A multi-level energy management system

Abstract

The present invention provides a multi-level energy management system, which solves the technical problem that the existing energy management system only distributes power demand, but has high electricity charges and high consumption costs during peak hours, and can be widely used in the field of energy management; specifically, it includes an energy supply module, an energy storage module, a load module and a control module, the energy supply module includes a photovoltaic power generation device and city power, the energy storage module includes a battery, the load module includes electrical equipment, and the control module includes a data acquisition unit, a data scheduling unit and a data optimization unit; the data acquisition unit is used to collect power data of other modules, the data scheduling unit is used to reasonably distribute the output power of the energy supply module to the energy storage module and the load module, and the data optimization unit is used to dynamically adjust the allocation of the data scheduling unit according to the current user's load module power consumption.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy management, and in particular relates to a multi-level energy management system. Background Art

[0002] The patent with publication number CN116720717B discloses a building energy management and regulation system and method, including a power record screening module, a power data analysis module, a scheduling matching module and a demand scheduling generation module; by obtaining a model curve of the power demand of the power-consuming object category in the building within a period, all power inputs that can be allocated to the building in this time period are managed and allocated according to the importance of the power-consuming object in life and safety, so as to meet the basic power supply needs and avoid the situation during the peak period of power usage where some important power-consuming objects and power-consuming equipment cannot obtain sufficient power energy due to power distribution problems, thereby affecting normal operation and life.

[0003] The above patent is only carried out by allocating electricity demand, but during the peak period of electricity consumption, the electricity price is high and the consumption cost is high. Summary of the invention

[0004] The purpose of the present invention is to solve the deficiencies of the above-mentioned technologies and provide a multi-level energy management system.

[0005] To this end, the present invention provides a multi-level energy management system, including an energy supply module, an energy storage module, a load module and a control module, the energy supply module includes a photovoltaic power generation device and city power, the energy storage module includes a battery, the load module includes an electrical device, and the control module includes a data acquisition unit, a data scheduling unit and a data optimization unit;

[0006] The data acquisition unit is used to collect power data of other modules, the data scheduling unit is used to reasonably distribute the output power of the energy supply module to the energy storage module and the load module, and the data optimization unit is used to dynamically adjust the allocation of the data scheduling unit according to the current user's load module power consumption.

[0007] Furthermore, a power distribution model based on a neural network is provided in the data scheduling unit, and the steps of establishing the power distribution model include:

[0008] The total output of the city power and photovoltaic power generation is E out , photovoltaic power generation power P PV , mains power P grid , power of electrical load P load , peak electricity rate coefficient C peak , off-peak electricity rate coefficient C offpeak , battery energy storage efficiency η battery , the amount of electricity stored in the battery E storage , the amount of electricity consumed by the electrical load Eload ,

[0009] E out =P PV +P grid ;

[0010] During peak electricity rates, it is preferred to use the electricity for load consumption to reduce the electricity bill during peak periods. During low electricity rates, it is preferred to store the electricity in batteries to optimize electricity costs by using low electricity rates.

[0011] When electricity rates are high, prioritize the consumption of electricity from the city and photovoltaic power generation, and minimize storage;

[0012] E load =min(P load ,E out );

[0013] E storage =0;

[0014] When electricity rates are low, electricity is stored in batteries first to reduce electricity costs during peak hours;

[0015] E storage =min(P PV +P grid ,(C peak -C offpeak )·P load );

[0016] E load =min(P load ,E out -E storage );

[0017] The storage efficiency of a battery is a constant η battery , used to calculate the amount of electricity actually stored in the battery:

[0018] E battery_input =E storage ·η battery ;

[0019] Among them, E battery_input It is the amount of electricity actually stored in the battery;

[0020] Obtain local historical data and establish a neural network model. The input of the neural network is: current time, current load, current photovoltaic power generation, current electricity fee, and current battery status; the output of the neural network is: battery storage power and load consumption power;

[0021] The training of a neural network can be done by minimizing the following loss function:

[0022] L = α(E battery_input -E storage ) 2 +β(E load -P load ) 2 ;

[0023] Among them, α and β are weight coefficients used to balance the accuracy of storage and consumption of electricity.

[0024] Furthermore, local historical data includes load, photovoltaic power generation, and historical electricity charges, and the current electricity charge status includes charging power and discharging power.

[0025] Furthermore, in the data optimization unit, the step of optimizing the power distribution model includes:

[0026] The total output power E at time t is recorded out (t) = P grid (t)+P PV (t), the amount of electricity stored at time t, E storage (t), the amount of electricity consumed at time t E load (t), the maximum capacity of the battery C battery ;

[0027] The optimization target of power distribution is set as:

[0028]

[0029] Among them: γ1, γ2, γ3 are adjustment coefficients, which control the weights between various targets. battery is the maximum capacity of the battery, C load (t) is the user's load demand at time t, Cost(t) is the electricity cost at time t;

[0030] And the distribution of electricity must meet the following constraints:

[0031] The charge and discharge of the battery cannot exceed the capacity limit of the battery:

[0032] 0≤E storage (t)≤C battery ;

[0033] The amount of power stored and consumed should be determined by the power output and load demand:

[0034] E storage (t)+E load (t) = E out (t);

[0035] Battery loss:

[0036] Ebattery_input (t) = η battery ·E storage (t);

[0037] During peak hours, the priority of electricity consumption is adjusted:

[0038] E out (t) = min(P load (t),E out (t),C peak (t)>C offpeak (t);

[0039] E load (t) = min(P load (t),E out (t)-E storange (t)),C peak (t)

[0040] ≤C offpeak (t);

[0041] Electricity fee model: Assuming that the electricity fee is different during peak and off-peak periods, the electricity fee is defined as:

[0042]

[0043] Assuming that the distribution of electricity is adjusted dynamically in multiple time periods, at a certain time t, the objective function becomes: J(t) = min(γ1·(C battery -E storange (t))+γ2·E load (t)+γ3·Cost(t)).

[0044] Furthermore, the step of optimizing the power distribution model also includes:

[0045] Define a state S(t) = [E storage(t) ,E load (t)], at time t, the state is updated by selecting an action (storing or consuming power) and the corresponding cost is calculated;

[0046] The recursive formula for dynamic programming is:

[0047]

[0048] Where a is the action of the battery, C(a,S(t)) is the immediate cost of taking action in the current state, and γ is a discount factor used to weigh future costs.

[0049] The present invention provides a multi-level energy management system, which has the following beneficial effects:

[0050] The present invention can dynamically optimize the distribution of electricity according to factors such as city electricity demand, photovoltaic power generation, load, electricity price fluctuations, and battery efficiency through the design of a neural network model. In practical applications, the neural network model is trained using historical data to effectively predict future electricity distribution and automatically adjust the strategy according to electricity prices and load conditions.

[0051] And by adopting complex dynamic optimization algorithms, it is possible to dynamically optimize the distribution of electricity according to the actual electricity consumption of users. Combined with dynamic programming or reinforcement learning methods, it is possible to effectively calculate the optimal storage and consumption strategy based on factors such as users' real-time needs, power generation, battery capacity, and electricity price fluctuations. DETAILED DESCRIPTION

[0052] The present invention is further described below in conjunction with specific examples to help understand the content of the present invention. The methods used in the present invention are all conventional methods unless otherwise specified; the raw materials and devices used are all conventional commercially available products unless otherwise specified.

[0053] The present invention provides a multi-level energy management system, including an energy supply module, an energy storage module, a load module and a control module. The energy supply module includes a photovoltaic power generation device and city power, the energy storage module includes a battery, the load module includes electrical equipment, and the control module includes a data acquisition unit, a data scheduling unit and a data optimization unit.

[0054] The data acquisition unit is used to collect power data of other modules. For example, the daily output power of the energy supply module, the current remaining power of the energy storage module, the daily power consumption of the load module and other data. The data scheduling unit is used to reasonably distribute the output power of the energy supply module to the energy storage module and the load module. The data optimization unit is used to dynamically adjust the allocation of the data scheduling unit according to the current user's load module power consumption to reduce the cost of use.

[0055] The data scheduling unit is provided with a power distribution model based on a neural network. The steps of establishing the power distribution model include:

[0056] The total output of the city power and photovoltaic power generation is E out , photovoltaic power generation power P PV , mains power P grid , power of electrical load P load , peak electricity rate coefficient C peak , off-peak electricity rate coefficient C offpeak , battery energy storage efficiency η battery , the amount of electricity stored in the battery E storage , the amount of electricity consumed by the electrical load E load .

[0057] E out =PPV +P grid ;

[0058] During the peak electricity bill period, the power is first consumed by the load to reduce the electricity bill during the peak period. During the low electricity bill period, the power is first stored in the battery to optimize the electricity cost by using the low electricity bill.

[0059] When electricity rates are high, priority is given to consuming electricity from the mains and photovoltaic power generation, and storage is minimized.

[0060] E load =min(P load ,E out );

[0061] E storage =0 (try not to store electricity);

[0062] When electricity rates are low, electricity is stored in batteries first to reduce electricity costs during peak hours.

[0063] E storage =min(P PV +P grid ,(C peak -C offpeak )·P load );

[0064] E load =min(P load ,E out -E storage );

[0065] The storage efficiency of a battery is a constant η battery , used to calculate the amount of electricity actually stored in the battery:

[0066] E battery_input =E storage ·η battery ;

[0067] Among them, E battery_input It is the amount of electricity actually stored in the battery.

[0068] In actual applications, the above formula is adjusted dynamically. As time goes by and electricity prices change, the system will automatically adjust the distribution of electricity according to real-time city power demand, load demand and electricity price levels.

[0069] Obtain local historical data (such as load, photovoltaic power generation, historical electricity charges, etc.) and establish a neural network model. The input of the neural network is: current time, current load, current photovoltaic power generation, current electricity charges (peak or off-peak period), current battery status (charging power, discharging power); the output of the neural network is: battery storage power, load consumption power.

[0070] The training of a neural network can be done by minimizing the following loss function:

[0071] L = α(E battery_input -E storage ) 2 +β(E load -P load ) 2 ;

[0072] Among them, α and β are weight coefficients used to balance the accuracy of storage and consumption of electricity.

[0073] Through the above formula and neural network model design, the distribution of electricity can be dynamically optimized according to factors such as city electricity demand, photovoltaic power generation, load, electricity price fluctuations, and battery efficiency. In practical applications, using historical data to train the neural network model can effectively predict future electricity distribution and automatically adjust the strategy according to electricity prices and load conditions.

[0074] When the data optimization unit is put into use by users, the data scheduling unit is used as the control logic to adjust the power consumption. In the subsequent use process, the current user's power consumption habits are continuously collected, and a dynamic optimization framework is built to optimize the power distribution model using constraint optimization. The optimization steps include:

[0075] The total output power E at time t is recorded out (t) = P grid (t)+P PV (t), the amount of electricity stored at time t, E storage (t), the amount of electricity consumed at time t E load (t), the maximum capacity of the battery C battery .

[0076] The optimization target of power distribution is set as:

[0077]

[0078] Among them: γ1, γ2, γ3 are adjustment coefficients, which control the weights between various targets. battery is the maximum capacity of the battery, C load (t) is the user's load demand at time t, and Cost(t) is the electricity cost at time t.

[0079] And the distribution of electricity must meet the following constraints:

[0080] 1. The charge and discharge of the battery cannot exceed the capacity limit of the battery:

[0081] 0≤E storage (t)≤C battery ;

[0082] 2. The amount of power stored and consumed should be determined by power output and load demand:

[0083] E storage (t)+E load (t) = E out (t);

[0084] 3.Battery loss:

[0085] E battery_input (t) = η battery ·E storage (t);

[0086] During peak hours, the priority of electricity consumption is adjusted:

[0087] E out (t) = min(P load (t),E out (t),C peak (t)>C offpeak (t);

[0088] E load (t) = min(P load (t),E out (t)-E storange (t)),C peak (t)≤C offpeak (t);

[0089] Electricity fee model: Assuming that the electricity fee is different during peak and off-peak periods, the electricity fee is defined as:

[0090]

[0091] Assuming that we want to dynamically adjust the distribution of electricity in multiple time periods, at a certain time t, the objective function becomes:

[0092] J(t)=min(γ1·(C battery -E storange (t))+γ2·E load (t)+γ3·Cost(t)).

[0093] Two methods are provided here for the subsequent optimization process, including dynamic programming (DP) and reinforcement learning (RL).

[0094] 1. Dynamic Programming (DP)

[0095] Dynamic programming can gradually optimize the battery storage and consumption strategy through recursive relations. Define a state S(t) = [E storage(t) ,E load (t)], at time t, the state is updated by selecting an action (storing or consuming power) and the corresponding cost is calculated.

[0096] The recursive formula for dynamic programming is:

[0097]

[0098] Where a is a possible action (storing or consuming electricity), C(a,S(t)) is the immediate cost of taking the action in the current state, and γ is a discount factor used to weigh future costs.

[0099] 2. Reinforcement Learning (RL)

[0100] Reinforcement learning selects the optimal power allocation strategy under different electricity prices, load demands, and power generation conditions by training the power allocation model. The state S(t) in the model may include information such as time, load demand, battery power, and electricity costs. The power allocation model maximizes future rewards by selecting appropriate actions (for example, choosing to store or consume power).

[0101] Here, Q-learning is used to train the power distribution model. The core update formula of Q-learning is:

[0102]

[0103] Where: Q(S(t),a) is the expected return of taking action a in state S(t), R(t) is the immediate reward, which can be defined based on the cost and benefit of power allocation, and α is the learning rate, which controls the update step size.

[0104] By adopting complex dynamic optimization algorithms, the distribution of electricity can be dynamically optimized according to the actual electricity consumption of users. Combined with dynamic programming or reinforcement learning methods, the optimal storage and consumption strategy can be effectively calculated based on factors such as users' real-time needs, power generation, battery capacity, electricity price fluctuations, etc.

[0105] The core of the data optimization unit is to minimize electricity costs while maximizing the battery storage efficiency and load power supply efficiency. Combined with reinforcement learning or dynamic programming models, the system's adaptability and robustness can be further improved to ensure effective operation in different electricity markets and electricity consumption modes.

[0106] However, what is described above is only a specific embodiment of the present invention and should not be used to limit the scope of implementation of the present invention. Therefore, the replacement of equivalent components, or equivalent changes and modifications made according to the scope of protection of the present invention should still fall within the scope covered by the claims of the present invention.

Claims

1. A multi-level energy management system, comprising an energy supply module, an energy storage module, a load module and a control module, wherein the energy supply module comprises a photovoltaic power generation device and city electricity, the energy storage module comprises a battery, the load module comprises an electrical device, and the control module comprises a data acquisition unit, a data scheduling unit and a data optimization unit; The data acquisition unit is used to collect the power data of other modules, and is characterized by: The data scheduling unit is used to reasonably distribute the output power of the energy supply module to the energy storage module and the load module, and the data optimization unit is used to dynamically adjust the allocation of the data scheduling unit according to the current user's load module power consumption.

2. The multi-level energy management system according to claim 1, characterized in that: The data scheduling unit is provided with a power distribution model based on a neural network. The steps of establishing the power distribution model include: The total output of the city power and photovoltaic power generation is E out , photovoltaic power generation power P PV , mains power P grid , power of electrical load P load , peak electricity rate coefficient C peak , off-peak electricity rate coefficient C offpeak , battery energy storage efficiency η battery , the amount of electricity stored in the battery E storage , the amount of electricity consumed by the electrical load E load , E out =P PV +P grid ; During peak electricity rates, it is preferred to use the electricity for load consumption to reduce electricity costs during peak periods. During low electricity rates, it is preferred to store the electricity in batteries to optimize electricity costs by using low electricity rates. When electricity rates are high, prioritize the consumption of electricity from the city and photovoltaic power generation, and minimize storage; E load =min(P load ,E out ); E storage ＝0； When electricity rates are low, electricity is stored in batteries first to reduce electricity costs during peak hours; E storage =min(P PV +P grid ,(C peak -C offpeak )·P load ); E load =min(P load ,E out -E storage ); The storage efficiency of a battery is a constant η battery , used to calculate the amount of electricity actually stored in the battery: AND battery_input =And storage ·η battery ; Among them, E battery_input It is the amount of electricity actually stored in the battery; Obtain local historical data and establish a neural network model. The input of the neural network is: current time, current load, current photovoltaic power generation, current electricity fee, and current battery status; the output of the neural network is: battery storage power and load consumption power; The training of a neural network can be done by minimizing the following loss function: L=α(E battery_input -AND storage ) 2 +β(E load -P load ) 2 4 Among them, α and β are weight coefficients used to balance the accuracy of storage and consumption of electricity.

3. The multi-level energy management system according to claim 2, characterized in that: Local historical data includes load, photovoltaic power generation, and historical electricity charges, and the current electricity charge status includes charging power and discharging power.

4. The multi-level energy management system according to claim 2 or 3, characterized in that: In the data optimization unit, the steps of optimizing the power distribution model include: The total output power E at time t is recorded out (t) = P grid (t)+P PV (t), the amount of electricity stored at time t, E storage (t), the amount of electricity consumed at time t E load (t), the maximum capacity of the battery C battery ; The optimization target of power distribution is set as: Among them: γ1, γ2, γ3 are adjustment coefficients, which control the weights between various targets. battery is the maximum capacity of the battery, C load (t) is the user's load demand at time t, Cost(t) is the electricity cost at time t; And the distribution of electricity must meet the following constraints: The charge and discharge of the battery cannot exceed the capacity limit of the battery: 0≤E storage (t)≤C battery ; The amount of power stored and consumed should be determined by the power output and load demand: E storage (t)+E load (t)=E out (t); Battery loss: AND battery_input (t)=η battery ·AND storage (t)? During peak hours, the priority of electricity consumption is adjusted: E out (t)=min(P load (t),E out (t),C peak (t)>C offpeak (t); E load (t)=min(P load (t),E out (t)-E storange (t)),C peak (t)≤C offpeak (t); Electricity fee model: Assuming that the electricity fee is different during peak and off-peak periods, the electricity fee is defined as: Assuming that the distribution of electricity is adjusted dynamically in multiple time periods, at a certain time t, the objective function becomes: J(t)=min(γ1·(C battery -E storange (t))+Y2·E load (t)+Y3·Cost(t))。 5. The multi-level energy management system according to claim 4, characterized in that: The steps for optimizing the power distribution model also include: Define a state S(t) = [ Estorage(t) , E load (t)], at time t, the state is updated by selecting an action (storing or consuming power) and the corresponding cost is calculated; The recursive formula for dynamic programming is: Where a is the action of the battery, C(a, S(t)) is the immediate cost of taking the action in the current state, and γ is a discount factor used to weigh future costs.

6. The multi-level energy management system according to claim 4, characterized in that: The steps for optimizing the power distribution model also include: Use Q-learning to train the power allocation model. The core update formula of Q-learning is: Where: Q(S(t), a) is the expected return of taking action a in state S(t), R(t) is the immediate reward, which can be defined based on the cost and benefit of power allocation, and α is the learning rate, which controls the update step size.

Images