Target-driven energy supply and demand intelligent optimization system and method

Abstract

The invention discloses a target-driven energy supply and demand intelligent optimization system and method. The system comprises an input module, a multi-target quantification module, a decision module, an execution module and a feedback module. The input module receives an input high-level target instruction; the multi-target quantification module receives a target instruction to form a target quantification signal; the decision module is connected with the dynamic multi-target quantization module and is used for receiving the target quantization signal and outputting a strategy execution signal; the execution module is connected with the reinforcement learning intelligent decision-making module and is used for converting the strategy execution signal into a control instruction of the energy equipment; the feedback module collects energy system state data and external environment data in real time, and drives target weight adaptive adjustment and strategy iteration updating. According to the target-driven energy supply and demand intelligent optimization system and method, the problems that an existing energy optimization system is difficult to dynamically convert an abstract high-level target into a quantization parameter and cannot adapt to environmental changes to adjust the priority of the target can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent decision-making and dynamic optimization control of energy systems, and in particular to a target-driven intelligent optimization system and method for energy supply and demand. Background Art

[0002] Current energy system optimization primarily relies on mathematical programming methods with fixed objective functions (e.g., minimizing costs and carbon emissions), requiring pre-defined objective weights. With the increasing proportion of renewable energy and the diversification of user demands, energy management faces the challenge of dynamic objectives: policy adjustments may require prioritizing supply in the short term, extreme weather conditions may require a shift to a reliability-first mode, and personalized user objectives (e.g., self-consumption rate) require immediate response. Existing technologies have significant drawbacks: First, fixed objectives lead to rigid strategies and an inability to respond to sudden changes in directives. For example, traditional models require manual optimization after emergency peak-shaving instructions from the power grid, resulting in significant response delays. Second, they have poor environmental adaptability, with most systems employing static constraint boundaries. When wind and solar power output plummets or load changes suddenly, fixed constraints can easily lead to strategy failure or even equipment exceeding limits. Third, multi-objective coordination is difficult, and manually setting weights makes it difficult to balance the dynamic relationships between objectives. There is also a lack of intelligent arbitration mechanisms when economic efficiency and low-carbon efficiency conflict.

[0003] Reinforcement learning has been introduced to the energy sector in recent years, but existing approaches still have limitations: reward function design relies on expert experience and cannot automatically link to high-level objectives; environmental feedback is used only for policy updates and does not drive objective adjustments in reverse. Existing technologies attempt to use multi-objective optimization algorithms to handle weight allocation, but fail to address the issue of dynamic objective quantification; reinforcement learning is used for microgrid scheduling, but the reward function is fixed and decoupled from the environmental state. Fundamentally, existing technologies lack an intelligent mechanism to connect the entire chain of "abstract objectives-quantified parameters-dynamic weights-environmental responses," making it difficult to achieve true goal-driven optimization in energy systems. This patent addresses these gaps by proposing a framework for intelligent target transformation and closed-loop learning to address the core challenges of dynamic target response and environmental adaptation. Summary of the Invention

[0004] In view of the shortcomings of the above-mentioned prior art, the present invention aims to provide a goal-driven intelligent energy supply and demand optimization system and method to address the difficulties existing energy optimization systems face in dynamically converting abstract high-level goals into quantitative parameters and in adaptively adjusting goal priorities in response to environmental changes. This invention utilizes a multi-objective quantification module to parse high-level instructions into quantitative parameters, generating dynamic weights and constraints based on feedback data. A decision module constructs a reward function based on weights and generates policies through interactive learning within the environment. An execution module converts policies into device instructions. A feedback module collects environmental data in real time to drive weight adjustment and policy iteration, forming a closed loop of "goal quantification-policy generation-execution feedback."

[0005] The present invention provides a target-driven intelligent optimization system for energy supply and demand, comprising: Input module, the input module receives input high-level target instructions; The multi-objective quantization module receives target instructions, parses the received high-level target instructions into quantized target parameters, and generates dynamic priority weights and constraints based on real-time environmental data to form target quantization signals; The decision module is connected to the dynamic multi-objective quantification module and is used to receive the target quantification signal, build a reward function guided by the target priority based on the signal, generate an optimization strategy through interactive learning with the environment, and output a strategy execution signal; The execution module is connected to the reinforcement learning intelligent decision-making module and is used to convert the strategy execution signal into the control instruction of the energy equipment; Feedback module, the feedback module collects energy system status data and external environment data in real time, forms environmental status signals and synchronously transmits them to the dynamic multi-objective quantification module and reinforcement learning intelligent decision-making module, driving the adaptive adjustment of target weights and iterative updates of strategies.

[0006] In one embodiment of the present invention, the multi-objective quantization module has a built-in semantic parsing unit and a target mapping relationship library, which are used to decompose the natural language description in the high-level target instructions into atomic target elements; by matching the predefined element-parameter conversion rules in the target mapping relationship library, the atomic target elements are converted into mathematically processable quantitative target parameters; at the same time, based on the supply and demand fluctuation amplitude, renewable energy forecast deviation rate and key equipment status in the real-time environmental data transmitted by the feedback module, the adjustment coefficient of the dynamic priority weight is automatically generated; the final output target quantization signal includes the quantitative target parameters, dynamic priority weights and a set of constraints that evolve in real time with the environmental data.

[0007] In one embodiment of the present invention, the reward function constructed by the decision module takes the dynamic priority weight in the target quantization signal as the core control factor; the quantitative target parameter is mapped to the sub-item baseline value of the reward function, and the sub-item weight coefficient is allocated according to the dynamic priority weight; by continuously receiving the environmental state signal of the feedback module, the impact of the energy device action on the system state is simulated; the policy network parameters are iteratively updated based on the state-action pair reward value in the historical policy library, so that the generated optimization strategy dynamically adapts to the change of target priority.

[0008] In one embodiment of the present invention, the feedback module includes a multi-source data fusion unit, which synchronously collects the real-time value of the load power of the energy system, the charge state of the energy storage unit, the output of the distributed power supply, the grid dispatch instructions and the weather forecast data; generates an environmental status signal in a unified format through timestamp alignment and data cleaning; and transmits the signal in parallel to the multi-objective quantification module and the decision module, driving the former to automatically increase the power supply reliability weight coefficient according to the load mutation event, and at the same time triggering the latter to start the strategy relearning process.

[0009] In one embodiment of the present invention, the multi-objective quantification module is provided with a target conflict arbitration mechanism. When the high-level target instructions contain conflicting elements: based on the system risk level assessment results in the environmental state signal, the weight coefficient of the secondary target is reduced or its constraint boundary is relaxed; by calculating the difficulty index of each target element and the system impact factor, a compromise factor is generated and injected into the target quantification signal; the compromise factor is synchronously transmitted to the reward function construction unit of the decision module to balance the conflicting target tendencies of the strategy.

[0010] In one embodiment of the present invention, the decision module adopts a hierarchical decision architecture, including a global strategy unit and a regional strategy unit; the global strategy unit receives the target quantization signal to generate a long-term optimization direction, and the regional strategy unit refines the specific action instructions according to the environmental state signal; through the target consistency coordination algorithm, the output of the regional strategy unit meets the target constraint of the global strategy unit; when the feedback module detects a deviation from the local area target, it triggers the global strategy unit to send a weight correction instruction to the regional strategy unit.

[0011] In one embodiment of the present invention, the constraint condition generation unit of the multi-objective quantification module includes a safety boundary prediction model; the equipment operation limit threshold is dynamically calculated based on the equipment aging coefficient, line load rate and ambient temperature and humidity in the environmental status signal; the limit threshold is converted into the constraint boundary of the optimization model, and is tightened or relaxed in real time with feedback data; when the risk of boundary violation is predicted, a preventive constraint reinforcement mark is automatically injected into the target quantification signal.

[0012] In one embodiment of the present invention, the execution module includes an instruction translation unit and a priority arbitration unit; the instruction translation unit decodes the policy execution signal into the underlying control protocol instructions of the energy device; the priority arbitration unit sorts the execution sequence of multiple device instructions according to the dynamic weight in the target quantization signal; when the feedback module returns an instruction conflict alarm, the low-weight instructions are interrupted according to the target priority weight and the scheduling queue is regenerated.

[0013] In one embodiment of the present invention, the feedback module sets a data credibility assessment unit to mark abnormal data in the environmental status signal; if sensor failure or communication interruption is detected, the historical data interpolation mechanism is activated to generate alternative data; at the same time, the data reliability coefficient is sent to the multi-objective quantification module to trigger it to reduce the sensitivity of dynamic weight adjustment; and a strategy conservatism enhancement factor is injected into the decision module to limit the scope of exploration behavior during the learning process.

[0014] The present invention also provides a target-driven intelligent optimization method for energy supply and demand, comprising: S1: Receive input high-level target instructions; S2: Receives target instructions, parses the received high-level target instructions into quantitative target parameters, and generates dynamic priority weights and constraints based on real-time environmental data to form target quantitative signals; S3: Receives the target quantization signal, constructs a reward function guided by the target priority based on the signal, generates an optimization strategy through interactive learning with the environment, and outputs a strategy execution signal; S4: converting the strategy execution signal into the control instruction of the energy device; S5: Generates environmental state signals and transmits them synchronously to the dynamic multi-objective quantification module and reinforcement learning intelligent decision-making module to drive adaptive adjustment of target weights and iterative updates of strategies.

[0015] The present invention provides a goal-driven intelligent optimization system and method for energy supply and demand. This system uses a multi-objective quantification module to parse high-level instructions into quantitative parameters, and generates dynamic weights and constraints based on feedback data. A decision module constructs a reward function using weights and generates strategies through interactive learning with the environment. An execution module converts strategies into device instructions. A feedback module collects environmental data in real time to drive weight adjustment and strategy iteration, forming a closed loop of "goal quantification-strategy generation-execution feedback." BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 This is a system architecture diagram of a target-driven energy supply and demand intelligent optimization system; Figure 2 A schematic diagram showing the workflow of a target-driven intelligent optimization system for energy supply and demand; Figure 3 The present invention is a method flow chart of a target-driven intelligent optimization method for energy supply and demand. DETAILED DESCRIPTION

[0018] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0019] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. Therefore, the illustrations only show components related to the present invention and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the type, quantity, and proportion of each component may be changed arbitrarily, and the component layout may also be more complex.

[0020] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the embodiments of the present invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring the embodiments of the present invention.

[0021] See Figure 1-Figure 3 , shown is a target-driven energy supply and demand intelligent optimization system and method of the present invention. A target-driven energy supply and demand intelligent optimization system of the present invention includes an input module, a multi-target quantization module, a decision module, an execution module and a feedback module. The input module receives input high-level target instructions; the multi-target quantization module receives the target instructions, and is used to parse the received high-level target instructions into quantitative target parameters, and generate dynamic priority weights and constraints according to real-time environmental data to form a target quantization signal; the decision module is connected to the dynamic multi-target quantization module, and is used to receive the target quantization signal, and construct a reward function guided by the target priority based on the signal, and generates an optimization strategy through interactive learning with the environment, and outputs a strategy execution signal; the execution module is connected to the reinforcement learning intelligent decision module, and is used to convert the strategy execution signal into a control instruction for the energy equipment; the feedback module collects energy system status data and external environment data in real time, forms an environmental status signal and synchronously transmits it to the dynamic multi-target quantization module and the reinforcement learning intelligent decision module, driving the adaptive adjustment of the target weight and the iterative update of the strategy.

[0022] like Figure 1As shown in the figure, a target-driven intelligent optimization system for energy supply and demand consists of five core modules to form a closed-loop optimization architecture: the input module serves as the target instruction entrance, receiving high-level target instructions input by users or upper-level management systems, which are usually natural language descriptions (such as "prioritize power supply reliability" and "maximize renewable energy consumption") or structured target parameter packages; the multi-target quantification module is connected to the input module, and decomposes the high-level target instructions into atomic target elements through the built-in target parsing engine, and converts the elements into mathematically processable quantitative target parameters based on the predefined target mapping rule library (such as quantifying "power supply reliability" as a load loss probability threshold), while receiving the environmental status signal transmitted by the feedback module in real time (including load fluctuation data, wind and solar power forecast output, and equipment health status), generating priority weight coefficients and constraint condition sets through a dynamic weight algorithm, and finally outputting the target quantization signal - the signal contains a triple of quantization parameters, dynamic weights and constraint boundaries; the decision module receives the target quantization signal Then, a reward function with dynamic weight as the core is constructed (for example, when the economic weight is 0.7, the reward function focuses on the cost item), and an optimization strategy is generated through a reinforcement learning mechanism that simulates environmental interaction, and the strategy execution signal is output to the execution module; the execution module translates the strategy execution signal into the underlying control protocol of the energy equipment (such as energy storage charging and discharging power instructions, interruptible load switch instructions), and its instruction translation unit is adaptable to a variety of industrial communication protocols; the feedback module collects energy system status data (such as bus voltage, energy storage charge state, photovoltaic inverter output) and external environmental data (such as temperature, irradiance, electricity price signals) through a distributed sensor network, and forms a timestamp-synchronized environmental status signal after multi-source data fusion processing. This signal is simultaneously transmitted to the multi-objective quantification module and the decision module - the former adjusts the target weight coefficient accordingly (such as increasing the power supply reliability weight when a sudden load increase is detected), and the latter triggers the strategy relearning process, forming a closed-loop optimization of "target input-quantification analysis-strategy generation-execution feedback".

[0023] Furthermore, the multi-objective quantification module includes three innovative sub-mechanisms: the semantic parsing unit uses natural language processing technology to decompose the fuzzy instructions of "prioritizing the reduction of peak-valley differences" transmitted by the input module into target elements ("peak shaving", "valley filling", "response speed"), and matches the preset element-parameter conversion rules (such as "peak shaving" corresponds to the load peak reduction rate parameter) through the target mapping relationship library to realize the conversion of unstructured instructions into mathematical parameters; after the dynamic weight generation unit receives the environmental status signal of the feedback module, it calculates the weight adjustment coefficient based on the real-time supply and demand balance index, renewable energy forecast deviation rate and equipment risk level - for example, when the wind and solar forecast deviation rate exceeds the threshold, the system stability weight is automatically increased by 0.2-0.5 benchmark points, and the corresponding constraint conditions (such as the minimum backup capacity of energy storage) are generated; the target conflict arbitration mechanism detects the instruction When there is a conflict between the goals (e.g., "lowest cost" and "zero wind curtailment"), an arbitration process is initiated based on the system risk level in the environmental state signal. First, the difficulty index of achieving each goal is calculated (e.g., the additional energy storage cost required for zero wind curtailment), then the system impact factors of target relaxation are evaluated (e.g., the impact of reducing the wind curtailment rate by 5% on the grid), and finally a compromise factor is generated (e.g., economic weight × 0.8 + low-carbon weight × 1.2) and injected into the target quantization signal. The target quantization signal output by this module adopts a hierarchical coding structure: the first layer is the quantization parameter value range (e.g., the cost cap is ￥10,000 / day), the second layer is the dynamic weight matrix ([economic 0.6, reliability 0.3, low-carbon 0.1]), and the third layer is the constraint condition set (e.g., the PV inverter output is ≤80% of the rated power), which is transmitted to the reward function construction interface of the decision module via a high-speed data bus.

[0024] Specifically, the decision-making module implements goal-driven adaptive optimization through a three-layer architecture: after the reinforcement learning agent receives the target quantitative signal, its dynamic weight matrix is ​​mapped to the reward function - for example, when the reliability weight is 0.4, the power outage loss penalty coefficient in the reward function is increased to 1.4 times the baseline value; the agent simulates action execution based on the environmental state signal of the feedback module (such as increasing the energy storage discharge power), calculates the state-action value function through the policy network, selects the action sequence that maximizes the cumulative reward, and outputs the policy execution signal; the hierarchical decision unit includes a global policy engine and a regional policy engine: the global engine generates long-term optimization directions based on the target quantitative signal (such as a daily peak shaving target of 300kW), and the regional engine refines the action instructions based on the environmental state signal (such as 150kW energy storage discharge in area A + 150kW load reduction in area B); The target consistency coordination algorithm verifies deviations between regional policy outputs and global targets in real time. When the feedback module detects a local deviation (e.g., Area B only actually reduces power by 80 kW), it triggers the global engine to send weight correction instructions to the regional engine (e.g., increasing the economic penalty coefficient for Area B). The policy migration unit stores a library of historical optimization strategies for typical scenarios (e.g., typhoon scenarios, electricity price spikes). When the environmental state signal matches the characteristics of a historical scenario (e.g., load fluctuation >30%), it loads the corresponding strategy as the initial solution and calculates the migration correction using a target difference algorithm. For example, if the economic weight in the current target is 0.3 higher than in the historical scenario, the probability of selecting the high-cost action in the strategy is reduced. When the environmental change exceeds the historical database (e.g., when a new load is connected), it automatically switches to online reinforcement learning mode, using the ε-greedy algorithm to balance exploration and exploitation. The resulting policy execution signal contains a set of device control instructions and a timed execution sequence (e.g., [PCS charges 100 kW at t1, cuts off non-critical loads by 50 kW at t2]).

[0025] In one embodiment of the present invention, the target conflict arbitration mechanism of the multi-objective quantification module starts a hierarchical processing process when a high-level instruction contradiction is detected: when the input module transmits the conflicting instructions of "cost minimization" and "zero wind abandonment", the arbitration unit first analyzes the system risk level in the environmental status signal - if the wind and solar power forecast output fluctuation rate exceeds the threshold and the energy storage backup capacity is lower than the critical value, it is determined to be a high-risk scenario and the constraint relaxation strategy is automatically activated; then the difficulty index of achieving each target element is calculated: based on the historical data, the energy storage cost curve required to achieve the "zero wind abandonment" target and the wind abandonment rate tolerance range corresponding to "cost minimization" are fitted, and the quantitative index is generated by combining the real-time electricity price signal and the equipment depreciation coefficient; the system of target relaxation is simultaneously evaluated. System impact factors: The grid topology is used to analyze the amplifying effect of increased wind curtailment on local voltage deviations, or the attenuation gradient of increased costs on user satisfaction. Ultimately, a dynamic compromise factor is generated and injected into the target quantitative signal. For example, when the economic impact factor is 0.7 and the environmental impact factor is 1.3, a weighted combination of economic weight × 0.8 and environmental weight × 1.2 is output. This compromise factor is synchronously transmitted to the reward function reconstruction interface of the decision module, and the penalty term coefficient is adjusted to 1.2 times the baseline value to strengthen the environmental tendency. The arbitration mechanism has a built-in feedback calibration loop. When the execution module returns that the actual wind curtailment rate exceeds expectations, the basic allocation value of the environmental weight in the next cycle is automatically increased, forming a progressive balance of conflicting goals.

[0026] like Figure 2As shown, the hierarchical decision-making architecture of the decision module operates through a dual-engine collaborative operation. The global strategy unit receives the target quantification signal and converts its dynamic weight matrix into a long-term optimization direction. For example, when the power supply reliability weight reaches 0.6, a global constraint of "daily load shortfall rate ≤ 0.5%" is generated. The regional strategy unit decomposes the action instructions based on local parameters in the environmental state signal (such as the energy storage SOC value in area A and the actual photovoltaic output in area B). The regional optimization model calculates the specific value range of the energy storage discharge power in area A to compensate for the load shortfall in area B and verifies the contribution of the action combination to the global target. The target consistency coordination algorithm compares the deviation between the regional output and the global direction in real time. If the feedback module detects that the actual shortfall rate in area C reaches 0.8% (a super-global constraint), the deviation compensation mechanism is triggered. First, the target deviation index of each region is calculated (for example, the execution deviation of the economic weight in area C is +0.3). The global strategy unit then generates a weight correction instruction: the power supply reliability penalty coefficient in area C is increased to 1.5 times the original value and injected into the reward function of the regional strategy unit. The corrected regional strategy engine re-solves and outputs additional control instructions (such as starting the diesel generator set in area C). When multi-region coordination fails (e.g., communication in zone D is interrupted), the global unit initiates target redistribution: dynamically adjusts the responsibility weights of each region based on its adjustable resource margin (remaining energy storage capacity in zone E / sheddable load in zone F), and broadcasts new constraint parameters through the environmental status signal channel of the feedback module. The safety boundary prediction model of the multi-objective quantification module establishes a dynamic mapping of the equipment operating limits: after receiving the environmental status signal from the feedback module, the constraint generation unit extracts the equipment aging coefficient (e.g., transformer insulation loss rate), the real-time line load rate, and the ambient temperature and humidity parameters; calculates the instantaneous tolerance upper limit of the equipment through the thermal stability equation—for example, when the ambient temperature rises to the preset threshold and the load rate exceeds 75%, it is deduced that the transformer's safe output needs to be reduced by 15%; converts the calculation results into a set of constraint boundaries for the optimization model: generates a hard constraint of "main transformer load rate ≤ 85%" and a gradient constraint of "photovoltaic inverter temperature rise ≤ 60°C"; the boundary evolves in real time with the feedback data: if the line load rate exceeds the warning value for two consecutive hours, the constraint is automatically tightened to the original boundary 90%; when an over-limit risk is predicted (such as a sudden drop in wind speed causing a 30% drop in wind power output), the safety assessment submodule performs a pre-fault simulation: it simulates the power flow distribution under the condition of N-1 failure of key equipment. If insufficient backup capacity is detected, a preventive constraint reinforcement flag is generated. This flag is a three-level coded instruction: the first level identifies the risk type (such as power shortage), the second level specifies the affected equipment (such as the #3 energy storage PCS), and the third level defines the constraint reinforcement range (such as charging and discharging power limits of ±20% to ±15%). The reinforcement flag is transmitted to the policy screening layer of the decision module via a target quantization signal, forcibly eliminating action options that may cause over-limit (such as prohibiting the #3 energy storage from charging at full power), and simultaneously triggering the instruction pre-verification process of the execution module.

[0027] Furthermore, the instruction processing mechanism of the execution module includes a two-way guarantee design: after receiving the strategy execution signal from the decision module, the instruction translation unit parses the time series and spatial distribution parameters of the action instruction set (such as [t1, PCS#1 charging 80kW] / [t2, cutting off load #3]); matches the device communication protocol through the protocol adapter library: converts the unified strategy signal into energy storage control instructions in Modbus TCP format or load switch instructions in IEC 104 protocol; the priority arbitration unit sorts the instruction queue according to the dynamic weight matrix in the target quantization signal - when the economic weight reaches 0.8, cost-sensitive instructions (such as off-peak charging) are executed first; if the feedback module returns an instruction conflict alarm (such as "PCS charging" and "diesel engine starting" need to share the bus), the arbitration unit starts the conflict resolution algorithm: first calculate the target contribution of the conflicting instructions (charging instruction contribution economic index 0.7 vs. The diesel engine command contribution reliability index is 0.9, and low-weight commands are interrupted based on the system urgency level (e.g., voltage deviation rate) in the environmental status signal. Once the interrupted command enters the temporary queue, it triggers local strategy regeneration in the decision module: The alternative action (e.g., calling backup energy storage) is recalculated based on the updated environmental status signal. The new command is then inserted into the execution sequence after secondary arbitration. Status tracking during execution is implemented: The command achievement rate is verified using real-time data collected by the feedback module (e.g., if the actual charging power deviates from the command by more than 5%). Failure to meet the target triggers a weight compensation mechanism, which increases the priority weight coefficient of subsequent commands from the device by 0.1-0.3 basis points.

[0028] like Figure 3 The figure shows a target-driven intelligent optimization method for energy supply and demand of the present invention. S1: receiving input high-level target instructions; S2: receiving the target instructions, parsing the received high-level target instructions into quantitative target parameters, and generating dynamic priority weights and constraints according to real-time environmental data to form a target quantization signal; S3: receiving the target quantization signal, constructing a reward function guided by target priority based on the signal, generating an optimization strategy through interactive learning with the environment, and outputting a strategy execution signal; S4: converting the strategy execution signal into a control instruction for energy equipment; S5: forming an environmental state signal and synchronously transmitting it to the dynamic multi-target quantization module and the reinforcement learning intelligent decision-making module to drive the adaptive adjustment of the target weight and the iterative update of the strategy.

[0029] The present invention provides a goal-driven intelligent optimization system and method for energy supply and demand. A multi-objective quantification module parses high-level instructions into quantitative parameters, and generates dynamic weights and constraints based on feedback data. A decision module constructs a reward function using weights and generates strategies through interactive learning with the environment. An execution module converts strategies into device instructions. A feedback module collects environmental data in real time to drive weight adjustment and strategy iteration, forming a closed loop of "goal quantification-strategy generation-execution feedback."

[0030] Therefore, the goal-driven energy supply and demand intelligent optimization system and method of the present invention solves the problem that existing energy optimization systems have difficulty in dynamically converting abstract high-level goals into quantitative parameters and cannot adaptively adjust goal priorities according to environmental changes.

[0031] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

Claims

1. A goal-driven intelligent optimization system for energy supply and demand, characterized by: include: An input module, wherein the input module receives input high-level target instructions; a multi-objective quantization module, which receives the target instruction, parses the received high-level target instruction into quantized target parameters, and generates dynamic priority weights and constraints based on real-time environmental data to form a target quantization signal; a decision module, connected to the multi-objective quantification module, configured to receive the target quantification signal, construct a reward function guided by target priority based on the signal, generate an optimization strategy through interactive learning with the environment, and output a strategy execution signal; an execution module, connected to the decision module, and configured to convert the strategy execution signal into a control instruction for an energy device; The feedback module collects energy system status data and external environment data in real time, forms an environmental status signal and transmits it synchronously to the multi-objective quantification module and decision module, driving the adaptive adjustment of target weights and iterative update of strategies.

2. A target-driven energy supply and demand intelligent optimization system according to claim 1, characterized in that: The multi-objective quantization module has a built-in semantic parsing unit and a target mapping relationship library, which are used to decompose the natural language description in the high-level target instructions into atomic target elements; by matching the predefined element-parameter conversion rules in the target mapping relationship library, the atomic target elements are converted into mathematically processable quantitative target parameters; at the same time, based on the supply and demand fluctuation amplitude, renewable energy forecast deviation rate and key equipment status in the real-time environmental data transmitted by the feedback module, the adjustment coefficient of the dynamic priority weight is automatically generated; the final output target quantization signal includes the quantitative target parameters, dynamic priority weights and a set of constraint conditions that evolve in real time with the environmental data.

3. The target-driven energy supply and demand intelligent optimization system according to claim 1, characterized in that: The reward function constructed by the decision module uses the dynamic priority weight in the target quantization signal as the core control factor; the quantitative target parameters are mapped to the sub-item baseline values ​​of the reward function, and the sub-item weight coefficients are allocated according to the dynamic priority weights; by continuously receiving the environmental state signals of the feedback module, the impact of the energy device action on the system state is simulated; based on the state-action pair reward values ​​in the historical strategy library, the strategy network parameters are iteratively updated, so that the generated optimization strategy dynamically adapts to the changes in the target priority.

4. The target-driven energy supply and demand intelligent optimization system according to claim 1, characterized in that: The feedback module includes a multi-source data fusion unit, which synchronously collects the real-time value of the energy system's load power, the charge state of the energy storage unit, the output of distributed power sources, grid dispatch instructions and weather forecast data; generates an environmental status signal in a unified format through timestamp alignment and data cleaning; and transmits this signal in parallel to the multi-objective quantification module and the decision-making module, driving the former to automatically increase the power supply reliability weight coefficient based on load mutation events, while triggering the latter to start the strategy relearning process.

5. The target-driven energy supply and demand intelligent optimization system according to claim 1, characterized in that: The multi-objective quantification module is equipped with a target conflict arbitration mechanism. When high-level target instructions contain conflicting elements, the module chooses to reduce the weight coefficient of the secondary target or relax its constraint boundary based on the system risk level assessment results in the environmental state signal. By calculating the difficulty index of each target element and the system impact factor, a compromise factor is generated and injected into the target quantification signal. The compromise factor is synchronously transmitted to the reward function construction unit of the decision module to balance the conflicting target tendencies of the strategy.

6. The target-driven energy supply and demand intelligent optimization system according to claim 1, characterized in that: The decision module adopts a hierarchical decision-making architecture, including a global strategy unit and a regional strategy unit. The global strategy unit receives a target quantization signal to generate a long-term optimization direction, and the regional strategy unit refines specific action instructions based on the environmental state signal. Through a target consistency coordination algorithm, the output of the regional strategy unit satisfies the target constraints of the global strategy unit. When the feedback module detects a deviation from the local area target, it triggers the global strategy unit to send a weight correction instruction to the regional strategy unit.

7. The target-driven energy supply and demand intelligent optimization system according to claim 1, characterized in that: The constraint condition generation unit of the multi-objective quantification module includes a safety boundary prediction model; dynamically calculates the equipment operation limit threshold based on the equipment aging coefficient, line load rate and ambient temperature and humidity in the environmental status signal; converts the limit threshold into the constraint boundary of the optimization model and tightens or relaxes it in real time with feedback data; and automatically injects a preventive constraint reinforcement mark into the target quantification signal when a boundary violation risk is predicted.

8. The target-driven energy supply and demand intelligent optimization system according to claim 1, characterized in that: The execution module includes an instruction translation unit and a priority arbitration unit; the instruction translation unit decodes the policy execution signal into the underlying control protocol instructions of the energy device; the priority arbitration unit sorts the execution sequence of multiple device instructions according to the dynamic weight in the target quantization signal; when the feedback module returns an instruction conflict alarm, it interrupts the low-weight instructions according to the target priority weight and regenerates the scheduling queue.

9. The target-driven energy supply and demand intelligent optimization system according to claim 1, characterized in that: The feedback module sets up a data credibility assessment unit to mark abnormal data in the environmental status signal; if sensor failure or communication interruption is detected, the historical data interpolation mechanism is activated to generate alternative data; at the same time, the data reliability coefficient is sent to the multi-objective quantification module to trigger it to reduce the sensitivity of dynamic weight adjustment; and a strategy conservatism enhancement factor is injected into the decision module to limit the scope of exploration behavior during the learning process.

10. A target-driven intelligent optimization method for energy supply and demand according to any one of claims 1 to 9, characterized in that: include: S1: Receive input high-level target instructions; S2: receiving the target instruction, parsing the received high-level target instruction into quantitative target parameters, and generating dynamic priority weights and constraints according to real-time environmental data to form a target quantitative signal; S3: Receive the target quantization signal, construct a reward function guided by the target priority based on the signal, generate an optimization strategy through interactive learning with the environment, and output a strategy execution signal; S4: converting the strategy execution signal into a control instruction for the energy device; S5: Form an environmental state signal and synchronously transmit it to the multi-objective quantification module and decision module to drive the adaptive adjustment of the target weight and the iterative update of the strategy.

Images