factor of the estimate of these amounts of energy. This possibility makes it possible to send the energy configuration amounts for several days to the allocation controller 18 so as to always have a configuration defined based on the weather forecast, even if the connection between the monitoring / configuration part 3 and the production / distribution part 4 is temporarily lost. In practice, situations of poor availability of the telecommunication network of newly electrified microgrids 2 often occur.
[0063] During the day, the electrical energy is mainly provided by the solar panels 12. This electrical energy makes it possible not only to meet the needs of the subscriber 1, but also to charge the battery 14 so as to be able to use the stored energy during the night. The consumption of the subscriber 1 takes precedence over the charging of the battery 14, so in very cloudy weather, the needs of the subscriber 1 can be met but the battery 14 will not be charged. If the battery 14 is not charged during the day, there may be no energy to meet the needs of the subscriber 1 at night. In addition, when the state of charge (SOC) of the battery 14 decreases below a predetermined charging threshold, a blackout occurs: the production station 11 is disconnected until the state of charge of the battery 14 returns to an acceptable level. This period may last for several hours, during which no subscriber 1 can have electricity. It is therefore very important to optimize the authorized amount of energy upstream of the subscriber 1 to protect the battery 14, charge said battery and make it possible to guarantee a better quality of service to the subscriber 1.
[0064] The allocation controller 18 has as a configuration parameter the amount of energy authorized to the subscriber. Thus, once the processing module 10 has calculated the amount of energy for the upcoming day, it is sent to the client data center 5, which is responsible for sending it to the allocation controller 18. A new configuration is sent every day. When a subscriber's consumption reaches the authorized maximum amount within a defined period of time, the subscriber is disconnected. No more energy is allocated to the subscriber until the next configured period.
[0065] The first object of the invention is to provide a daily amount of energy while avoiding blackouts. This means that the state of charge of the battery 14 must not fall below a predetermined charge threshold S%. In this case,
[0066] S=40.
[0067] The predetermined charge threshold of 40% makes it possible not to damage the battery 14. A battery charge state less than or equal to S% causes the production station to be disconnected until the battery 14 is charged and reaches (S+ΔS)%.
[0068] (S+ΔS)% is a parameter configured in the battery 14 .
[0069] in this case,
[0070] ΔS=10,
[0071] That is, after a power outage, the production station 11 distributes energy again starting from a 50% state of charge.
[0072] A second object of the invention is to increase the availability of energy to subscribers and in particular to use the assumed additional energy, if any.
[0073] On sunny days, due to insufficient energy demand from subscriber 1 and battery 14, a certain amount of energy that solar panel 12 could have generated is not generated: this is the assumed additional energy.
[0074] Thus, for a given day, the assumed additional energy is the energy that was not produced by the solar panels 12 during the given day but could have been produced if the consumption of the subscriber 1 and / or the energy demand of the battery 14 were greater.
[0075] refer to Figure 2 The concept of such assumed additional energy is illustrated. For a given day, the assumed additional energy 23 corresponds to the difference between the solar potential 24 of said given day and the power 25 provided by the solar panels during said given day (and integrated over said given day). In other words, the assumed additional energy is calculated by subtracting the subscriber's consumption and the battery charging requirements during the day from the solar potential.
[0076] Therefore, a second object of the invention is to increase the amount of energy distributed during the day in the presence of assumed additional energy.
[0077] Thus, the allocation method defines the amount of maximum energy available to the subscriber such that the state of charge at sunrise after the given date does not drop below a predetermined charging threshold, thereby avoiding a blackout. An assumed excess energy is also estimated. There is enough assumed excess energy so that the subscriber's needs during the day can be met, the battery's charging needs can be met so that the battery is fully charged, and the subscriber's needs during the night can be met.
[0078] Thus, the amount of energy includes an amount of daytime energy intended to be dispensed during the daytime of a given date (when it is daytime) and an amount of nighttime energy intended to be dispensed during the nighttime of a given date.
[0079] It is sought to optimize the amount of daytime energy based on the value of the assumed additional energy. To this end, if the assumed additional energy is greater than a predetermined energy threshold E (e.g. equal to 5 kWh), the amount of daytime energy is increased for the subscriber, otherwise if the assumed additional energy is negative, the amount of daytime energy is decreased. When said condition is met, a new amount of daytime energy is obtained.
[0080] At night, energy is formed by the battery. Therefore, the aim is to make the final state of charge of the battery greater than S% at sunrise, in order to avoid any blackouts. To this end, it is necessary to reduce or increase the amount of night energy. When said conditions are met, a new amount of night energy is obtained.
[0081] A more detailed description of an embodiment of the distribution method according to the invention is now given.
[0082] For each given date, the allocation method comprises an optimization step implemented before the given date, said optimization step comprising: defining the amount of energy for the given date, then predicting the final state of charge and the assumed additional energy, and adjusting the amount of energy based on the state of charge and the assumed additional energy. The allocation method then consists in allocating the amount of energy obtained to the subscribers of the microgrid during the given date.
[0083] Implementing the allocation method includes performing the steps of Figure 3 The main algorithms can be seen in .
[0084] The main algorithm uses a set of prediction algorithms to predict the final state of charge and the assumed additional energy.
[0085] The prediction algorithm gets at input the potential consumption of the subscriber. This consumption value is changed by an amount corresponding to the estimated maximum amount available to the subscriber, in order to determine the evolution of the assumed additional energy value and the final state of charge value based on the amount of energy modification. Therefore, the first step is to initialize the main algorithm with the amount of energy configured by default in the system (e.g. 50 kWh). Next, this amount of energy is modified to satisfy the stop condition of the main algorithm.
[0086] Therefore, the main algorithm starts with an initialization phase, in which the amount of "default" energy is defined (step E0). The amount of energy includes the amount of daytime energy and the amount of nighttime energy.
[0087] The default energy amount is, for example, equal to 50 kWh, ie, for example, a daytime energy amount of 30 kWh and a nighttime energy amount of 20 kWh.
[0088] The prediction algorithm then predicts the final state of charge SOC_f and the assumed additional energy Ex for a given day (step E1 ).
[0089] Adjusting the amount of energy comprises first checking whether the assumed additional energy belongs to a predetermined energy interval, and if this is not the case, increasing or decreasing the amount of daytime energy, and then predicting again the final state of charge and the assumed additional energy, and repeating these steps until the assumed additional energy actually belongs to the predetermined energy interval.
[0090] Therefore, after implementing the prediction algorithm, the following conditions are checked for the assumed additional energy Ex:
[0091] Ex∈[0;E]?
[0092] E is a predetermined energy threshold.
[0093] If the conditions are not met, and therefore if:
[0094] If Ex>E or Ex<0 (step E2), the main algorithm moves to step E3.
[0095] In step E3, if:
[0096] Ex<0, then:
[0097] step∈[0;1[and
[0098] Qed=Qed×step,where
[0099] Qed is the amount of daytime energy, and where,
[0100] Step is the first adjustment factor.
[0101] In this case, for example, step=0.9.
[0102] In step E3, if:
[0103] Ex>E, then:
[0104] step∈[1;2[and:
[0105] Qed=Qed×step.
[0106] In this case, for example, step=1.1.
[0107] The main algorithm then returns to step E1.
[0108] Therefore, if the assumed additional energy does not belong to the interval [0; E] and if the assumed additional energy is strictly negative, the amount of daytime energy is multiplied by a factor between 0 (inclusive) and 1 (exclusive), here equal to 0.9, and is therefore reduced.
[0109] If the assumed additional energy does not belong to the interval [0; E] and if the assumed additional energy is strictly greater than the predetermined energy threshold E, the amount of daytime energy is multiplied by a factor between 1 (inclusive) and 2 (exclusive), here equal to 1.1, and is therefore increased.
[0110] In this case, for example:
[0111] E=5kWh.
[0112] Therefore, if the prediction algorithm predicts that the assumed additional energy is strictly greater than 5 kWh, there is surplus energy that can be used, and therefore the amount of daytime energy increases. The amount of daytime energy is equal to:
[0113] 1.1*30=33kWh.
[0114] After step E3 , the main algorithm returns to step E1 : for the new amount of daytime energy, the prediction algorithm again predicts a hypothetical additional energy value Ex and a final state of charge value SOC_f.
[0115] Check again whether the conditions for the assumed additional energy are met:
[0116] Ex∈[0;E]?
[0117] If it is not satisfied, and therefore if the assumed additional energy still does not belong to the interval [0; E], the allocation method returns to step E3.
[0118] If the condition is met (step E4 ), the amount of daytime energy to be distributed is correctly defined.
[0119] It is then necessary that the conditions relating to the final state of charge are met at the end of the night on a given date (ie at sunrise after the given date).
[0120] Adjusting the amount of energy comprises checking whether the final state of charge belongs to a predetermined charging interval, and if this is not the case, increasing or decreasing the amount of energy at night, then predicting the final state of charge again and assuming the additional energy, and repeating these steps until the final state of charge belongs to the predetermined charging interval.
[0121] Therefore, the following conditions are checked for the final state of charge:
[0122] SOC_f∈[S%; (S+ΔS)%]?
[0123] SOC_f is the final state of charge and S is the predetermined charging threshold.
[0124] In this case, S=40 and ΔS=10.
[0125] If the said condition is satisfied, and thus if the final state of charge belongs to the interval [S%; (S + ΔS)%] (step E7), the quantity of energy for a given date is defined and the main algorithm ends (step E8).
[0126] If the said condition is not satisfied, and thus if the final state of charge does not belong to the interval [S%; (S + ΔS)%], i.e., if:
[0127] SOC_f < S% or SOC_f > (S + ΔS)% (step E5), the main algorithm moves to step E6.
[0128] In step E6, if:
[0129] SOC_f < S%, then:
[0130] step ∈ [0; 1[ and:
[0131] Qen = Qen × step, where,
[0132] Qen is the quantity of energy at night, and
[0133] step is the second adjustment factor.
[0134] In this case, for example, step = 0.9.
[0135] In step E6, if:
[0136] SOC_f > (S + ΔS)%, then:
[0137] step ∈ [1; 2[ and:
[0138] SOC_f = SOC_f × step.
[0139] In this case, for example, step = 1.1.
[0140] Therefore, if the final state of charge does not belong to the interval [S%; (S + ΔS)%] and if the final state of charge is strictly lower than the predetermined charge threshold, the quantity of energy at night is multiplied by a factor between 0 (inclusive) and 1 (exclusive), here equal to 0.9, and is thus reduced. If the predicted result of the final state of charge is lower than 40%, the quantity of energy at night is reduced:
[0141] 0.9 * 20 = 18 kWh.
[0142] If the final state of charge does not belong to the interval [S%; (S+ΔS)%] and if the final state of charge is strictly greater than the predetermined charging threshold plus 10%, i.e. 50%, the amount of night energy is multiplied by a factor between 1 (inclusive) and 2 (exclusive), here equal to 1.1, and is therefore increased.
[0143] After step E6 , the main algorithm returns to step E1 : for the new amount of energy, the prediction algorithm again predicts the assumed additional energy value and the final state of charge value.
[0144] It is checked again whether the conditions for the assumed additional energy are met and then, if this is the case, whether the conditions for the state of charge are met.
[0145] When the conditions for the state of charge are met and therefore:
[0146] When SOC_f∈[S%;(S+ΔS)%](step E7),
[0147] The amount of energy for a given day is defined and the main algorithm ends (step E8 ).
[0148] According to this example, the estimated amount of available energy would therefore be 51 kWh, of which 33 kWh would be during the day and 18 kWh would be at night.
[0149] A description is now given of a set of prediction algorithms for predicting the final state of charge and the assumed additional energy.
[0150] These algorithms work together in the following ways: Figure 4 shown.
[0151] Four prediction algorithms are used, which work together to obtain the final state of charge and the assumed additional energy.
[0152] This set of prediction algorithms 30 uses as input an initial state of charge SOC_i, which is equal to the state of charge of the battery at the beginning of daylight on a given date, the weather forecast and the subscriber's consumption during the date J-1 before a given date J. The weather forecast may be for a given date J only or, as will be seen later, for all dates between a given date J and date J+X, making it possible to define the amount of energy to be allocated for several days.
[0153] This set of prediction algorithms 30 comprises an algorithm 31 for predicting the energy demand of the battery, an algorithm 32 for predicting the solar energy potential, an algorithm 33 for predicting the consumption of the subscribers and algorithms 34a, 34b for predicting the evolution of the battery charge state.
[0154] The algorithm 31 for predicting the energy demand of the battery uses the initial state of charge as input.
[0155] The algorithm 32 for predicting solar energy potential uses weather forecasts as input.
[0156] The algorithm 33 for predicting a subscriber's consumption uses as input the subscriber's consumption during a date J-1 prior to a given date J.
[0157] The algorithm 34a for predicting the evolution of the battery's state of charge uses as input the energy demand of the battery (predicted by the algorithm 31 for predicting the energy demand of the battery), the solar energy potential (predicted by the algorithm 32 for predicting the solar energy potential) and the subscriber's consumption during the day on a given date (predicted by the algorithm 33 for predicting the subscriber's consumption) in order to predict the evolution of the battery's state of charge during battery charging.
[0158] The algorithm 34b for predicting the evolution of the battery state of charge uses as input the subscriber's consumption during the night of a given date (predicted by the algorithm 33 for predicting the subscriber's consumption), the charge state at sunset on a given date J and the assumed additional energy Ex (both predicted by the algorithm 34a for predicting the evolution of the battery state of charge) in order to predict the evolution of the battery state of charge during battery discharge.
[0159] This set of prediction algorithms 30 therefore makes it possible to predict the final state of charge SOC_f and the assumed additional energy Ex.
[0160] Thus, knowing the state of charge value at sunrise, the weather forecast for the day and the energy consumption of the subscriber the previous day, a prediction algorithm can be used to estimate the assumed additional energy and the state of charge at sunrise the next day, 24 hours later. This makes it possible to estimate the state of charge value and the assumed additional energy value when implementing the main algorithm and compare said values with thresholds.
[0161] This method is particularly advantageous, in particular because the final state of charge and the assumed additional energy for a given date J and one or more days after the given date J can be predicted before the given date J.
[0162] Thus, in case the communication between the production / distribution part and the monitoring / configuration part is lost, for example due to loss of the Internet connection, it is still possible to define the amount of optimized energy for a date after a given date.
[0163] Thus, the distribution method can work autonomously, ie without any Internet connection, and thus without receiving data, for X days.
[0164] Therefore, at each new Internet connection, the system recovers the weather forecast for the following X days and estimates the amount of energy for these X days.
[0165] Therefore, reference Figure 5 , implementing this set of prediction algorithms makes it possible to evaluate the final state of charge and the assumed additional energy for a given date J+X, and therefore define the amount of energy to be allocated until date J+X based on the weather forecast for X days, the initial state of charge for a given date J and the subscriber's consumption during the dates J-1 preceding the given date J.
[0166] A prediction algorithm is implemented (step E20_J) which makes it possible to obtain the final state of charge and the assumed additional energy for date J based on the weather forecast for date J, the initial state of charge for date J and the subscriber's consumption during date J-1 preceding date J.
[0167] The prediction algorithm is implemented again (step E20_J+1), which makes it possible to obtain the final charge state and the assumed additional energy for date J+1 based on the weather forecast for date J+1, the final charge state for date J and the consumption of the subscriber during date J.
[0168] These steps are repeated so that finally the prediction algorithm is implemented again (step E20_J+X), which makes it possible to obtain the final charge state and the assumed additional energy for date J+X based on the weather forecast for date J+X, the final charge state for date J+X-1 and the subscriber's consumption during date J+X-1.
[0169] A description of each prediction algorithm is now given.
[0170] The algorithm for predicting a subscriber's consumption may simply consist of considering that the subscriber's consumption during a given date is equal to the subscriber's consumption during the day before the given date.
[0171] A supervised regression machine learning algorithm may also be used. The goal is to satisfy the subscriber as much as possible in terms of energy consumption while complying with the available energy. Therefore, by knowing the consumption that the subscriber is about to incur, this makes it possible to determine the minimum amount of energy that the system wishes to supply to the subscriber.
[0172] The algorithm used to predict solar energy potential works as follows.
[0173] Solar potential is the amount of power a solar panel is able to produce given the weather conditions at a given time. Solar potential depends on the characteristics of the solar panel, its location, cloud cover, season and time of day, which affect the height of the sun and therefore the amount of energy received by the solar panel.
[0174] In this case, each solar panel includes an irradiance sensor and a sensor for sensing the temperature of the photovoltaic cell. Using these two sensors and the following formula, the solar energy potential can be calculated:
[0175]
[0176] in:
[0177] gs is the solar energy potential (in kW);
[0178] irr is the irradiance measurement (in W / m 2 as units);
[0179] tempPV is the temperature measurement (in °C) produced by the sensor used to sense the temperature of the photovoltaic cell.
[0180] Therefore, the solar energy potential can be calculated in real time. Therefore, a known history of solar energy potential data is available.
[0181] refer to Figure 6 , the solar energy potential is predicted as follows.
[0182] A weather forecast is made for a given date (step E30). Using a weather data API (such as Open WeatherMap), this makes it possible to collect hourly weather forecasts including cloud coverage (%), temperature, wind speed, precipitation, weather conditions (sunny, rainy, cloudy, etc.): step E31.
[0183] These historical weather forecast data and known solar energy potential values are used to train a machine learning algorithm (step E32). Once the model has been trained using the weather forecast, the solar energy potential value for the upcoming day can be predicted.
[0184] The learning algorithm used is the XGBoost algorithm, which is a supervised regression machine learning model based on the boosted gradient tree algorithm. Boosted gradient is a supervised learning algorithm that combines the results of a set of simple and weak models to obtain better predictions. A holistic approach can also be used by coupling XGBoost with a neural network. The two algorithms are trained independently, and in the prediction, there is a vote to select the best prediction. The voting-based regressor averages the individual predictions - here, the predictions from the neural network and XGBoost to form the final prediction.
[0185] The supervised learning algorithm makes it possible to obtain an estimate of the irradiance (in W / m 2 as units) (step E33).
[0186] The above formula for the solar potential is then used together with the irradiance estimate (step E34 ). The solar potential for date J is thus calculated (step E35 ).
[0187] The proposed method makes it possible to obtain solar energy potential values with hourly granularity.
[0188] The algorithm for predicting the energy requirements of a battery works as follows.
[0189] The charging demand of a battery is the energy required to charge the battery.
[0190] The energy sources in the solution proposed here are solar panels and batteries. Batteries are the only energy source at night, and therefore for the batteries to be able to provide energy at night, they must be able to be fully charged during the day using solar energy.
[0191] During the night, the power is negative and the battery is discharged. During the day, the power is positive and the battery is charged.
[0192] In order to meet the charging needs of the battery, it is necessary to have a model that determines the energy that the solar panel must provide to the battery to bring the state of charge to 100%, given the state of charge at sunrise. A "clear" day is when the solar panel is producing the most energy, usually more than the subscriber's needs, and the battery's state of charge reaches 100%.
[0193] refer to Figure 7 , the algorithm therefore comprises obtaining an initial state of charge equal to the state of charge of the battery at sunrise on a given date (for example at 6 a.m.) (step E40), implementing a linear regression model (step E41), and determining the amount of energy (in kWh) required to charge the battery starting from a given state of charge value (step E42).
[0194] The algorithm for predicting the evolution of the battery state of charge works as follows.
[0195] Starting from an initial state of charge value and a value of energy delivered to the battery, a relationship between the state of charge point and the energy delivered to the battery is found; this relationship is obtained using a machine learning algorithm, which in this case is a linear regression.
[0196] The evolution of the state of charge is linearly related to the energy required for this evolution of the state of charge. Using the state of charge value at sunrise and the energy demand of the battery at the input of the model, the evolution of the state of charge during this day when the battery is charged is calculated.
[0197] Likewise, a machine learning algorithm, in this case a linear regression, is trained to obtain a model that determines the change in state of charge when the battery starts delivering energy from its state of charge value at sunset.
[0198] Therefore, two models are used to determine the variation of the battery state of charge, one applicable during the day (in order to predict the evolution of the battery state of charge during battery charging) and another applicable during the night (in order to predict the evolution of the battery state of charge during battery discharging).
[0199] Of course, the invention is not limited to the embodiments described, but covers all variants coming within the ambit of the invention as defined by the claims.
Claims
1. A method for distributing electrical energy produced by a production plant (11) comprising at least one solar panel (12) and at least one battery (14) to a power grid (2) of at least one subscriber (1), said method comprising, for a given day including day and night, the following optimization steps implemented before said given day: - define the amount of energy for said given day; - predicting a final state of charge (SOC_f) equal to the state of charge of the battery at the end of the night on the given date and an assumed additional energy (Ex), the energy not produced by the solar panels during the day on the given date but which could be produced if the subscriber's consumption and / or the energy demand of the battery increased; - adjusting said amount of energy based on said final state of charge and said assumed additional energy; After said optimization step, said allocation method further comprises a step of allocating said amount of energy to subscribers of said electrical network during said given date.
2. The distribution method according to claim 1, wherein: The amount of energy includes an amount of daytime energy intended to be distributed during the daytime of the given date and an amount of nighttime energy intended to be distributed during the nighttime of the given date.
3. The distribution method according to claim 2, wherein: Adjusting the amount of energy comprises the following steps: checking whether the assumed additional energy (Ex) belongs to a predetermined energy interval, and if this is not the case, increasing or decreasing the amount of daytime energy, and then predicting the final state of charge and the assumed additional energy again, and repeating these steps until the assumed additional energy belongs to the predetermined energy interval.
4. The distribution method according to claim 2, wherein: Adjusting the amount of energy comprises the following steps: checking whether the final state of charge (SOC_f) belongs to a predetermined charging interval, and if this is not the case, increasing or decreasing the amount of night energy, then predicting the final state of charge and the assumed additional energy again, and repeating these steps until the final state of charge belongs to the predetermined charging interval.
5. The dispensing method according to claim 1, wherein: The prediction of the final state of charge and the assumed additional energy implements a set of prediction algorithms (30) that cooperate with each other, the set of prediction algorithms using as input an initial state of charge (SOC_i), which is equal to the state of charge of the battery at the beginning of the day on the given date, a weather forecast and the subscriber's consumption during the day before the given date.
6. The distribution method according to claim 5, wherein: The final state of charge (SOC_f) and the assumed additional energy (Ex) are predicted before the given date, for the given date, and for one or more days after the given date.
7. The distribution method according to claim 5, wherein: The set of forecasting algorithms (30) comprises: a forecasting algorithm (31) for the energy demand of the battery, a forecasting algorithm (32) for the solar energy potential, a forecasting algorithm (33) for the consumption of the subscribers and a forecasting algorithm (34a, 34b) for the evolution of the state of charge of the battery.
8. The distribution method according to claim 7, wherein: The prediction algorithm of the energy demand of the battery uses the initial state of charge (SOC_i) as input.
9. The distribution method according to claim 7, wherein: The prediction algorithm for solar energy potential uses weather forecasts as input.
10. The distribution method according to claim 8, wherein: The prediction algorithm of the solar energy potential implements a supervised learning algorithm which allows generating an estimate of the irradiance from which the solar energy potential is calculated.
11. The distribution method according to claim 7, wherein: The prediction algorithm of the subscriber's consumption uses as input the subscriber's consumption during the day preceding the given date.
12. The distribution method according to claim 7, wherein: The prediction algorithm of the evolution of the battery's state of charge uses as input the battery's energy demand, the solar energy potential and the subscriber's consumption during the day on the given date in order to predict the evolution of the battery's state of charge during charging of the battery.
13. The distribution method according to claim 7, wherein: The prediction algorithm of the evolution of the state of charge of the battery uses as input the consumption of the subscriber during the night of the given date, the state of charge at the end of the day of the given date and the assumed additional energy (Ex) in order to predict the evolution of the state of charge of the battery during discharge of the battery.
14. A processing module (10) in which the dispensing method according to one of the preceding claims is implemented.
15. A computer-readable recording medium on which a computer program including instructions is recorded, the instructions causing the processing module according to claim 14 to execute the steps of the distribution method according to any one of claims 1 to 13.
16. A computer program product comprising a computer program including instructions causing a processing module according to claim 14 to execute the steps of the distribution method according to one of claims 1 to 13.
Citation Information
Patent Citations
Storage-battery control device, storage-battery charge / discharge system, photovoltaic generation system, and storage-battery control method
CN107710542A
Method for distributing electrical energy from solar energy to a plurality of groups
CN109217356A
Energy storage system
KR101597993B1
