Network load coordinated regulation strategy based on multi-objective genetic algorithm
By building a reinforcement learning model and multi-objective genetic algorithm based on thermal power units and optimizing the grid control strategy, the problem that traditional grid dispatching is difficult to strike a balance between high efficiency and environmental protection in new power systems is solved, and low-carbon economic operation and load control are achieved.
Patent Information
- Application Number
- CN202410271345.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-11
- Publication Date
- 2025-09-12
AI Technical Summary
Traditional power grid dispatching is difficult to simultaneously meet the efficiency, environmental protection and economic requirements of adjustable loads in new power systems. Especially under the dual carbon goals, it is difficult to effectively regulate the volatility and uncertainty of new energy sources such as wind power and photovoltaics, resulting in high power grid operating costs and large carbon emissions.
A reinforcement learning model is constructed based on the historical active power, reactive power and carbon emissions of thermal power units. Combined with a multi-objective genetic algorithm, the output curves of thermal power, hydropower and energy storage units are optimized. With the goal of minimizing grid operating costs and carbon emissions, a grid-load coordinated control strategy is constructed.
It has achieved low-carbon economic operation in the new power system, reduced grid operating costs and carbon emissions, met the needs of adjustable loads, and promoted the development of the power system towards intelligence and cleanliness.
Smart Images
Figure FT_1 
Figure SMS_1 
Figure SMS_3
Abstract
Description
Technical Field
[0001] The present invention relates to the fields of reinforcement learning and load collaborative regulation, and specifically discloses a grid-load collaborative regulation strategy based on a multi-objective genetic algorithm. Background Art
[0002] With the construction of new power systems and power markets, the supply side faces many challenges. Against the backdrop of the global pursuit of dual carbon goals, namely carbon peak and carbon neutrality, the green and low-carbon transformation of the power system has become key. The commissioning of a variety of new adjustable loads not only breaks the current situation where the power grid can only passively adapt to load changes, but also puts forward new requirements for the adjustment of power services on the supply side and the reduction of carbon emissions. The current power supply and demand situation is tense, and traditional power grid dispatching and operation methods are difficult to meet the needs of the development of adjustable loads, let alone meet the carbon emission requirements under the dual carbon goals. Therefore, a model with the minimum grid operating cost and the lowest carbon emissions as the control target should be established to solve the problems of grid peak and frequency regulation caused by fluctuations in renewable energy output, and promote the development of power systems in a more intelligent and clean direction.
[0003] Current research focuses primarily on the uncertainties of wind power grid integration and the impact of renewable energy on the power system, but fails to address the simultaneous environmental impact, efficiency, and economic benefits of coordinated dispatch. Therefore, a new trend is emerging: a strategy that minimizes grid operating costs and carbon emissions, using traditional thermal power units, hydropower units, and distributed power sources to balance the volatility and uncertainty of renewable energy-based power sources.
[0004] Therefore, based on the above background, the present invention first constructs a grid carbon emission model based on reinforcement learning based on the historical active power, reactive power, and corresponding carbon emissions of thermal power units, accurately measures the carbon emissions of thermal power units in the grid, and then, based on wind power, photovoltaic, and load information, and under the condition of meeting the grid operation constraints, with the minimum electricity operation cost and the minimum carbon emissions as the goals, constructs a grid control strategy based on a multi-objective genetic algorithm. Finally, the output curves of thermal power units, hydropower units, and energy storage units are given to achieve the goal of low-carbon and economical operation of the new power system. Summary of the Invention
[0005] (1) Purpose of the invention
[0006] The purpose of this invention is to overcome the shortcomings of traditional power grid dispatching and provide a grid-load coordinated control strategy method based on a multi-objective genetic algorithm. First, a reinforcement learning-based grid carbon emission model is constructed based on the historical active power, reactive power, and corresponding carbon emissions of thermal power units to accurately measure the carbon emissions of thermal power units in the power grid. Then, based on wind power, photovoltaic power, and load information, and with the goal of minimizing electricity operating costs and carbon emissions while meeting grid operation constraints, a multi-objective genetic algorithm-based grid control strategy is constructed. Finally, output curves for thermal power units, hydropower units, and energy storage units are generated to achieve the goal of low-carbon and economical operation of the new power system.
[0007] (2) Technical solution
[0008] In order to achieve the above-mentioned purpose, the technical solution adopted by the method of the present invention is: first, according to the operating cost F of the thermal power unit, Gj , hydropower unit operating cost F hj , Energy storage unit operating cost F Cj And line network loss F loss A grid operation cost objective function is constructed, and then a grid carbon emission objective function is constructed based on the grid carbon emissions generated by thermal power units during the scheduling period. Then, based on the output of thermal power units and the corresponding carbon emission data, the LSTM algorithm is used to train a carbon emission model corresponding to the unit output of thermal power units, and the carbon emissions of thermal power units in the grid are accurately measured.
[0009] When training the carbon emission model corresponding to the unit output of the thermal power unit, the active power, reactive power, and fuel carbon emission factor of the thermal power unit Gj are first collected as input samples, and the corresponding carbon emission data are input into the hidden layer. Then, based on the input sample data and cell state information, the LSTM deep learning network is trained to finally obtain the carbon emission model.
[0010] Under the constraints of active power balance constraints, wind power generation constraints, photovoltaic power generation constraints, hydropower generation constraints, thermal power unit constraints, energy storage constraints, line constraints and load loss constraints, according to the carbon emissions of the thermal power units in the power grid, a power grid control strategy based on a multi-objective genetic algorithm is given to obtain the output curves of thermal power units, hydropower units and energy storage units, and a control strategy that meets the low-carbon and economical operation needs of the new power system is obtained.
[0011] (3) Beneficial effects
[0012] The algorithm proposed in the present invention can regulate the grid load to achieve the minimum grid operation cost and the lowest carbon emissions. In the new power system environment, the high proportion of random loads caused by the increase in the proportion of high-volatility power sources represented by wind power and photovoltaics, as well as controllable air conditioners and electric vehicles connected will increase the grid operation cost, and the traditional grid dispatching and operation methods are difficult to meet the needs of adjustable load development, and cannot meet the carbon emission requirements under the dual carbon goals. Therefore, based on the above two points, firstly, based on the historical active power, reactive power, and corresponding carbon emissions of thermal power units, a grid carbon emission model based on reinforcement learning is constructed to accurately measure the carbon emissions of thermal power units in the grid. At the same time, based on wind power, photovoltaics, and load information, under the condition of meeting the grid operation constraints, with the goal of minimizing electricity operation costs and minimizing carbon emissions, a grid control strategy based on a multi-objective genetic algorithm is constructed, and the output curves of thermal power units, hydropower units, and energy storage units are given to obtain a control strategy that meets the low-carbon and economical operation needs of the new power system. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Figure 1 It is a grid-load coordinated control strategy diagram based on a multi-objective genetic algorithm. Create a flow chart. DETAILED DESCRIPTION
[0014] The grid operation cost F1 mainly includes the thermal power unit operation cost F Gj , the operating cost of the hydropower unit F hj , the operating cost of the energy storage unit F Cj And line network loss F loss , specifically as shown in formula (1). N G 、N h 、N C 、N loss They are the number of thermal power units, hydropower units, energy storage units and the total number of power grid lines.
[0015]
[0016] Thermal power unit operating cost F Gj The calculation model is shown in formula (2), where aGj, bG j , cGj is the energy consumption characteristic parameter of thermal power unit Gj. In the new power system environment, thermal power units will bear more power balancing tasks, and their start and stop will be more frequent, so their start and stop costs need to be considered. is a 0-1 variable, indicating the operating state of the unit Gj at time t. When the value is 1, it indicates that the unit is in the power generation state, and when the value is 0, it indicates that the unit is in the stopped state. Gj The cost of starting and stopping unit Gj once.
[0017]
[0018] Hydropower unit operating cost F hj The calculation model is shown in formula (3). The hydropower unit only has the start-up and shutdown costs, where It is a 0-1 variable, indicating the operating status of the hydropower unit hj at time t. When the value is 1, it indicates that the hydropower unit is in the power generation state, and when the value is 0, it indicates that the hydropower unit is in the stopped state. hj is the cost of starting and stopping the hydropower unit hj once.
[0019]
[0020] Energy storage unit operating cost F Cj The calculation model is shown in formula (4), where is the charging and discharging power cost coefficient of the energy storage unit Cj.
[0021]
[0022] Line network loss F loss j The calculation model is shown in formula (5), where: is the current value of line j at time, R j , X j are the resistance and reactance of the line respectively.
[0023]
[0024] During the dispatching period, the carbon emissions of the power grid are mainly generated by thermal power units. The constructed carbon emissions objective function of the power grid is shown in formula (6):
[0025]
[0026] In the formula is the output of the thermal power unit at time t, is the unit power carbon emission of thermal power unit j at time t, which is mainly related to the output of thermal power unit The carbon emission factor of the fuel of thermal power units is related to the product, as shown in formula (7):
[0027]
[0028] is the fuel carbon emission factor of thermal power unit j at time t. Based on the output and corresponding carbon emission data of thermal power units, this paper uses the LSTM algorithm to train the carbon emission model corresponding to the unit output of thermal power units shown in formula (7).
[0029] First, the active power, reactive power, and fuel carbon emission factor of the thermal power unit Gj are collected as input samples, and the corresponding carbon emission data are input into the hidden layer.
[0030] Then, based on the input sample data and cell state information, the LSTM deep learning network is trained. The training process is shown in formula (8):
[0031]
[0032] In formula (8), σ is the activation function; f t is the output of the forget gate, W f 、b f is the corresponding forget gate matrix; i t is the output of the input gate, W i 、b i is the corresponding input gate weight matrix; C t-1 is the old cell state information, Add candidate status information for selection, C t For updated cell information, W C 、b C is the corresponding neuron matrix; o t is the output of the output gate, W o 、b o is the corresponding output gate matrix; h t is the output result.
[0033] The constraints mainly include active power balance constraints, wind power generation constraints, photovoltaic power generation constraints, hydropower generation constraints, thermal power unit constraints, energy storage constraints, line constraints and load loss constraints.
[0034] The system power must meet the balance condition, as shown in formula (9). is the charging power of the j-th energy storage unit at time t, is the discharge power of the j-th energy storage unit at time t, is the power of the j-th load at time t, The output of the j-th wind farm at time t is: The jth photovoltaic station outputs power at all times. For the output of the j-th hydropower station at time t, is the output of the jth thermal power unit at time t. N c is the number of energy storage units in the charging state, Nc is the number of energy storage units in the discharging state, N L is the load quantity, N PW is the number of wind farms, N PV is the number of photovoltaic power stations, N h is the number of hydropower stations, N G is the number of thermal power units.
[0035]
[0036] Wind power generation constraints mainly include wind power ramping constraints and wind power abandonment constraints, as shown in formula (10). is the wind power ramp rate, are the minimum and maximum ramp rates of wind power, respectively. Δt is the time interval between two wind power values at time t. is the wind power output at time t.
[0037]
[0038] Wind power penetration rate P′ PW The ratio of the actual wind power output to the total load in the cycle is taken as shown in formula (11):
[0039]
[0040] Photovoltaic power generation constraints mainly include operation constraints, ramp constraints, and curtailment constraints, as shown in formula (12). is the lower limit of photovoltaic power generation power, Photovoltaic power upper limit, P′ PV,t is the photovoltaic power generation power at time t, are the photovoltaic down and up climbing rate limits, P′ PVD,t is the abandoned optical power at time t.
[0041]
[0042] Hydropower generation constraints mainly include turbine output constraints and available water constraints, as shown in formula (13). are the minimum and maximum outputs of the turbine generator, are the maximum and minimum water consumption of the hydropower station, respectively. The use of hydropower resources is determined by the upper and lower limits of the water level that determine power dispatch.
[0043]
[0044] The constraints of thermal power units mainly include operation constraints and ramp constraints, as shown in formula (14). is the lower limit of the output of the i-th thermal power unit, is the upper limit of the output of the i-th thermal power unit, is the lower climbing rate limit value of the i-th thermal power unit, is the upper limit value of the climbing rate of the i-th thermal power unit.
[0045]
[0046] Energy storage constraints mainly include energy storage power constraints, energy storage charge rate constraints, and energy storage capacity constraints. The energy storage power constraint is shown in formula (15). is the maximum charging power of energy storage, is the maximum discharge power of energy storage.
[0047]
[0048] The energy storage charge constraint is shown in formula (16). Setting this constraint can avoid overcharging and over-discharging of energy storage. max , SOC min is the maximum and minimum charge rate of energy storage, SOC0 is the initial charge rate of energy storage system, E t is the energy storage capacity.
[0049]
[0050] The energy storage capacity constraint is shown in Equation (17), which ensures that the energy storage capacity meets the requirements at all times. Max , E Min are the upper and lower limits of the energy storage capacity (here 100% of the rated capacity and 10% of the rated capacity respectively), Δt is the time interval, η1 and η2 are the efficiency of energy storage discharge and charging respectively.
[0051]
[0052] Line constraints mainly include line power equality constraints, line transmission power constraints and node phase angle constraints. As shown in formula (18), P ij,t is the transmission power of the line with i and j as endpoints at time t, B ij is the susceptance of the line, θ i,t ,θ j,t is the phase angle of node ij at time t, and P is the thermal stability limit of the line.
[0053]
[0054] The load loss constraint means that the load loss power of a node in the system cannot be greater than the load power of the node. Specifically, it is shown in formula (4-58): 0≤P LDj,t ≤P Lj,t
[0055] Thus far, the technical solutions of the present invention have been described in conjunction with the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the invention of this specification, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.
Claims
1. A grid-load coordinated control strategy based on a multi-objective genetic algorithm, characterized by: First, according to the operating cost F of the thermal power unit Gj , hydropower unit operating cost F hj , Energy storage unit operating cost F Cj And line network loss F loss A grid operation cost objective function is constructed, and then a grid carbon emission objective function is constructed based on the grid carbon emissions generated by thermal power units during the dispatching period. Then, based on the output of thermal power units and the corresponding carbon emission data, the LSTM algorithm is used to train a carbon emission model corresponding to the unit output of thermal power units to accurately measure the carbon emissions of thermal power units in the grid. Finally, the output curves of thermal power units, hydropower units, and energy storage units are given through a grid control strategy based on a multi-objective genetic algorithm to meet the low-carbon and economical operation needs of the new power system.
2. A grid-load coordinated control strategy based on a multi-objective genetic algorithm according to claim 1, characterized in that: The main steps of constructing the grid operation cost objective function include: Step 1: The grid operation cost F1 mainly includes the thermal power unit operation cost F Gj , the operating cost of the hydropower unit F hj , the operating cost of the energy storage unit F Cj And line network loss F loss . Step 2: Thermal power unit operating cost F Gj The calculation model is as follows, where a Gj 、b Gj 、c Gj is the consumption characteristic parameter of thermal power unit Gj. is a 0-1 variable, indicating the operating state of the unit Gj at time t. When the value is 1, it indicates that the unit is in the power generation state, and when the value is 0, it indicates that the unit is in the stopped state. Gj The cost of starting and stopping unit Gj once. Step 3: Hydropower unit operating cost F hj , the calculation model is as follows, where It is a 0-1 variable, indicating the operating status of the hydropower unit hj at time t. When the value is 1, it indicates that the hydropower unit is in the power generation state, and when the value is 0, it indicates that the hydropower unit is in the stopped state. hj is the cost of starting and stopping the hydropower unit hj once. Step 4: Energy storage unit operating cost F Cj The calculation model is as follows, where is the charging and discharging power cost coefficient of the energy storage unit Cj. Step 5: Line network loss F loss j The calculation model is as follows, where: is the current value of line j at time, R j , X j are the resistance and reactance of the line respectively. Step 6: Get the grid operation cost F1: Where N G 、N h 、N C 、N loss They are the number of thermal power units, hydropower units, energy storage units and the total number of power grid lines.
3. The grid-load coordinated control strategy based on a multi-objective genetic algorithm according to claim 1 is characterized in that: The LSTM deep learning network is used to obtain the carbon emissions model corresponding to the unit output of thermal power units to construct the carbon emissions target function of the power grid. The main steps include: Step 1: First, collect the active power, reactive power, and fuel carbon emission factor of the thermal power unit Gj as input samples, and input the corresponding carbon emission data into the hidden layer. Then, based on the input sample data and cell state information, train the LSTM deep learning network. The training process is as follows: Where σ is the activation function; f t is the output of the forget gate, W f 、b f is the corresponding forget gate matrix; i t is the output of the input gate, W i 、b i is the corresponding input gate weight matrix; C t-1 is the old cell state information, Add candidate status information for selection, C t For updated cell information, W C 、b C is the corresponding neuron matrix; o t is the output of the output gate, W o 、b o is the corresponding output gate matrix; h t is the output result. Step 2: Based on the output and corresponding carbon emissions data of thermal power units, use the LSTM algorithm to train a carbon emissions model corresponding to the unit output of thermal power units: is the fuel carbon emission factor of thermal power unit j at time t. Step 3: During the dispatch period, grid carbon emissions are mainly generated by thermal power units. The constructed grid carbon emissions objective function is as follows: In the formula is the output of the thermal power unit at time t, is the unit power carbon emission of thermal power unit j at time t, which is mainly related to the output of thermal power unit The carbon emission factor of the fuel of thermal power units is related to the product.
4. The grid-load coordinated control strategy based on a multi-objective genetic algorithm according to claim 1 is characterized in that: To clearly construct the constraint function under the conditions, the main steps include: Step 1: The active power balance constraint is shown in the following equation. is the charging power of the j-th energy storage unit at time t, is the discharge power of the j-th energy storage unit at time t, is the power of the j-th load at time t, The output of the j-th wind farm at time t is: The jth photovoltaic station outputs power at all times. For the output of the j-th hydropower station at time t, The output of the j-th thermal power unit at time t. Nc is the number of energy storage units in charging state, Nc is the number of energy storage units in the discharge state, NL is the load quantity, NPW is the number of wind farms, NPV is the number of photovoltaic power stations, Nh is the number of hydropower stations, NG is the number of thermal power units. Step 2: Wind power generation constraints mainly include wind power ramping constraints and wind power curtailment constraints, as shown in the following formula. is the wind power ramp rate, are the minimum and maximum ramp rates of wind power respectively. Δt is the time interval between two wind power outputs at time t. is the wind power output at time t. Wind power penetration rate P′ PW The ratio of the actual wind power output to the total load in the cycle is as shown in the following formula: Step 3: PV power generation constraints mainly include operation constraints, ramp constraints, and curtailment constraints, as shown in the following formula. is the lower limit of photovoltaic power generation power, The upper limit of photovoltaic power generation power, is the photovoltaic power generation power at time t, They are the photovoltaic down and up climbing rate limit values, is the abandoned optical power at time t. Step 4: Hydropower generation constraints mainly include turbine output constraints and available water constraints, as shown in the following formula. are the minimum and maximum outputs of the turbine generator, are the maximum and minimum water consumption of the hydropower station, respectively. The use of hydropower resources is determined by the upper and lower limits of the water level that determine power dispatch. Step 5: The constraints of the thermal power unit mainly include operation constraints and ramp constraints, as shown in the following formula. is the lower limit of the output of the i-th thermal power unit, is the upper limit of the output of the i-th thermal power unit, is the lower climbing rate limit value of the i-th thermal power unit, is the upper limit value of the ramp rate of the i-th thermal power unit. Step 6: Energy storage constraints mainly include energy storage power constraints, energy storage charge rate constraints, and energy storage capacity constraints. The energy storage power constraints are shown in the following formula. is the maximum charging power of energy storage, is the maximum discharge power of energy storage. Step 7: Energy storage charge constraint is shown in the following formula. Setting this constraint can avoid overcharge and overdischarge of energy storage. max , SOC min is the maximum and minimum charge rate of energy storage, SOC0 is the initial charge rate of energy storage system, E t is the energy storage capacity. Step 8: The energy storage capacity constraint is shown in the following formula to ensure that the energy storage capacity meets the requirements at all times. Max , E Min are the upper and lower limits of the energy storage capacity (here 100% of the rated capacity and 10% of the rated capacity respectively), Δt is the time interval, η1 and η2 are the efficiency of energy storage discharge and charging respectively. Step 9: Line constraints mainly include line power equality constraints, line transmission power constraints and node phase angle constraints. As shown in the following formula, P ij,t is the transmission power of the line with i and j as endpoints at time t, B ij is the susceptance of the line, θ i,t ,θ j,t is the phase angle of node ij at time t, and P is the thermal stability limit of the line. Step 10: Load loss constraint means that the load loss power of a node in the system cannot be greater than the load power of the node. The specific formula is as follows: 0≤P LDj,t ≤P Lj,t 5. The grid-load coordinated control strategy based on a multi-objective genetic algorithm according to claim 1 is characterized in that: Under the constraints of active power balance constraints, wind power generation constraints, photovoltaic power generation constraints, hydropower generation constraints, thermal power unit constraints, energy storage constraints, line constraints and load loss constraints, the grid operation cost objective function and the grid carbon emission objective function are constructed. Then, under the influence of the two objective functions, the output curves of thermal power units, hydropower units and energy storage units are given through the grid control strategy based on the multi-objective genetic algorithm, thus completing the control optimization with minimum electricity operation cost and minimum carbon emissions.