Industrial park comprehensive energy management method and system based on hierarchical system

Abstract

The invention discloses an industrial park comprehensive energy management method and system based on a hierarchical system, and belongs to the technical field of energy management and automatic control. The method comprises the following steps: integrating external data, and predicting load and renewable energy power generation power by using a machine learning algorithm; making a start-stop and output plan of each energy device in the park through an MILP model; and based on short-term prediction and an intra-day scheduling plan, establishing an MPC model, solving a real-time control instruction of each energy device, and controlling operation of various energy production, conversion and storage devices. According to the method, an algorithm combining reinforcement learning and model prediction control is creatively applied, the prediction model and the cost function of MPC are adaptively adjusted on line through reinforcement learning, the equipment power is dynamically adjusted in combination with short-term rolling prediction, uncertainty is effectively dealt with, more accurate second-level power balance is achieved, and the reliability of the system is improved. And the adaptability and the operation economy of the system to high-proportion renewable energy sources are effectively improved through hierarchical collaborative optimization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy management and automatic control, in particular to an industrial park comprehensive energy management method and system based on a hierarchical and recursive system. BACKGROUND

[0002] As a major distributed renewable energy, photovoltaic power generation is increasingly widely used in parks. However, the intermittency and randomness of photovoltaic output bring great challenges to the safe, stable and economic operation of the park energy system. In order to ensure power supply reliability, the park usually configures diesel generators as backup power and introduces energy storage systems to smooth power fluctuations. At present, the "light storage-diesel" system composed of photovoltaic, energy storage and diesel generators is a common energy supply mode in industrial parks.

[0003] The existing industrial park energy management system has obvious deficiencies in dealing with high proportion of renewable energy access. Most systems use single-time-scale control strategies based on rules or simple optimization, such as only performing day-ahead scheduling oriented to economy or only performing real-time control targeting stability. This fragmented operation mode is difficult to balance the economy of system operation and the accuracy of real-time control. Specifically, the day-ahead scheduling plan is based on long-term prediction, but the prediction error will accumulate over time, resulting in deviation between actual operation and plan; and the real-time control layer may deviate from the optimal economic operation trajectory and increase the total operation cost of the system if it lacks coordination with the upper layer scheduling to maintain instantaneous balance. In addition, the introduction of multi-element energy storage technologies such as hydrogen energy increases the complexity of the system, and traditional optimization methods are difficult to accurately model and coordinate the multi-energy conversion process.

[0004] Therefore, the current industrial park comprehensive energy system urgently needs an efficient energy management technology that can effectively coordinate different time scales, connect different optimization targets, and adaptively respond to wind and light load uncertainty. SUMMARY

[0005] The present application relates to the technical field of energy management and automatic control, in particular to an industrial park comprehensive energy management method and system based on a hierarchical and recursive system.

[0006] Technical solution: One aspect of the present application provides an industrial park comprehensive energy management method based on a hierarchical and recursive system, comprising:

[0007] Prediction and perception step: integrate external data and use machine learning algorithms to predict load and renewable energy generation power;

[0008] Intraday scheduling step: based on long-term prediction results, formulate the start-stop and output plan of each energy device in the park through a MILP model;

[0009] Real-time control step: based on short-term prediction and intraday scheduling plan, establish MPC model, online solve real-time control instruction of each energy device;

[0010] Execution step: according to the real-time control instruction, control the operation of various energy production, conversion and storage devices;

[0011] The method further comprises feeding back the operation data of each energy device to the prediction and perception step and / or the real-time control step to form a closed-loop control.

[0012] Further, the intraday scheduling step comprises:

[0013] Establish a MILP model, which takes the long-term prediction result as input, takes the minimum total operation cost of the system as the first objective function, and outputs the first optimization relationship of device start-stop and output plan; Wherein the expression of the first objective function is:

[0014] ,

[0015] Wherein, represents the total number of time periods of the scheduling period, is the cost or benefit generated by the power exchange with the power grid; is the total cost of the diesel generator g; is the operation and maintenance cost of the energy storage system;

[0016] The constraint condition of the objective function is constructed, including device physical operation constraint and system power balance constraint, wherein the expression of the physical operation constraint is:

[0017] ,

[0018] Wherein, represents the actual output power of the th diesel generator in the time period , represents the minimum technical output of the th diesel generator, represents the rated power of the th diesel generator, represents the start-stop state of the th diesel generator in the time period , represents the maximum downward ramping rate of the th diesel generator, represents the charging power of the energy storage system in the time period , represents the maximum allowed charging power of the energy storage system, represents the discharging power of the energy storage system in the time period , represents the maximum allowed discharging power of the energy storage system, represents the state of charge of the energy storage system in the time period , represents the charging efficiency of the energy storage system, represents the discharging efficiency of the energy storage system, represents the length of each dispatching time period, represents the rated capacity of the energy storage system, represents the lower safety limit of the state of charge of the energy storage system, represents the upper safety limit of the state of charge of the energy storage system, respectively, represents the power exchanged with the power grid by the system in the time period , represents the maximum power allowed to be sold to the power grid, represents the maximum power allowed to be purchased from the power grid;

[0019] The expression of the system power balance constraint is:

[0020] ,

[0021] wherein, represents the total number of diesel generators in the system, represents the predicted photovoltaic output in the time period , represents the predicted load demand in the time period ;

[0022] The MILP model is solved by a branch and bound algorithm to generate an intraday dispatching plan, which includes the optimal start-stop plan , the output plan , the optimal charging-discharging plan of the energy storage system , and the optimal exchange power plan of the power grid . The four plans are set as the set values and physical constraints are issued to the real-time control layer.

[0023] Further, the real-time control step includes:

[0024] The MPC model takes short-term prediction data, the upper-layer intraday dispatching plan and constraints as inputs, minimizes the deviation between real-time power and planned power as the second objective function, and outputs the second optimization relationship of the device real-time power instruction; wherein the expression of the second objective function is:

[0025] ,

[0026] wherein, is the power of the system at time quantified risk cost of deviation; a quantified indicator of the execution force of the plan, whose value directly reflects how high the risk of battery out-of-control in the future will be caused by the current formulated charging and discharging plan; the total economic cost and equipment wear cost caused by the execution of the generator and energy storage power regulation actions in the plan; a quantified indicator of the execution force of the plan, whose value directly reflects how high the risk of battery out-of-control in the future will be caused by the current formulated charging and discharging plan; quantify and penalize the non-smoothness or volatility of the generator active power output plan;

[0027] solving the MPC model by using a rolling optimization and feedback correction algorithm to obtain an optimal control instruction sequence for future time steps, wherein the control instructions include diesel generator power instructions, lithium battery energy storage system power instructions, hydrogen energy storage system power instructions, and fine-tuning instructions for grid exchange power.

[0028] Further, the process of solving the MILP model by using a branch and bound algorithm includes:

[0029] initializing a search tree, and the root node is a linear programming relaxation of the original MILP problem;

[0030] relaxing the integer constraints in the MILP model to form a linear programming problem and solve it to obtain a relaxation solution and a cost lower bound;

[0031] if the relaxation solution does not satisfy the integer constraints, select a fractional integer variable for branching to generate two branch sub-problems;

[0032] the first sub-problem: forcibly set =0, i.e., the generator is shut down at this time;

[0033] the second sub-problem: forcibly set =1, i.e., the generator is started at this time;

[0034] for each branch sub-problem, recursively solve, bound, and use the cost upper bound of the current optimal integer solution to prune sub-problems that are unlikely to produce better solutions;

[0035] when the search is completed or the termination condition is met, output the optimal integer solution as the daily scheduling plan.

[0036] Further, the execution step includes:

[0037] issuing the first instruction in the optimal control instruction sequence to each energy device, and each energy device operates according to the instruction.

[0038] Further, the prediction and perception step includes:

[0039] By acquiring historical operating data of various energy devices, future weather forecast data, and dynamic market electricity price data, machine learning models are used to predict long-term photovoltaic output, load, and market electricity price.

[0040] Furthermore, the steps for optimizing the MPC model using the rolling optimization and feedback correction algorithm include:

[0041] The second objective function and the physical constraints of the system are input into the embedded optimization solver for calculation. The actual operating state measurement at the current moment is used as the initial state. Combined with the set value and short-term prediction data, a finite-time open-loop optimization problem is solved to obtain the optimal control command sequence in the future control time domain.

[0042] Extract the first control command from the optimal control command sequence and issue it to the execution device;

[0043] At the next moment, based on the new actual operating status measurements of the system, the optimization is repeated.

[0044] Another aspect of the present invention provides an integrated energy management system for industrial parks based on a hierarchical system, comprising:

[0045] The prediction and perception layer, based on acquired historical operating data and external data, performs long-term photovoltaic output forecasting, load forecasting, and market electricity price forecasting, providing data support for the intraday dispatch layer;

[0046] The intraday dispatch layer calculates the diesel generator start-up and shutdown and energy storage plans based on long-term photovoltaic output, load and market electricity price, and sends the optimized setpoints and constraints to the real-time control layer.

[0047] The real-time control layer, based on short-term forecast data and intraday scheduling instructions, solves the real-time power control signals of diesel generators and energy storage online and sends them to the execution and equipment layer.

[0048] The execution and equipment layer, including photovoltaic arrays, diesel generators, lithium battery energy storage and hydrogen energy storage systems, receives and executes instructions issued by the real-time control layer;

[0049] The data feedback closed-loop system is used to monitor the operating status data of each energy device in real time and upload the operating status data to the intraday scheduling layer and the real-time control layer to form a closed-loop optimization.

[0050] Beneficial effects: Compared with the prior art, the significant advantages of this invention are:

[0051] 1. Significantly improved economic benefits: This invention, through the constructed MILP optimization model, achieves optimal economic planning for the entire lifecycle of diesel generator start-up and shutdown, energy storage charging and discharging, and energy exchange with the power grid, significantly reducing the total operating cost of the system;

[0052] 2. Enhanced real-time control accuracy and adaptive capability: This invention constructs an MPC model, combining the advantages of rolling optimization and feedback correction of model predictive control with the online learning capability of reinforcement learning. This enables the system to dynamically adapt to disturbances such as load fluctuations and uncertainties in photovoltaic output, achieving more accurate tracking of intraday scheduling plans, reducing the deviation between plans and actual results, and thus achieving more refined power balance on a second / minute time scale, thereby improving power quality.

[0053] 3. Improved system reliability and resilience: This invention utilizes a multi-timescale hierarchical collaborative design and closed-loop feedback mechanism, enabling upper-level planning to guide lower-level control, while the execution effect of the lower level can provide feedback to correct the upper-level plan, forming an organic whole; at the same time, it integrates risk management and emergency mechanisms, which can prioritize ensuring power supply reliability when a serious power deviation is detected, thereby enhancing the system's ability to cope with emergencies.

[0054] 4. Optimized utilization of diversified energy storage, especially hydrogen energy storage: This invention models the hydrogen energy storage system as an integrated unit, clarifies its internal energy conversion constraints, and comprehensively considers the rapid response of lithium batteries and the long-term, large-scale energy storage advantages of hydrogen energy storage in multi-timescale optimization. This realizes the complementary advantages between different energy storage technologies, improves the system's ability to absorb renewable energy, and is in line with the national energy strategy.

[0055] 5. Strong engineering applicability: The invention has a clear hierarchy, and both the algorithm and model have clear implementation paths. The intraday scheduling layer adopts the MILP solution algorithm, which ensures the global optimality or near-optimality of the optimization results. The MPC framework of the real-time control layer is easy to implement rolling optimization on industrial controllers and has good prospects for engineering application. Attached Figure Description

[0056] Figure 1 A flowchart illustrating a hierarchical, layered approach to integrated energy management in industrial parks.

[0057] Figure 2 This is a flowchart illustrating the process of solving a MILP model using the branch and bound algorithm.

[0058] Figure 3 This is a schematic diagram illustrating the principle of rolling optimization using model predictive control. Detailed Implementation

[0059] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and not intended to limit the scope of the invention. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the embodiments of the present invention, and not all structures.

[0060] In the following description, specific details such as target system architecture and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of this application with unnecessary detail.

[0061] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0062] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0063] Furthermore, in the description of this application and the appended claims, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0064] References to "one embodiment" or "some embodiments" in this specification mean that one or more embodiments of this application include the target features, structures, or characteristics described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized.

[0065] Example 1

[0066] The flowchart of the integrated energy management method for industrial parks based on a hierarchical system described in this embodiment is as follows: Figure 1 As shown, the method includes the following steps:

[0067] S1: Prediction and Sensing Steps: Integrate external data and use machine learning algorithms to predict load and renewable energy generation capacity;

[0068] S2: Intraday scheduling steps: Based on long-term forecast results, the start-up, shutdown and power output plans of various energy equipment in the park are formulated through the MILP model;

[0069] S3: Real-time control steps: Based on short-term forecasts and intraday scheduling plans, establish an MPC model and solve for the real-time control commands of each energy device online;

[0070] S4: Execution steps: Control the operation of various energy production, conversion and storage devices according to real-time control instructions;

[0071] S5: The method further includes feeding back the operating data of each energy device to the prediction and sensing step and / or the real-time control step to form a closed-loop control.

[0072] Furthermore, the prediction and perception steps include:

[0073] By acquiring historical operating data of various energy devices, future weather forecast data, and dynamic market electricity price data, machine learning models are used to predict long-term photovoltaic output, load, and market electricity price.

[0074] Operational history is obtained through data loggers and smart meters; external information is acquired through meteorological sensors to provide future weather forecasts; and dynamic market electricity prices are provided through data interfaces. This data provides information support for subsequent optimization decisions.

[0075] Historical photovoltaic power output, corresponding historical meteorological data, and temporal characteristics are used as a training set and input into a selected machine learning model for training. In this example, no specific restrictions are placed on the machine learning model. After receiving future weather forecast data, the trained machine model outputs a long-term photovoltaic power output prediction. The model learns load variation patterns from historical data and combines this with future calendars, production plans, and weather forecasts to generate load predictions. Market electricity price .

[0076] Furthermore, the intraday scheduling steps include:

[0077] A MILP model is established, which takes long-term forecast results as input, minimizes the total system operating cost as the first objective function, and outputs the first optimal relationship between equipment start-up / shutdown and power output plans; the expression for the first objective function is:

[0078] ,

[0079] in, This indicates the total number of time periods in the scheduling cycle. The cost or benefit of exchanging power with the power grid; The total cost of the diesel generator g; For the operation and maintenance costs of energy storage systems;

[0080] The constraints for the objective function are constructed, including physical operation constraints of the equipment and system power balance constraints, wherein the expression for the physical operation constraints is:

[0081] ,

[0082] in, Indicates the first diesel generator during the period Actual output power Indicates the first The minimum technical output of a diesel generator. Indicates the first The rated power of the diesel generator. Indicates the first diesel generator during the period Start-stop status, Indicates the first The maximum downhill climbing rate of the diesel generator. Indicates the time period of the energy storage system The charging power, This indicates the maximum allowable charging power of the energy storage system. Indicates the time period of the energy storage system The discharge power, This indicates the maximum allowable discharge power of the energy storage system. Indicates the time period of the energy storage system The state of charge, This indicates the charging efficiency of the energy storage system. This indicates the discharge efficiency of the energy storage system. This indicates the duration of each scheduling period. Indicates the rated capacity of the energy storage system. The safe lower limit of the state of charge of an energy storage system. These represent the safe upper limit of the state of charge of the energy storage system. Indicates the system during the time period Power exchanged with the power grid This indicates the maximum power that is allowed to be sold to the grid. This indicates the maximum power that is permitted to be purchased from the grid;

[0083] The expression for the system power balance constraint is:

[0084] ,

[0085] in, This indicates the total number of diesel generators in the system. Indicates time period Predicted photovoltaic output Indicates time period The predicted load demand is represented by;

[0086] The MILP model is optimized using a branch and bound algorithm to generate an intraday scheduling plan, which includes the optimal start-stop plan for the diesel generators. , contribution plan Optimal charge and discharge plan for energy storage system Optimal switching power plan for the power grid The four plans are set as parameters and physical constraints are sent to the real-time control layer.

[0087] Combination Figure 2 As shown, the process of solving the MILP model using the branch and bound algorithm further includes:

[0088] S21: Initialization:

[0089] Initialize a search tree with the root node being a linear programming relaxation of the original MILP problem.

[0090] During initialization, all integer constraints are temporarily ignored;

[0091] S22: Relaxation Solution and Boundary Calculation:

[0092] Relaxing the integer constraints in the MILP model to form a linear programming problem and solving it yields the relaxed solution and the lower bound of the cost.

[0093] Solve the linear programming relaxation problem corresponding to the current node, temporarily ignoring all integer constraints. This relaxed solution provides a lower bound on the global optimal cost of the original MILP problem. Although some key variables in the solution include: diesel generator start / stop commands... Energy storage charging and discharging While it may be fractional and lacks physical operability, it quickly assesses the theoretical minimum possible value of the total system operating cost over the entire scheduling cycle. At each step, all device power constraints, energy storage SOC constraints, and system power balance constraints are rigorously verified to ensure that the search direction remains within the feasible region of the problem.

[0094] S23: Branch:

[0095] If the relaxed solution does not satisfy the integer constraint, then a fractional integer variable is chosen for branching, for a certain time step. Diesel generator start-stop variable This generates two branching subproblems:

[0096] First sub-problem: Enforcement order =0, meaning the generator stops at that moment;

[0097] Second sub-problem: Enforcement order =1, meaning the generator is started at that moment.

[0098] This step systematically enumerates all possible start-stop combinations of the diesel generator at different points in time, as well as the charging and discharging states of the energy storage system. Essentially, it intelligently traverses all possible operating conditions of the system, which is key to the core decision of accurately optimizing the unit's start-stop scheme.

[0099] S24: Boundary Delineation and Pruning:

[0100] For each subproblem, solve and bound recursively, and prune subproblems that cannot produce better solutions by using the upper bound of the cost of the current best integer solution.

[0101] Infeasible pruning: If the relaxation problem of a certain branch violates the system's safe operation constraints, then all solutions under that branch are deemed infeasible and are pruned.

[0102] Boundary pruning: If the cost of a relaxed solution in a certain branch is higher than the total running cost of the best feasible integer solution found so far, it means that continuing to search for that branch will not yield a more economical solution, and therefore it is pruned.

[0103] Integer solution update: When a branch's relaxation solution has exactly all integer variables, it constitutes a feasible solution such as an explicit start-stop plan and power allocation. If its total cost is lower, it is updated to the new optimal solution.

[0104] S25: Iteration and Convergence:

[0105] Perform the above steps S22 to S24 in a loop. When the search is completed or the termination condition is met, such as when convergence is achieved or the maximum number of iterations is reached, output the optimal integer solution as the intraday scheduling plan.

[0106] The optimal integer solution is the globally optimal strategy that minimizes the total operating cost throughout the entire scheduling cycle while strictly satisfying all physical and safety constraints. This strategy explicitly provides the optimal start-stop sequence for the diesel generator, the optimal charging and discharging plan for the energy storage system, and the energy interaction plan with the power grid, perfectly balancing the system's economy and safety.

[0107] Furthermore, the real-time control steps include:

[0108] The MPC model takes short-term forecast data, upper-level intraday scheduling plans, and constraints as inputs, minimizes the deviation between real-time power and planned power as the second objective function, and outputs a second optimization relationship for the real-time power command of the device; the expression for the second objective function is:

[0109] ,

[0110] in, For at any time The quantification of the risk cost of deviation; For at any time The sum of the total economic cost and equipment depreciation cost resulting from the generator and energy storage power regulation actions in the execution plan; As a quantitative indicator of the plan's execution capability, its value directly reflects how high the risk of the battery running out of control in the future will be due to the current charging and discharging plan; Quantify and penalize the unevenness or severe fluctuations in the generator's active power output plan;

[0111] The MPC model is optimized and solved using rolling optimization and feedback correction algorithms to obtain the optimal control command sequence for future time steps. The control commands include diesel generator power commands, lithium battery energy storage system power commands, hydrogen energy storage system power commands, and fine-tuning commands for grid switching power.

[0112] Furthermore, the steps for optimizing the MPC model using the rolling optimization and feedback correction algorithm include:

[0113] The second objective function and the physical constraints of the system are input into the embedded optimization solver for calculation. The actual operating state measurement at the current moment is used as the initial state. Combined with the set value and short-term prediction data, a finite-time open-loop optimization problem is solved to obtain the optimal control command sequence in the future control time domain.

[0114] Extract the first control command from the optimal control command sequence and issue it to the execution device;

[0115] At the next moment, based on the new actual operating status measurements of the system, the optimization is repeated.

[0116] In the real-time control phase, key real-time data is read through smart meters and sensors, including: the actual output power of the diesel generator, the actual charging and discharging power and remaining capacity (SOC) of the energy storage system (lithium battery / hydrogen storage), the real-time power of key load points, and the actual output of the photovoltaic array. Setpoints and constraints are accepted: based on the "planning blueprint" for a future period output from the intraday scheduling phase, this includes the planned output curve of the diesel generator, the planned SOC trajectory of the energy storage system, and the planned values ​​and upper and lower limits of power exchanged with the grid. These constitute the tracking targets and safety boundaries for real-time control.

[0117] The planned output curve of the diesel generator is a stepped or piecewise constant curve, based on the output plan selected according to the start-stop status.

[0118] The planned SOC trajectory of the energy storage system is derived through the integral optimal charge and discharge plan.

[0119] Planned values ​​and upper and lower limits of power exchanged with the power grid: Planned values Optimal switching power plan derived directly from optimized output Positive values ​​represent electricity purchases, and negative values ​​represent electricity sales; the upper and lower limits are inherent physical constraints or contractual stipulations of the system. The intraday scheduling steps have already complied with these constraints during the solution process, so when outputting the plan, these constraints will be sent as upper and lower limits to the real-time control layer, informing it of the legal fluctuation range of power interaction with the grid.

[0120] Short-term forecast data refers to more accurate forecasts of photovoltaic power output fluctuations and load fluctuations within the next few minutes, based on the latest ultra-short-term meteorological data.

[0121] The optimization problem of the second objective function is solved using an industrial computer or high-performance PLC within the system. This problem is not for the current instant, but for a finite time domain in the future. The optimization objective is:

[0122] (a) Economy and tracking: Minimize the deviation cost of tracking the upper-level plan, while taking into account the fuel consumption and equipment wear of diesel engine regulation;

[0123] (b) Stability: Punishes severe fluctuations in power commands and protects the equipment.

[0124] (c) Safety: A penalty term is introduced to address the risk of SOC runaway in energy storage, ensuring that energy storage capacity is used rationally and avoiding overcharging and discharging. The solver will calculate a series of optimal control command sequences for future time steps.

[0125] The process for finding the optimal control command sequence is as follows:

[0126] The second objective function and the physical constraints of the system are expressed in a rigorous mathematical form. This mathematically formalized problem is then input into an embedded optimization solver for calculation. After the solver completes the calculation, it outputs the result, applies the first instruction, and enters the next cycle.

[0127] Execute and adjust ( time time):

[0128] (1) Only the first step is executed: The system will not execute the entire calculated future instruction sequence, but will only send the first step instruction in the sequence to the diesel generator governor and energy storage converter for execution, which ensures the real-time control.

[0129] (2) Reassess: one second later ( time)

[0130] The system again senses the new state of the system through the sensors and compares the actual measured value with that of the previous cycle. By comparing the predicted values ​​at different times, the plan deviation caused by prediction errors and model mismatch can be quantified.

[0131] (3) Scroll Forward: The system scrolls the entire optimization window forward by one second. It completely discards the previous one. The old plan that was about to begin, The latest measurement value at each moment is the initial state. Combined with the updated short-term forecast, a new round of optimization cycle begins.

[0132] Figure 3 This diagram illustrates the principle of rolling optimization using model predictive control. Taking a rolling time window as an example, it shows how, at time k, the optimal control sequence in the future time domain is solved based on the current measured value, the upper-level setpoint, and short-term predictions, with only the first step of the instruction being executed. At time k+1, the time window rolls forward, and optimization is performed again based on the new measured values. Key concepts such as the prediction time domain, control time domain, setpoint trajectory, and optimization trajectory are labeled in the diagram.

[0133] In each control cycle, the controller performs the following steps: (1) collects real-time measurement data of the system; (2) receives setpoints and constraints from the upper layer, as well as ultra-short-term prediction data; (3) solves the low-level objective function to obtain an optimal control command sequence in the future time domain; (4) issues the first command of the sequence to the diesel generator and energy storage system; (5) in the next cycle, the above process is repeated. Through reinforcement learning in this algorithm, the weight matrix Q, R and other parameters of the MPC can be adjusted online to adapt to environmental changes.

[0134] Intraday scheduling and real-time control steps are collaboratively optimized through real-time data feedback, forming a complete control system. The transmission of planned instructions and the feedback of real-time data are tightly coupled: the upper-level plan provides objectives for the lower-level control, and the execution effect of the lower-level control, in turn, verifies and corrects the upper-level plan. This structure effectively balances the computational complexity of the optimization problem and the real-time requirements of system operation, jointly ensuring the optimal operation of the photovoltaic-storage-diesel system in terms of economy, stability, and reliability. The hydrogen energy storage system is modeled as an integrated unit, and its model relationships include the energy conversion efficiency and balance constraints among the three sub-processes: hydrogen production in the electrolyzer, hydrogen storage in the hydrogen storage tank, and power generation in the fuel cell.

[0135] Furthermore, the execution steps include:

[0136] The first instruction in the optimal control instruction sequence is sent to each energy device, and each energy device operates according to the instruction.

[0137] The energy equipment includes photovoltaic arrays, diesel generators, lithium battery energy storage, and hydrogen energy storage systems. The photovoltaic arrays consist of multiple photovoltaic modules connected in series and parallel. Electrical energy is collected through a combiner box and then connected to the AC bus via a photovoltaic inverter, enabling energy exchange with the diesel generator, lithium battery energy storage, hydrogen energy storage system, loads, and grid connection points within the system. The diesel generators consist of an engine, generator, and control system. They are connected to the AC bus via a circuit breaker and can be started and stopped according to dispatch commands for power supply or peak shaving. The lithium battery energy storage system consists of battery clusters, a battery management system (BMS), and a bidirectional converter (PCS). It is connected to the AC bus via the PCS for charging and discharging management of electrical energy. The hydrogen energy storage system consists of an electrolyzer, a hydrogen storage tank, and a fuel cell. The electrolyzer consumes electrical energy to produce and store hydrogen, while the fuel cell uses the stored hydrogen to generate electricity, which is then connected to the AC bus via the converter.

[0138] Check whether the above steps S1 to S5 can be performed in sequence to achieve the target. If not, enter the optimization loop and optimize again until the optimization result meets the final goal and whether the instruction flow and data feedback flow at each level are smooth.

[0139] The criteria for determining whether a standard is met include:

[0140] (a) Correct timing: Whether the data can be transmitted and processed within the specified control period.

[0141] (b) Integrity: No data packets are lost or erroneous. For example, can the real-time control layer continuously receive sensor data, and can the execution layer devices accurately receive and respond to control commands?

[0142] (c) Protocol consistency: The data format and interface protocol between each layer are completely matched.

[0143] (d) Instruction validity: Whether the plans generated by the intraday scheduling layer conform to basic operational common sense.

[0144] (e) Constraint satisfaction: Does any solution generated by the optimization algorithm strictly satisfy all physical constraints?

[0145] (f) Handling extreme scenarios: Simulate a sudden drop in photovoltaic power or a sudden increase in load, and see if the system can respond correctly according to the preset logic.

[0146] (g) After one week or one month of actual operation, whether the calculated total cost is lower than the cost of using the traditional rule control strategy, or whether the expected percentage of energy saving and cost reduction has been achieved.

[0147] (h) Whether the deviation between the actual power of the real-time control layer and the plan issued by the intraday scheduling layer is controlled within the allowable range. If the deviation is too large, it indicates that the real-time tracking effect is poor.

[0148] (i) Whether the actual SOC of the energy storage system is always kept within the safe range and has never been overcharged or over-discharged. At the same time, check whether the daily SOC curve is smooth and avoid drastic fluctuations to extend its lifespan.

[0149] (j) Monitor whether key parameters such as diesel generator power, gradeability, and grid exchange power exceed the constraints. The requirement is that the number of times the limit is exceeded is 0.

[0150] Verification: The model output, i.e., the scheduling plan, is compared with the historical operating data of the real system or a high-fidelity simulation platform to verify whether the model can truly reflect the system operation, and the model parameters are calibrated based on the verification results.

[0151] Model simulation experiments and operational strategy evaluation and analysis: Using the validated model, namely the high-fidelity closed-loop simulation system integrating MILP and MPC, simulation experiments were conducted by changing external input parameters, including weather forecasts and market electricity prices. The impact of different electricity price policies or photovoltaic penetration rates on the system's economic indicators and reliability was quantitatively analyzed, providing a basis for the formulation of optimized system operation strategies.

[0152] Example 2

[0153] The integrated energy management system for industrial parks based on a hierarchical system described in this embodiment includes:

[0154] The prediction and perception layer, based on acquired historical operating data and external data, performs long-term photovoltaic output forecasting, load forecasting, and market electricity price forecasting, providing data support for the intraday dispatch layer;

[0155] The intraday dispatch layer calculates the diesel generator start-up and shutdown and energy storage plans based on long-term photovoltaic output, load and market electricity price, and sends the optimized setpoints and constraints to the real-time control layer.

[0156] The real-time control layer, based on short-term forecast data and intraday scheduling instructions, solves the real-time power control signals of diesel generators and energy storage online and sends them to the execution and equipment layer.

[0157] The execution and equipment layer, including photovoltaic arrays, diesel generators, lithium battery energy storage, and hydrogen energy storage systems, receives and executes instructions issued by the real-time control layer.

[0158] The data feedback closed-loop system is used to monitor the operating status data of each energy device in real time and upload the operating status data to the intraday scheduling layer and the real-time control layer to form a closed-loop optimization.

[0159] This invention relates to a hierarchical integrated energy management system for industrial parks, aiming to achieve multi-energy complementarity and intelligent control. The system adopts a four-layer structure, from top to bottom: a prediction and perception layer, an intraday scheduling layer, a real-time control layer, and an execution and equipment layer, forming a closed-loop control through bottom-layer data feedback. The prediction and perception layer integrates external data such as meteorology and electricity prices, and uses machine learning algorithms to predict load and wind and solar power generation. The intraday scheduling layer, based on long-term prediction results, optimizes unit combinations and formulates start-up, shutdown, and output plans for each piece of equipment using a mixed-integer programming model. The real-time control layer innovatively applies an algorithm combining reinforcement learning and model predictive control. This algorithm uses reinforcement learning to adaptively adjust the prediction model and cost function of MPC online, combined with short-term rolling predictions to dynamically adjust equipment power, effectively addressing uncertainties and achieving more accurate second-level power balance. The execution layer includes diverse physical equipment such as diesel generators, lithium batteries, hydrogen storage, and photovoltaics. The core technology of this solution lies in constructing a cross-timescale prediction-scheduling coupling model and integrating diverse energy storage technologies such as hydrogen energy. Through hierarchical collaborative optimization, it effectively improves the system's adaptability to high proportions of renewable energy and its operational economy.

Claims

1. A method for integrated energy management of industrial park based on hierarchical and hierarchical system, characterized in that, The method comprises the following steps: a prediction and perception step: integrating external data and predicting load and renewable power generation power by using a machine learning algorithm; an intra-day scheduling step: formulating a start-stop and output plan of each energy device in the park based on a long-term prediction result by using a MILP model; a real-time control step: establishing an MPC model based on a short-term prediction and the intra-day scheduling plan, and solving real-time control instructions of each energy device online; an execution step: controlling multiple energy production, conversion and storage devices according to the real-time control instructions; wherein the method further comprises feeding back operation data of each energy device to the prediction and perception step and / or the real-time control step to form a closed-loop control.

2. The hierarchical tier-based industrial park integrated energy management method according to claim 1, wherein, The intra-day scheduling step comprises: establishing a MILP model, which takes the long-term prediction result as input, takes the minimum total system operation cost as a first objective function, and outputs a first optimization relationship of the device start-stop and output plan; wherein the expression of the first objective function is: ， wherein, denotes the total number of time periods of the dispatch cycle, is the cost or benefit resulting from the exchange of power with the grid; is the total cost of the diesel generator g; is the operation and maintenance cost of the energy storage system; constructing constraint conditions of the objective function, including device physical operation constraints and system power balance constraints, wherein the expression of the physical operation constraint is: ， in, Indicates the first diesel generator during the period The actual output power Indicates the first The minimum technical output of a diesel generator. Indicates the first The rated power of the diesel generator. Indicates the first diesel generator during the period Start-stop status, Indicates the first The maximum downhill climbing rate of the diesel generator. Indicates the time period of the energy storage system The charging power, This indicates the maximum allowable charging power of the energy storage system. Indicates the time period of the energy storage system The discharge power, This indicates the maximum allowable discharge power of the energy storage system. Indicates the time period of the energy storage system The state of charge, This indicates the charging efficiency of the energy storage system. This indicates the discharge efficiency of the energy storage system. This indicates the duration of each scheduling period. Indicates the rated capacity of the energy storage system. This indicates the lower limit of the safe state of charge of the energy storage system. These represent the safe upper limit of the state of charge of the energy storage system. Indicates the system during the time period Power exchanged with the power grid This indicates the maximum power that is allowed to be sold to the grid. This indicates the maximum power that is permitted to be purchased from the grid; the expression of the system power balance constraint is: ， wherein, represents the total number of diesel generators in the system, represents the time period of the predicted photovoltaic output, represents the time period of the predicted load demand; The MILP model is solved by a branch and bound algorithm to generate the intraday scheduling plan, which includes the optimal start-stop plan of the diesel generator , output plan , optimal charge-discharge plan of the energy storage system , optimal exchange power plan of the power grid , four plans are set as the set value and physical constraints are issued to the real-time control layer.

3. The hierarchical tier-based industrial park integrated energy management method of claim 2, wherein, The real-time control step comprises: the MPC model takes the short-term prediction data, the upper-layer intra-day scheduling plan and the constraints as input, takes the minimum deviation of real-time power and planned power as a second objective function, and outputs a second optimization relationship of the device real-time power instruction; wherein the expression of the second objective function is: ， in, For at any time The quantification of the risk cost of deviation; For at any time The sum of the total economic cost and equipment depreciation cost resulting from the generator and energy storage power regulation actions in the execution plan; As a quantitative indicator of the plan's execution capability, its value directly reflects how high the risk of the battery running out of control in the future will be due to the current charging and discharging plan; Quantify and penalize the unevenness or severe fluctuations in the generator's active power output plan; the MPC model is optimized and solved by using a rolling optimization and feedback correction algorithm to obtain an optimal control instruction sequence of future time steps, wherein the control instructions include diesel generator power instructions, lithium battery energy storage system power instructions, hydrogen energy storage system power instructions, and fine adjustment instructions of grid exchange power.

4. The hierarchical tier-based industrial park integrated energy management method according to claim 3, wherein, The process of solving the MILP model by using a branch and bound algorithm comprises: initializing a search tree, and the root node is a linear programming relaxation of the original MILP problem; relaxing the integer constraints in the MILP model to form a linear programming problem and solve it to obtain a relaxation solution and a cost lower bound; if the relaxation solution does not satisfy the integer constraints, a fractional integer variable is selected for branching to generate two branch sub-problems; First sub-problem: Injunction = 0, i.e. the generator is off at that instant Second sub-problem: Injunction = 1, i.e. the generator is on at that moment; for each branch sub-problem, recursively solve, bound, and use the cost upper bound of the current optimal integer solution to prune sub-problems that cannot produce better solutions; when the search is completed or the termination condition is met, output the optimal integer solution as the intra-day scheduling plan.

5. The hierarchical tier-based industrial park integrated energy management method according to claim 4, wherein, The execution step comprises: the first instruction in the optimal control instruction sequence is issued to each energy device, and each energy device operates according to the instruction.

6. The hierarchical tier-based industrial park integrated energy management method of claim 5, wherein, The prediction and perception step comprises: obtaining historical operation data of each energy device, future weather forecast data, and dynamic market electricity price data, and predicting long-term photovoltaic output, load and market electricity price by using a machine learning model.

7. The hierarchical tier-based industrial park integrated energy management method of claim 5, wherein, The step of optimizing and solving the MPC model by using a rolling optimization and feedback correction algorithm comprises: The second objective function and the physical constraints of the system are input into an embedded optimization solver for calculation, taking the actual operation state measurement value at the current time as the initial state, combining the set value and short-term prediction data, solving an open-loop optimization problem in a finite time domain, and obtaining the optimal control instruction sequence in the future control time domain; The first control instruction is extracted from the optimal control instruction sequence and issued to the execution equipment; At the next time, based on the new system actual operation state measurement value, the optimization is repeated.

8. A hierarchical layered based industrial park integrated energy management system, characterized in that, It includes: The prediction and perception layer performs long-term photovoltaic output prediction, load prediction, and market electricity price prediction based on the obtained historical operation data and external data, providing data support for the intraday scheduling layer; The intraday scheduling layer solves the diesel generator start-stop and energy storage plan based on the long-term photovoltaic output, load, and market electricity price, and issues the optimized set value and constraints to the real-time control layer; The real-time control layer solves the real-time power control signal of the diesel generator and energy storage based on the short-term prediction data and intraday scheduling instructions, and issues it to the execution and equipment layer; The execution and equipment layer includes photovoltaic arrays, diesel generators, lithium battery energy storage, and hydrogen energy storage systems, which receive and execute the instructions issued by the real-time control layer; The data feedback closed-loop system is used to monitor the operation state data of each energy equipment in real time, and upload the operation state data to the intraday scheduling layer and the real-time control layer to form a closed-loop optimization.

Images