Zero-carbon park energy management method and system based on AI and medium

Abstract

The invention provides an AI-based zero-carbon park energy management method and system, and a medium, and aims at achieving the purposes of efficient utilization and zero-carbon emission of energy in a park through an intelligent scheduling strategy. According to the method, energy consumption demand estimation, carbon emission optimization, multi-energy system management and other technologies are utilized to carry out real-time monitoring and optimal scheduling on various energy use conditions of the park. Specifically, the system collects the power use condition in the park, and performs energy consumption demand estimation and carbon emission calculation through an AI model. And then, according to a demand estimation result and a carbon emission optimization model, an optimal energy scheduling strategy is selected, real-time feedback is obtained after scheduling execution, and a decision model is further optimized. And finally, realizing collaborative optimization of an environmental protection target and economic benefits by adaptively adjusting a discount factor and a weight balance mechanism.

Description

Technical Field

[0001] The present invention relates to the field of energy management technology, and in particular to an AI-based zero-carbon park energy management system. Background Art

[0002] As the global climate change problem becomes increasingly serious, achieving zero carbon emissions has become an important goal pursued by various parks and industrial sites. The energy management system of traditional parks mainly relies on manual operations and fixed rules, which makes it difficult to adapt to complex and changing energy needs, and it is difficult to achieve coordinated optimization among multiple energy systems. Especially in the context of the gradual popularization of renewable energy, how to reasonably allocate and utilize a variety of energy resources (such as solar energy, wind energy, battery energy storage, etc.) while taking into account the economic benefits and environmental protection goals of the park has become a difficult problem that needs to be solved urgently.

[0003] In recent years, with the development of artificial intelligence (AI) technology, intelligent scheduling methods based on deep learning and reinforcement learning have gradually been applied to the field of energy management. Using AI technology, it is possible to predict energy demand within the park and dynamically optimize multi-energy systems. By introducing reinforcement learning, energy scheduling strategies can be automatically optimized based on historical data and real-time feedback, giving priority to low-carbon energy, minimizing carbon emissions while improving energy efficiency. However, most of the current energy management systems lack intelligent scheduling functions, making it difficult to achieve a dynamic balance between economic benefits and carbon emission targets, making it difficult to achieve an ideal compromise between economic and environmental goals. Summary of the invention

[0004] In order to achieve the above-mentioned object of the present invention, the present invention provides an AI-based zero-carbon park energy management system, comprising:

[0005] Data collection and monitoring module: collect and integrate raw data;

[0006] State, action and reward value module: adjust actions based on state and demand, with reward value as the target;

[0007] Deep learning module: uses the DQN algorithm and combines it with the gradient strategy for deep learning to balance immediate rewards and future rewards;

[0008] Intelligent energy scheduling module: Use the trained model to select the optimal action under the current state.

[0009] In a preferred embodiment of the present invention, the data acquisition and monitoring module includes:

[0010] For each sensor S in the park i , the data value measured at time t can be expressed as:

[0011] X i(t) = F i (V i (t))+∈ i (t);

[0012] V i (t) is the i-th sensor S i The collected power data, F i (V i (t)) is the measurement function that converts power data into data values, ∈ i (t) is the sensor S i The measurement error at time t;

[0013] Measurement data X i (t) The edge computing device of the data concentrator aggregates and compresses the data to obtain the compressed data

[0014]

[0015] represents the compressed data, N is the number of time steps used for averaging, and X i (t-kΔt) represents the measurement data of the i-th sensor in the period before time t. This period is determined by the product of multiple time steps and the sampling interval. k is the index of the time step, which is an integer from 0 to N-1. Δt is the sampling interval, and the smaller the time, the higher the frequency.

[0016] In a preferred embodiment of the present invention, the state, action and reward value module includes:

[0017] The compressed data Input to system state s t In the process, the estimated energy consumption demand, the estimated generation demand of renewable energy and the current state of the energy storage system are calculated to characterize the current state of energy management;

[0018]

[0019] in, Energy demand estimation is calculated by machine learning model based on historical energy consumption data and real-time environmental parameters. The power generation demand of renewable energy is estimated by edge computing through the time series analysis model. battery (t) is the current state of the energy storage system;

[0020] Action Set a t It represents a regulation strategy, which adjusts the system action according to the instruction, including charging and discharging and selecting power priority, which is expressed as:

[0021] at ∈{charging, discharging, using mains electricity, using renewable energy};

[0022] Introducing the reward value R t To quantify the effect of the current dispatch strategy, with the goal of minimizing carbon emissions and energy costs, it is defined as:

[0023] R t =-(α·CO2(t)+β·Energy Cost(t));

[0024] CO2(t) represents the carbon emissions at time t, and Energy Cost(t) represents the energy cost at time t; α and β are the weights of minimizing carbon emissions and minimizing energy costs, respectively, and are manually adjusted according to demand.

[0025] In a preferred embodiment of the present invention, the deep learning module includes:

[0026] Apply the state, action, and reward value to the deep learning model and update the parameters in combination with the DQN algorithm:

[0027]

[0028] Among them, Q(s t ,a t ) is the Q value, which is used to represent the state s t Next, perform action a t The expected total reward that can be obtained is used to measure the total reward in the future under the current state and action selection. The goal is to find an optimal Q-value function through training so that the action selected in any state can maximize the Q-value, thereby obtaining the optimal strategy;

[0029] R t is the reward value, which represents the immediate reward at time t, γ(s t ,a t ) is the discount factor, which is used to balance the weights of current rewards and future rewards. A larger γ(s t ,a t ) value will pay more attention to the future long-term returns, while a smaller γ(s t ,a t ) values ​​will focus more on immediate short-term returns;

[0030] It is the Q value of the action with the maximum Q value in the set of actions that can be selected in the next state. By taking the maximum value, the DQN algorithm can learn the optimal strategy, that is, to select the action with the largest Q value in each state, so as to obtain the maximum long-term reward;

[0031] In the policy gradient method, the discount factor is considered as part of the policy and the model is optimized by the gradient descent algorithm:

[0032]

[0033] γ(s t ,a t ) means in state s t and perform action a t The adaptive discount factor, γ base is the base discount factor, the default value of the discount factor without additional adjustment, η is the learning rate, is the gradient operation of the discount factor;

[0034] is the expected cumulative return starting from time t, represents the discounted cumulative return at each future time step, R t+k represents the immediate reward at time step t+k.

[0035] In a preferred embodiment of the present invention, the intelligent energy scheduling module includes:

[0036] Using the parameters optimized by the model, select the optimal action a through the optimal Q value t+1 ;

[0037] a t+1 = argmaxQ(s t ,a t );

[0038] Execute the selected actiona t+1 , control the energy storage system and energy use, argmaxQ(s t ,a t ) means to choose among all possible actions to make Q(s t ,a t ) The largest action, after the action is executed, the environment feedbacks the new state s t+1 and update the reward value R t+1 , and recorded in the experience pool for further training optimization.

[0039] The present invention also discloses a computer system, comprising:

[0040] a memory for storing processor-executable instructions;

[0041] Among them, the processor is configured to implement the AI-based zero-carbon campus energy management system when executing the executable instructions.

[0042] The present invention also discloses a computer-readable storage medium comprising:

[0043] a memory having a computer program stored thereon;

[0044] A processor is used to execute the program in the memory to implement the AI-based zero-carbon campus energy management system.

[0045] The present invention also discloses an AI-based zero-carbon park energy management method, including:

[0046] S1, collected and integrated raw data;

[0047] S2, adjusts actions based on the state and needs, with the reward value as the target;

[0048] S3, uses the DQN algorithm and combines it with the gradient strategy for deep learning to balance immediate rewards and future rewards;

[0049] S4, using the trained model, selects the optimal action in the current state.

[0050] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0051] The present invention significantly reduces the carbon emissions of the park and gradually achieves the environmental protection goal of a zero-carbon park by optimizing the energy scheduling strategy, giving priority to the use of low-carbon or zero-carbon energy (such as renewable energy) and reasonably controlling the use of battery energy storage systems.

[0052] Based on the energy consumption prediction and scheduling optimization of AI technology, the present invention can dynamically adjust the energy use strategy according to the real-time energy demand and environmental conditions of the park, reduce unnecessary energy waste, and thus significantly improve energy utilization efficiency.

[0053] The dynamic weight and adaptive discount factor adjustment mechanism is designed to maximize environmental benefits while ensuring the minimum economic benefits. Even when energy costs are high or environmental conditions are not ideal, the present invention can adaptively optimize scheduling to ensure the economic sustainability of the park.

[0054] By introducing reinforcement learning algorithms, the present invention realizes intelligent energy management. The system can automatically learn and optimize energy scheduling strategies, autonomously adapt to changes in energy demand and external environmental fluctuations in the park over time, reduce manual intervention, and realize automated intelligent energy management.

[0055] The AI-based real-time monitoring and prediction mechanism can quickly respond to changes in energy consumption within the park and fluctuations in external environmental factors (such as weather and market electricity prices), and adjust energy scheduling strategies in a timely manner to ensure that the system can still operate efficiently in a changing environment.

[0056] The energy management system is suitable for various smart parks, industrial parks and other places. It can effectively reduce carbon emissions and reduce operating costs. It is in line with the trend of sustainable development, has high economic value and social significance, and makes positive contributions to achieving the carbon neutrality goal.

[0057] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0059] Figure 1 It is a system schematic diagram of the present invention.

[0060] Figure 2 This is a schematic diagram of energy efficiency after intelligent scheduling. DETAILED DESCRIPTION

[0061] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0062] like Figure 1 As shown, the present invention provides an AI-based zero-carbon park energy management system, including:

[0063] 1. Data Collection and Monitoring

[0064] 1.1 IoT Sensor Network

[0065] IoT sensors are deployed in the park to monitor energy data in real time, mainly the usage of electricity.

[0066] Use sensors to collect and monitor the energy consumption data of the park for a long time. For each sensor S in the park i , the data value measured at time t can be expressed as:

[0067] X i (t) = F i (V i (t))+∈ i (t);

[0068] V i (t) is the i-th sensor S i The collected power data, F i (V i(t)) is the measurement function that converts power data into data values, ∈ i (t) is the sensor S i Measurement error at time t.

[0069] 1.2 Data Concentrator

[0070] Collect and integrate measurement data from various sensors, perform preliminary processing through edge computing devices, and reduce the pressure of data uploading.

[0071] The Schneider Electric TM171 series PLC (programmable logic controller) is used to control energy equipment and collect data in the park. It can receive data from different sensors and execute control logic.

[0072] Measurement data X i (t) The edge computing device of the data concentrator aggregates and compresses the data to obtain the compressed data

[0073]

[0074] represents the compressed data, N is the number of time steps used for averaging, and X i (t-kΔt) represents the measurement data of the i-th sensor in the period before time t. This period is determined by the product of multiple time steps and the sampling interval. k is the index of the time step, which is an integer from 0 to N-1. Δt is the sampling interval, and the smaller the time, the higher the frequency.

[0075] 2AI Model

[0076] Energy consumption demand estimation model: Use deep learning algorithm to establish an energy consumption demand estimation model to estimate the energy consumption demand of the park in different time periods.

[0077] 2.1.1 State, Action, and Reward Value

[0078] System Status t Includes current energy demand estimates, electricity demand estimates from renewable energy sources, and the current state of the energy storage system to characterize the current state of energy management;

[0079]

[0080] in, Energy demand estimation is calculated by machine learning model based on historical energy consumption data and real-time environmental parameters. The power generation demand of renewable energy is estimated by edge computing through the time series analysis model. battery(t) is the current state of the energy storage system.

[0081] 2.1.2 Action Set

[0082] Action Set a t Represents a regulation strategy, including charging and discharging and selecting power source priority, expressed as:

[0083] a t ∈{charging, discharging, using mains electricity, clean energy};

[0084] 2.1.3 Reward Value

[0085] Reward value R t With the goal of minimizing carbon emissions and energy costs, it is defined as:

[0086] R t =-(α·CO2(t)+β·Energy Cost(t));

[0087] CO2(t) represents the carbon emissions at time t, and Energy Cost(t) represents the energy cost at time t; α and β are the weights of minimizing carbon emissions and minimizing energy costs, respectively, and are manually adjusted according to demand.

[0088] 2.2 Reinforcement Learning Algorithms

[0089] 2.2.1 Deep Q-Learning

[0090] The Q value is updated through the DQN algorithm:

[0091]

[0092] Among them, Q(s t ,a t ) is the Q value, which is used to represent the state s t Next, perform action a t The expected total reward that can be obtained is used to measure the total reward in the future under the current state and action selection. The goal is to find an optimal Q-value function through training so that the action selected in any state can maximize the Q-value, thereby obtaining the optimal strategy;

[0093] R t is the reward value, which represents the immediate reward at time t, γ(s t ,a t ) is the discount factor, which is used to balance the weights of current rewards and future rewards. A larger γ(s t ,a t ) value will pay more attention to the future long-term returns, while a smaller γ(s t ,a t) values ​​will focus more on immediate short-term returns;

[0094] It is the Q value of the action with the maximum Q value in the set of selectable actions in the next state. By taking the maximum value, the DQN algorithm can learn the optimal strategy, that is, select the action with the largest Q value in each state, so as to obtain the maximum long-term reward.

[0095] 2.2.2 Hybrid Loss Function

[0096] In the policy gradient method, the discount factor is considered as part of the policy and optimized by the gradient descent algorithm:

[0097]

[0098] γ(s t ,a t ) means in state s t and perform action a t The adaptive discount factor, γ base is the base discount factor, the default value of the discount factor without additional adjustment, η is the learning rate, is the gradient operation of the discount factor;

[0099] is the expected cumulative return starting from time t, represents the discounted cumulative return at each future time step, R t+k represents the immediate reward at time step t+k.

[0100] When making strategic decisions, the system gives priority to the company's economic sustainability, that is, to ensure that the economic benefits obtained during the operation reach or exceed the minimum profit threshold. Only after this profit condition is met will more resources and weights be invested in the realization of environmental protection goals to ensure the minimization of carbon emissions and energy consumption. The zero-carbon emission goal is a sustainable development strategy that neither sacrifices the company's economic survival space nor continuously promotes the achievement of environmental protection goals. Through the adjustment of dynamic weights and discount factors, the system can flexibly balance short-term economic needs and long-term environmental protection goals in different scenarios, so that the company can obtain continuous economic benefits and competitive advantages in achieving the vision of zero carbon emissions.

[0101] 3. Optimize Scheduling Strategy

[0102] Intelligent energy dispatching platform: Based on energy demand estimation and carbon emission optimization results, intelligently dispatch energy usage within the park, such as giving priority to the use of renewable energy and battery energy storage systems.

[0103] Using the parameters optimized by the model, select the optimal action a through the optimal Q value t+1 ;

[0104] a t+1 = argmaxQ(s t ,a t );

[0105] Execute the selected actiona t+1 , control the energy storage system and energy use, argmaxQ(s t ,a t ) means to choose among all possible actions to make Q(s t ,a t ) The largest action, after the action is executed, the environment feedbacks the new state s t+1 and reward R t , and recorded in the experience pool for further training optimization.

[0106] 4. Experimental Data Analysis

[0107] like Figure 2 As shown, from the chart we can observe the percentage changes of three performance indicators "before optimization" and "after optimization": energy consumption, energy cost and carbon emissions.

[0108] 4.1 Energy consumption:

[0109] Before optimization: energy consumption is at a high level (100%).

[0110] After optimization: Energy consumption is significantly reduced, down to 46%.

[0111] The intelligent scheduling strategy achieved significant energy consumption reduction by optimizing the use period and priority of energy. This shows that the system effectively reduced unnecessary energy waste and improved energy efficiency.

[0112] 4.2 Energy costs:

[0113] Before optimization: high energy cost (close to 100%).

[0114] After optimization: Energy costs reduced by up to 80%.

[0115] Through intelligent scheduling, the system gives priority to the use of low-cost renewable energy and rationally arranges the charging and discharging of the energy storage system, significantly reducing energy procurement costs.

[0116] 4.3 Carbon emissions:

[0117] Before optimization: The carbon emission level is high (100%).

[0118] After optimization: Carbon emissions were reduced to about 67%.

[0119] The optimization strategy effectively reduces carbon emissions by giving priority to renewable energy and reducing reliance on traditional energy. This shows that the AI ​​scheduling strategy helps reduce the carbon footprint of the park and move towards the zero-carbon goal.

[0120] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. An AI-based zero-carbon park energy management system, characterized in that: include: Data collection and monitoring module: collect and integrate raw data; State, action and reward value module: adjust actions based on state and demand, with reward value as the target; Deep learning module: uses the DQN algorithm and combines it with the gradient strategy for deep learning to balance immediate rewards and future rewards; Intelligent energy scheduling module: Use the trained model to select the optimal action under the current state.

2. The AI-based zero-carbon campus energy management system according to claim 1 is characterized in that: The data acquisition and monitoring module includes: For each sensor S in the park i , the data value measured at time t can be expressed as: X i (t)=F i (V i (t))+∈ i (t); V i (t) is the i-th sensor S i The collected power data, F i (V i (t)) is the measurement function that converts power data into data values, ∈ i (t) is the sensor S i The measurement error at time t; Measurement data X i (t) The edge computing device of the data concentrator aggregates and compresses the data to obtain the compressed data represents the compressed data, N is the number of time steps used for averaging, and X i (t-kΔt) represents the measurement data of the i-th sensor in the period before time t. This period is determined by the product of multiple time steps and the sampling interval. k is the index of the time step, which is an integer from 0 to N-1. Δt is the sampling interval, and the smaller the time, the higher the frequency.

3. The AI-based zero-carbon campus energy management system according to claim 1 is characterized in that: The state, action and reward value modules include: The compressed data Input to system state s t In the process, the estimated energy consumption demand, the estimated generation demand of renewable energy and the current state of the energy storage system are calculated to characterize the current state of energy management; in, Energy demand estimation is calculated by machine learning model based on historical energy consumption data and real-time environmental parameters. The power generation demand of renewable energy is estimated by edge computing through the time series analysis model. battery (t) is the current state of the energy storage system; Action Set a t It represents a regulation strategy, which adjusts the system action according to the instruction, including charging and discharging and selecting power priority, which is expressed as: a t ∈{charging, discharging, using mains electricity, using renewable energy}; Introducing the reward value R t To quantify the effect of the current dispatch strategy, with the goal of minimizing carbon emissions and energy costs, it is defined as: R t =-(α·CO2(t)+β·Energy Cost(t)); CO2(t) represents the carbon emissions at time t, and Energy Cost(t) represents the energy cost at time t; α and β are the weights of minimizing carbon emissions and minimizing energy costs, respectively, and are manually adjusted according to demand.

4. The AI-based zero-carbon campus energy management system according to claim 1 is characterized in that: The deep learning module includes: Apply the state, action, and reward value to the deep learning model and update the parameters in combination with the DQN algorithm: Among them, Q(s t ,a t ) is the Q value, which is used to represent the state s t Next, perform action a t The expected total reward that can be obtained is used to measure the total reward in the future under the current state and action selection. The goal is to find an optimal Q-value function through training so that the action selected in any state can maximize the Q-value, thereby obtaining the optimal strategy; R t is the reward value, which represents the immediate reward at time t, γ(s t ,a t ) is the discount factor, which is used to balance the weights of current rewards and future rewards. A larger γ(s t ,a t ) value will pay more attention to the future long-term returns, while a smaller γ(s t ,a t ) values ​​will focus more on immediate short-term returns; It is the Q value of the action with the maximum Q value in the set of actions that can be selected in the next state. By taking the maximum value, the DQN algorithm can learn the optimal strategy, that is, to select the action with the largest Q value in each state, so as to obtain the maximum long-term reward; In the policy gradient method, the discount factor is considered as part of the policy and the model is optimized by the gradient descent algorithm: γ(s t ,a t ) means in state s t and perform action a t The adaptive discount factor γ base is the base discount factor, the default value of the discount factor without additional adjustment, η is the learning rate, ▽ γ is the gradient operation of the discount factor; is the expected cumulative return starting from time t, represents the discounted cumulative return at each future time step, R t+k represents the immediate reward at time step t+k.

5. The AI-based zero-carbon campus energy management system according to claim 1 is characterized in that: The intelligent energy scheduling module includes: Using the parameters optimized by the model, select the optimal action a through the optimal Q value t+1 ; a t+1 =argmaxQ(s t ,a t ); Execute the selected actiona t+1 , control the energy storage system and energy use, argmaxQ(s t ,a t ) means to choose among all possible actions to make Q(s t ,a t ) The largest action, after the action is executed, the environment feedbacks the new state s t+1 and the new reward value R t+1 , and recorded in the experience pool for further training optimization.

6. A computer system, characterized in that: include: a memory for storing processor-executable instructions; Wherein, the processor is configured to implement an AI-based zero-carbon campus energy management system as described in any one of claims 1 to 5 when executing the executable instructions.

7. A computer-readable storage medium, characterized in that: include: a memory having a computer program stored thereon; A processor, used to execute the program in the memory to implement an AI-based zero-carbon campus energy management system as described in any one of claims 1 to 5.

8. A zero-carbon park energy management method based on AI, characterized in that: include: S1, collected and integrated raw data; S2, adjusts actions based on the state and needs, with the reward value as the target; S3, uses the DQN algorithm and combines it with the gradient strategy for deep learning to balance immediate rewards and future rewards; S4, using the trained model, selects the optimal action in the current state.

Images