Park integrated energy system scheduling method and device based on time sequence feature extraction and strategy optimization

Abstract

The invention provides a park integrated energy system scheduling method and device based on time sequence feature extraction and strategy optimization, and the method comprises the steps: obtaining park energy state data at a plurality of historical moments, the park energy state data reflecting the output condition of clean energy in a park and the energy load condition in the park; the energy load comprises an electric load and a thermal load; inputting the park energy state data at the plurality of historical moments into a trained time sequence feature extraction model to obtain time sequence features output by the time sequence feature extraction model; the input data are input into the strategy optimization model, an energy system scheduling strategy is determined based on output data of the strategy optimization model, the input data comprise time sequence features and park energy state data at the current moment, and the energy system scheduling strategy comprises the demand response execution amount and the equipment adjustment amount in the park energy system. The accuracy of the scheduling strategy of the park integrated energy system can be improved.

Description

Technical Field

[0001] This invention relates to the field of energy dispatching technology, and in particular to a method and apparatus for dispatching a comprehensive energy system in a park based on time-series feature extraction and strategy optimization. Background Technology

[0002] An industrial park is a designated area with centralized and unified planning, where enterprises and companies of a specific industry or type are set up and managed uniformly. As a core carrier of urban energy consumption and carbon emissions, the park's energy system is undergoing a transformation from a traditional single fossil fuel-based model to a comprehensive energy service model that integrates multiple elements such as "source-grid-load-storage." While the park's integrated electricity-heat energy system, relying on the tiered utilization and complementary scheduling of electricity and heat as core energy sources, has the potential to improve clean energy utilization and reduce system operating costs, in actual engineering applications, clean energy output exhibits certain fluctuations, and the park's electricity load (including commercial office, industrial production, and residential loads) and heat load (including winter heating, domestic hot water, and industrial heat loads) show significant time-series coupling characteristics.

[0003] Existing energy system dispatching methods struggle to capture the trends and cyclical characteristics of clean energy output fluctuations, and often employ a strategy of "independent optimization of electricity and heat subsystems." These methods not only suffer from high wind and solar curtailment rates, leading to energy resource waste, but also potentially cause grid voltage fluctuations and frequency deviations, resulting in insufficient dispatching reliability. Furthermore, user demand response is largely limited to peak shaving and valley filling of electrical loads, failing to incorporate the thermal inertia characteristics of thermal loads to achieve load shifting, thus failing to fully exploit the system's multi-energy complementarity potential.

[0004] In summary, the scheduling schemes generated by existing park energy system scheduling methods are not very accurate. Summary of the Invention

[0005] This invention provides a method and apparatus for scheduling integrated energy systems in industrial parks based on time-series feature extraction and strategy optimization, which addresses the shortcomings of low accuracy in existing industrial park energy system scheduling methods and improves the accuracy of industrial park energy system scheduling schemes.

[0006] This invention provides a method for scheduling a comprehensive energy system in a park based on time-series feature extraction and strategy optimization. The method includes: The park's energy status data at multiple historical moments are acquired. The park's energy status data reflects the output of clean energy and the energy load in the park, including electrical load and thermal load. The energy status data of the park at the multiple historical moments are input into the trained time series feature extraction model to obtain the time series features output by the time series feature extraction model. The input data is fed into the strategy optimization model. Based on the output data of the strategy optimization model, the energy system scheduling strategy is determined. The input data includes the time-series characteristics and the current energy status data of the park. The energy system scheduling strategy includes the demand response execution amount and equipment adjustment amount in the park's energy system.

[0007] According to the method for scheduling a comprehensive energy system in a park based on time-series feature extraction and strategy optimization provided in this application, the park energy status data includes the power of clean energy, electrical load, heat load, electricity price, state of charge of energy storage system, indoor temperature and outdoor temperature.

[0008] According to the integrated energy system scheduling method for industrial parks based on time-series feature extraction and strategy optimization provided in this application, the training process of the strategy optimization model includes: The sample time-series features are input into the strategy optimization model, and the sample energy system scheduling strategy is determined based on the sample output data of the strategy optimization model. Based on the state transition results corresponding to the sample energy system scheduling strategy, the evaluation scores of the sample energy system scheduling strategy on multiple indicators are determined. The evaluation scores of the multiple indicators are weighted and fused to obtain the objective function value; The policy optimization model is updated based on the objective function value.

[0009] According to the integrated energy system scheduling method for parks based on time-series feature extraction and strategy optimization provided in this application, the indicators include a first indicator, which includes park energy cost, carbon emissions, and temperature deviation.

[0010] According to the integrated energy system scheduling method for parks based on time-series feature extraction and strategy optimization provided in this application, the index further includes a second index, which reflects the degree of violation of the energy system scheduling strategy relative to the constraints of the energy system. The constraints of the energy system include comfort constraints, electrical energy supply and demand balance constraints, thermal energy supply and demand balance constraints, and operational constraints of multiple devices in the energy system.

[0011] According to the integrated energy system scheduling method for industrial parks based on time-series feature extraction and strategy optimization provided in this application, the time-series feature extraction model is trained based on multiple sets of training data, each set of training data including sample park energy state data and state prediction labels corresponding to the sample park energy state data; the training process of the time-series feature extraction model includes: Obtain energy status data of the sample park, input the energy status data of the sample park into the time series feature extraction model, and obtain the sample time series features output by the time series feature extraction model; The temporal features of the sample are input into the prediction model to obtain the state prediction result output by the prediction model; Based on the state prediction results and the state prediction labels corresponding to the energy state data of the sample park, the training loss is determined. The temporal feature extraction model is updated based on the training loss. Before inputting the energy status data of the sample park into the time-series feature extraction model, the process includes: performing dynamic normalization processing on the energy status data of the sample park.

[0012] The present invention also provides a scheduling device for a park integrated energy system based on time-series feature extraction and strategy optimization, the device comprising: The historical data acquisition module is used to acquire park energy status data at multiple historical moments. The park energy status data reflects the output of clean energy in the park and the energy load in the park. The energy load includes electrical load and thermal load. The time-series feature extraction module is used to input the park energy status data of the multiple historical moments into the trained time-series feature extraction model to obtain the time-series features output by the time-series feature extraction model. The scheduling strategy output module is used to input input data into the strategy optimization model, and determine the energy system scheduling strategy based on the output data of the strategy optimization model. The input data includes the time series characteristics and the current energy status data of the park. The energy system scheduling strategy includes the demand response execution amount and equipment adjustment amount in the park energy system.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the park integrated energy system scheduling method based on time-series feature extraction and strategy optimization as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the park integrated energy system scheduling method based on time-series feature extraction and strategy optimization as described above.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the park integrated energy system scheduling method based on time-series feature extraction and strategy optimization as described above.

[0016] This invention provides a method and apparatus for scheduling a comprehensive energy system in a park based on time-series feature extraction and strategy optimization. It acquires park energy status data from multiple historical moments, reflecting the output of clean energy and the energy load (including electrical and thermal loads) within the park. This data is input into a trained time-series feature extraction model to obtain time-series features. These features are then input into a strategy optimization model. Based on the output of the strategy optimization model, an energy system scheduling strategy is determined, including demand response execution and equipment adjustment quantities within the park's energy system. By using the time-series feature extraction model to extract the trend and periodic characteristics of clean energy output fluctuations, as well as the time-series coupling characteristics of the park's electrical and thermal loads, and using these as inputs to the scheduling strategy model, the accuracy of the generated scheduling strategy can be effectively improved. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the integrated energy system scheduling method for parks based on time-series feature extraction and strategy optimization provided in this application.

[0019] Figure 2 This is a comparison of the model training convergence curves in the integrated energy system scheduling method for parks based on time-series feature extraction and strategy optimization provided in this application.

[0020] Figure 3 This is a typical daily renewable energy output and load data graph from the simulation experiment of the park integrated energy system scheduling method based on time-series feature extraction and strategy optimization provided in this application.

[0021] Figure 4 This is a simulation experiment of the solar power optimization scheduling results in the park integrated energy system scheduling method based on time-series feature extraction and strategy optimization provided in this application.

[0022] Figure 5 This is a simulation experiment of the typical daily thermal power optimization scheduling result in the park integrated energy system scheduling method based on time-series feature extraction and strategy optimization provided in this application.

[0023] Figure 6This is a comparison chart of operating cost errors in the simulation experiment of the integrated energy system scheduling method for parks based on time-series feature extraction and strategy optimization provided in this application.

[0024] Figure 7 This is a robustness analysis diagram of the reward function parameters in a simulation experiment of a park integrated energy system scheduling method based on time-series feature extraction and strategy optimization.

[0025] Figure 8 This is a schematic diagram of the structure of the integrated energy system scheduling device for parks based on time-series feature extraction and strategy optimization provided in this application.

[0026] Figure 9 This is a schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0028] The following is combined Figure 1-7 This application describes a method for scheduling a comprehensive energy system in a park based on time-series feature extraction and strategy optimization. For example... Figure 1 As shown, the integrated energy system scheduling method for industrial parks based on time-series feature extraction and strategy optimization provided in this application includes the following steps: S110. Obtain park energy status data at multiple historical moments. The park energy status data reflects the processing of clean energy in the park and the energy load in the park. The energy load includes electrical load and heat load. S120. Input the energy status data of the park at multiple historical moments into the trained time series feature extraction model to obtain the time series features output by the time series feature extraction model. S130. Input the input data into the strategy optimization model. Based on the output data of the strategy optimization model, determine the energy system scheduling strategy. The input data includes time-series characteristics and the current energy status data of the park. The energy system scheduling strategy includes the demand response execution amount and equipment adjustment amount in the park's energy system.

[0029] The method provided in this application is aimed at the scheduling of the integrated energy system of the park. The integrated energy system of the park is a multi-element integrated energy system that coordinates "source-grid-load-storage", including clean energy (such as photovoltaic power generation and wind power), combined heat and power (CHP) units, energy storage systems, gas boilers, and electric boilers, and is connected to the external power grid and natural gas grid. The load of the integrated energy system of the park includes electrical load and heat load.

[0030] In one possible implementation, the park scheduling problem can be mapped as an MDP (Markov Decision Process) quadruple (S, A, R, Ptrans). The state space S represents the park's energy state data at multiple historical moments. In some embodiments, these historical moments can be set as 24 time steps prior to the current moment. By introducing 24 historical time-series data, a time-series feature extraction model (LSTM) is used to capture the trend characteristics of clean energy output and load changes. This clean energy can be photovoltaic power generation, or it can include both photovoltaic and wind power.

[0031] The park's energy status data reflects the output of clean energy and the energy load within the park. Specifically, it can include clean energy power, electrical load, heat load, electricity price, state of charge of energy storage systems, indoor temperature, and outdoor temperature. The park's energy status data at multiple historical moments can be represented by a single state vector as follows: ; in The power sequence of the photovoltaic system within the time interval [t-23,t]; The sequence of the base load of the electrical load within the time interval [t-23,t]; The basic load of the heat load is the sequence of the base load within the time interval [t-23,t]. The sequence of electricity prices within the time interval [t-23,t]; The state of charge (BESS) at time t; Let be the indoor temperature at time t; Let be the outdoor temperature at time t.

[0032] The action space A in the quadruple represents the various possible energy system scheduling strategies, with action variables selected as the core equipment adjustment quantities and demand response execution quantities. To avoid increased decision complexity caused by discrete start-stop variables, the actions in action space A can be defined as follows: ; in, This indicates the power generation adjustment amount of the combined heat and power unit; This indicates the amount of charging and discharging power adjustment in the battery energy storage system; This indicates the power consumption adjustment amount of the electric boiler; This indicates the adjustment amount of the heat production capacity of the gas-fired boiler; This indicates the load reduction adjustment amount on the power side; Indicates the load transfer adjustment amount; This indicates the amount of heat load adjustment.

[0033] In other words, the generated energy system dispatch strategy includes The values ​​that each element in the array can take.

[0034] The method provided in this application uses a temporal feature extraction (LSTM) model and a policy optimization (PPO) model to generate an action (energy system scheduling policy) in the action space A based on the state space S (energy state data of the park at multiple historical moments).

[0035] Specifically, the input to the time-series feature extraction model is the energy status data of the park at multiple historical moments. The output of the time-series feature extraction model can reflect the time-series changes in the output of clean energy and the energy load in the park. The time-series features and the current energy status data of the park are input into the strategy optimization model to obtain the energy system scheduling strategy. Based on the energy system scheduling strategy, the demand response execution and equipment adjustment in the park's energy system are controlled and adjusted.

[0036] The method provided in this application acquires energy status data of a park at multiple historical moments. This data reflects the output of clean energy and the energy load (including electrical and thermal loads) within the park. The data is input into a trained time-series feature extraction model to obtain time-series features. These features are then input into a strategy optimization model. Based on the output of the strategy optimization model, an energy system scheduling strategy is determined, including demand response execution and equipment regulation within the park's energy system. By using the time-series feature extraction model to extract the trend and periodic characteristics of clean energy output fluctuations, as well as the time-series coupling characteristics of the park's electrical and thermal loads, and using these as inputs to the scheduling strategy model, the accuracy of the generated scheduling strategy can be effectively improved.

[0037] In one possible implementation, the temporal feature extraction model can be a 3-layer network with an input dimension of 24×7, 64 hidden layer neurons, and an output dimension of 24. Understandably, the structure of the temporal feature extraction model can adopt the network architecture of existing temporal feature extraction models.

[0038] To improve the performance of the time-series feature extraction model and output more accurate time-series features, one possible implementation of the method provided in this application involves training the time-series feature extraction model based on multiple sets of training data. Each set of training data includes energy state data of a sample park and corresponding state prediction labels. The training process of the time-series feature extraction model includes: Obtain energy status data of the sample park, input the energy status data of the sample park into the time series feature extraction model, and obtain the sample time series features output by the time series feature extraction model; Input the temporal features of the samples into the prediction model and obtain the state prediction results output by the prediction model; Based on the state prediction results and the state prediction labels corresponding to the energy state data of the sample park, the training loss is determined. The temporal feature extraction model is updated based on training loss. Before inputting the energy status data of the sample park into the time series feature extraction model, the process includes: performing dynamic normalization on the energy status data of the sample park.

[0039] In the data preprocessing stage, one year of historical data of the park, namely PV output, load, electricity price and temperature, were collected and divided into training set and validation set in a 7:3 ratio. The state variables were normalized and the time series feature extraction model was trained with the goal of minimizing the one-step prediction error of PV output and load. When the training objective was achieved, such as when the error of the validation set was ≤5%, the offline training of the time series feature extraction model was completed.

[0040] By determining the training loss based on the state prediction results and the state prediction labels corresponding to the energy state data of the sample park, and updating the time series feature extraction model based on the training loss, the extracted time series features can accurately reflect the time series coupling characteristics of photovoltaic power generation output fluctuations and park loads, providing more accurate information for energy system scheduling strategy decisions and improving the accuracy of energy system scheduling strategies.

[0041] In one possible implementation, the energy status data of the sample park can be dynamically normalized to avoid the collapse of the strategy and value function caused by the fixed statistics of traditional batch normalization. This reduces the fluctuation range of the training data from ±15% to ±5%, indirectly reducing the algorithm performance error caused by data distribution offset.

[0042] The training process of the policy optimization model includes: The sample time-series features are input into the strategy optimization model, and the sample energy system scheduling strategy is determined based on the sample output data of the strategy optimization model. Based on the state transition results corresponding to the energy system scheduling strategies of the sample energy systems, the evaluation scores of the energy system scheduling strategies on multiple indicators are determined. The evaluation scores of multiple indicators are weighted and fused to obtain the objective function value; The optimization model is based on the objective function value update strategy.

[0043] The reward function R in the quadruple is the objective function value used to evaluate the merits of the strategy output by the strategy optimization model. In one possible implementation, the multiple indicators used to evaluate the energy system scheduling strategy include a first indicator, which includes park energy costs, carbon emissions, and comfort.

[0044] The evaluation score of the first indicator reflects the benefits obtained from the energy system dispatch strategy. Specifically, the lower the energy cost of the park, the higher the evaluation score of this indicator; the lower the carbon emissions, the higher the evaluation score of this indicator; and the higher the comfort level, the higher the evaluation score of this indicator.

[0045] Specifically, the park's energy costs Covering the interaction costs between power grids Natural gas procurement costs BESS depreciation cost and demand response compensation costs Where T is the scheduling period; t is the time step; and i is the benchmark rate of return. Let t be the interaction cost between the park and the power grid; Let t be the natural gas procurement cost; Let BESS be the depreciation cost at time t; Let t be the demand response compensation cost. The formula for calculating the park's energy cost is as follows: ; The interaction cost between the park and the power grid includes the electricity purchase and sales processes. Taking into account the time-of-use pricing policy and the power grid's electricity sales discount (electricity purchase is based on the catalog price, and electricity sales are settled at 80% of the purchase price), the formula for calculating the power grid's electricity purchase and sales cost at time t is as follows: ; in, The time-of-use electricity price at time t; Let t be the power exchange between the park and the power grid. >0 indicates electricity purchased from the power grid. <0 indicates selling electricity to the grid; The time step is usually set to 1 hour; This is the electricity sales discount factor.

[0046] The natural gas consumption of the park's energy system covers the power generation of the CHP unit and the heat production of GB. Considering the impact of natural gas calorific value and equipment efficiency, the formula for calculating the natural gas procurement cost at time t is as follows: ; in The price per cubic meter of natural gas is 3.2 yuan / m³. 3 Referencing domestic industrial natural gas prices; Let t be the power generation capacity of the CHP unit; To optimize the power generation efficiency of the CHP unit, parameters for a 500kW gas turbine CHP unit are adapted. This refers to the lower heating value of natural gas. Let GB be the heat generation power at time t; GB heating efficiency.

[0047] Battery energy storage systems (BESS) need to be depreciated using a unit power cycle method, taking into account their charge-discharge cycle life and daily maintenance costs, to reasonably calculate costs. The depreciation and maintenance costs of the BESS at time t are as follows. The calculation formula is as follows: ; in Let t be the depreciation and maintenance cost of BESS. The depreciation factor per unit power for BESS; The BESS unit power maintenance factor; Let t be the absolute value of the charge and discharge power of BESS (ensuring that the cost is positive, when >0 indicates discharge. <0 indicates charging).

[0048] Demand response encompasses three categories: electrical load shifting, electrical load reduction, and thermal load regulation. The demand response compensation cost at time t represents the sum of compensation costs generated by the three types of demand response measures: load transfer, load reduction, and heat load adjustment. It is calculated using the following formula: ; in This is the unit compensation coefficient for electrical load transfer; Let t be the amount of electrical load transferred at time t; This is the compensation coefficient for the unit of electrical load reduction; Let t be the amount of electrical load reduction at time t; This is the unit compensation coefficient for heat load adjustment; Let t be the amount of heat load adjustment at time t.

[0049] Carbon emissions encompass both implicit carbon emissions from electricity purchases and direct carbon emissions from natural gas combustion. The calculation formula is as follows: ; in This represents the system's total carbon emissions over 24 hours. Carbon emission factors for supplying electricity to the power grid; Let t be the power output purchased by the park from the power grid; Carbon emission factors from natural gas combustion; Let t be the amount of natural gas consumed at time t.

[0050] To calculate the average deviation between the actual indoor temperature and the set temperature over 24 hours, and to quantify the average degree of temperature fluctuation, the average temperature deviation is calculated as follows: ; Where T represents the total time interval; t represents the specific moment. The actual indoor temperature at time t is collected in real time by a room temperature sensor. The temperature setpoint at time t; The absolute deviation between the actual temperature and the set temperature at time t.

[0051] Furthermore, in some cases, higher returns are not always better, because energy systems are subject to constraints. In another possible implementation of the method provided in this application, the multiple indicators used to evaluate energy system scheduling strategies also include a second indicator. The second indicator reflects the degree to which the sample energy system scheduling strategy violates the constraints of the energy system. Specifically, the constraints of the energy system include comfort constraints, electrical energy supply and demand balance constraints, thermal energy supply and demand balance constraints, and operational constraints of multiple devices in the energy system.

[0052] Comfort constraints are used to control indoor temperature fluctuations and the duration of electrical load interruptions, ensuring thermal comfort while preventing power outages from affecting normal energy use. To limit the maximum 24-hour average indoor temperature deviation from the set upper limit, the temperature deviation upper limit constraint is as follows: ; in This represents the average deviation of indoor temperature over 24 hours. This represents the upper limit of the average temperature deviation.

[0053] To limit the total duration of power outages within 24 hours and prevent excessive power outages from affecting users' normal energy consumption, the power outage duration constraints are as follows: ; in t is the duration of each time interval; Let be the indicator function, representing the amount of electrical load interruption at time t. That is, if there is an interruption =1, otherwise =0; t If there is a load interruption at time t, the interruption duration is included in this period. This represents the maximum permissible duration of electrical load interruption. Thermal comfort is controlled by the average temperature deviation to prevent the temperature from deviating from the set value for an extended period. Energy reliability is controlled by the total duration of electrical load interruption to prevent excessively frequent power outages, ultimately ensuring user energy comfort.

[0054] With the goal of balancing the supply and demand of electricity and heat energy, and ensuring the matching of power production and consumption within the system, the relevant constraints are as follows. Considering the power output of photovoltaic (PV), combined cooling, heating and power (CHP) generation, grid interaction, battery energy storage system (BESS) charging and discharging, and the supply and demand balance of electrical load, the power balance constraints are as follows: ; in The actual output of PV at time t; The baseline value of the park's electrical load at time t; Let t be the power consumption of the electric boiler (EB) at time t.

[0055] Considering the heat production from CHP, gas-fired boilers (GB), EB, and the heat dissipation from the heating network, the heat power balance constraints are as follows: ; in Let t be the heat output of the CHP unit; Let be the heat generation power of EB at time t; The baseline value of the park's heat load at time t; The heat loss of the heating network at time t; Let t be the fluctuation of the park's heat load at time t.

[0056] To ensure the safe and stable operation of the equipment, the output of the core equipment (CHP, BESS, EB, GB) must meet the constraints of rated capacity, start-stop characteristics, etc.

[0057] The CHP unit's generating power must meet the rated capacity limit, and the start-up and shutdown states must comply with the minimum operating time requirement. The constraints are as follows: ; in This is the minimum continuous operating time for the CHP unit, used to constrain the duration the unit needs to run continuously after startup, and to avoid frequent start-ups and shutdowns. This represents the start / stop status of the CHP unit at time t. Used to determine whether the unit should be started at time t. =1 and =0, then the left side is This indicates that subsequent requirements must be met after startup. Duration of operation; right side It is t to The cumulative operating status of the unit within a given time period; if it continues to run after startup, the sum on the right side is... The equation holds true; if the machine stops midway, the sum on the right side will be less than 1 / 3. The constraints are not met.

[0058] BESS charging and discharging power and state of charge (SOC) must meet the safe range, and the initial and final SOC of each day must be consistent to avoid long-term undercharging or overcharging. The constraints are as follows: ; ; in The State of Charge (SOC) of the Battery Energy Storage System (BES) at time t describes the proportion of remaining battery capacity to total capacity. SOC is a physical quantity that describes the percentage of remaining battery capacity, typically ranging from 0 to 1 (or 0% to 100%). 0 indicates the battery is fully discharged with no remaining capacity; 1 (or 100%) indicates the battery is fully charged and has reached its maximum capacity. By monitoring and controlling SOC, the remaining battery capacity can be understood, allowing for reasonable charge and discharge management and ensuring the safe and efficient operation of the battery energy storage system. Let BESS be the state of charge at time (t+1), reflecting the dynamic change of SOC over time; The charge and discharge power of BESS at time t; The time step represents the time interval between adjacent times t and (t+1); Let BESS be the charge / discharge efficiency at time t; The rated capacity of the BESS describes the maximum total amount of electrical energy that the battery can store; This is a boundary constraint on the charge state of the BESS. It means that at the initial time t=1 and the final time t=T+1 of the scheduling cycle, the charge state of the BESS must remain consistent and equal to a specific charge state value. The purpose of this is to prevent the BESS from being in a state of undercharge or overcharge for a long time, to protect the battery's lifespan and the system's stable operation, and to ensure that the battery's state of charge can return to its initial level after a scheduling cycle ends, so that the next cycle can be scheduled.

[0059] The power consumption of EB and the heat generation of GB must meet the rated capacity limit, and both must avoid operating at full load simultaneously. The constraints are as follows: 1; in Let be the power of the electric boiler at time t; This refers to the maximum permissible power of the electric boiler. The thermal power of the gas-fired boiler at time t; This represents the maximum permissible thermal power of the gas-fired boiler.

[0060] When the indicators include both the first and second indicators, the formula for calculating the objective function value can be expressed as: ; in Let be the objective function value at time t; Cost weighting; Let be the total cost at time t; Carbon emission weights; Let be the carbon emissions at time t; To prioritize comfort, the impact of temperature deviation on rewards should be made more prominent, thus prioritizing the satisfaction of comfort needs. The temperature deviation at time t; To constrain the penalty weights; Let t be the penalty cost for violating the constraint.

[0061] The input to the strategy optimization model is the time-series features output by the time-series feature extraction model and the current real-time energy state data of the park. In one possible implementation, the output of the strategy optimization model is the action probability distribution, i.e., Ptrans in the four-tuple. Ptrans describes the probability of the system transitioning from the current state to the next state after executing an action in the action space A, given the current state (including photovoltaic power, load, electricity price, energy storage state of charge, indoor and outdoor temperature, etc.). Based on the current state (including information such as photovoltaic power, load, electricity price, energy storage state of charge, indoor and outdoor temperature), after executing an action (such as adjusting the power generation of the cogeneration unit, adjusting the charging and discharging of energy storage), the probability distribution of the system transitioning to a new state at the next time step is calculated. Based on the state probability corresponding to each action, the state value corresponding to each action can be calculated, and the action with the highest state value is selected as the energy system scheduling strategy.

[0062] During the training phase of the policy optimization model, the agent performs actions in the simulation environment. At each step, historical time-series features are extracted through LSTM and combined with the real-time state to generate actions, and the trajectory (st, at, rt, st+1) is recorded. The policy update is based on the collected trajectory to calculate the advantage function. The policy is updated by minimizing the PPO clipped objective function. The process is repeated until the reward fluctuation is ≤3%.

[0063] In one possible implementation, after determining the energy system scheduling strategy based on the output data of the strategy optimization model, the normalization parameters of the time-series feature extraction model can be delegated based on the current energy system scheduling strategy. That is, the real-time status at time t and historical 24-step data are obtained through the park energy management system (EMS). In the time-series feature extraction stage, historical data is input into the pre-trained LSTM model, and the time-series feature vector is output. In the optimal action generation stage, the time-series features and real-time status are fused and input into the PPO policy network to output the optimal action that satisfies the constraints, which is then converted into equipment control instructions. In the feedback and adjustment stage, after executing the instructions, the actual rewards and constraint violations are calculated, and the LSTM normalization parameters are updated to provide a basis for decision-making in the next time step, thereby realizing closed-loop scheduling.

[0064] To verify the improvement of the method provided in this application, the system operating cost error is used to evaluate the performance accuracy of the method provided in the optimization of integrated energy management in the park. This error is calculated by comparing the system operating cost obtained by different optimization algorithms with the system operating cost obtained by the theoretically optimal algorithm (mixed integer linear programming, MILP). It can accurately reflect the slight differences in economic performance of different algorithms, and is especially suitable for the horizontal comparison and evaluation of the economic efficiency of multiple algorithms in the same system scenario.

[0065] Specifically, system operating cost error The calculation formula is as follows: ; in This represents the system operating cost error. The smaller the value, the closer the economic performance of other optimization algorithms (such as PSO, PPO, etc.) is to the theoretical optimal performance obtained by the MILP algorithm. The system operating cost is represented by other optimization algorithms (such as particle swarm optimization (PSO), stochastic programming (SP), traditional PPO, and the method provided in this application). This represents the theoretically optimal system operating cost obtained by the MILP algorithm; The relative deviation term is calculated by dividing the difference between the algorithm cost and the theoretical optimal cost by the theoretical optimal cost, thus eliminating the impact of cost magnitude differences on error assessment.

[0066] Error calculations were performed by comparing the system operating costs calculated by each algorithm with the theoretical optimal cost of the MILP algorithm. The results show that the PSO algorithm... The mean is 5.12%, the SP algorithm is 4.87%, the traditional PPO algorithm is 3.65%, while the method provided in this application... The average error is only 1.27%. From the error curve trend, the method provided in this application not only has the smallest error, but also has a daily fluctuation range of only 0.32%, indicating that its economic performance is closer to the theoretical optimum and its stability is better when facing renewable energy fluctuations and load changes.

[0067] Meanwhile, robustness analysis of key parameters in the objective function further reveals the impact of parameter settings on algorithm performance error. If W1 is too small, the sensitivity to cost optimization decreases. The W2 value increased from 1.27% to 2.91%; if it were too high, carbon emissions would increase by 4.2%, temperature deviation would reach 1.8℃, and training error fluctuation would widen from 0.32% to 0.63%. If W2 was too low, carbon emissions would increase by 6.8%; if it were too high, it would increase the consumption of clean energy, etc. The rate rises to 1.89%, reducing economic efficiency. If W3 is too small, the temperature deviation increases from 1.3℃ to 2.2℃ (exceeding the comfort range of ≤2℃); if it is too large, it increases the need for additional heating network power. The rate of error increases to 1.65%, compromising economic efficiency. If W4 is too small, the constraint violation rate rises from 1.2% to 4.1%, making the algorithm prone to getting trapped in local optima. The error rate rises to 2.38%; if it is too high, even a slight violation will result in a heavy penalty, the reward function fluctuates wildly, the error fluctuation range drops to 0.72%, and the algorithm stability decreases.

[0068] Furthermore, this application avoids the strategy and value function collapse problem caused by traditional batch normalization of fixed statistics by dynamically normalizing the system state, reducing the fluctuation range of training data from ±15% to ±5%, indirectly reducing algorithm performance errors caused by data distribution offset. Combined with system operating cost error assessment and robustness analysis of reward function parameters, it can be determined that the method provided in this application can accurately control economic performance deviations and ensure the balance and operational stability of multi-objective optimization in integrated electric-thermal energy systems in industrial parks, meeting the dual requirements of accuracy and reliability for engineering applications. In summary, the method provided in this application extracts time-series data features from multiple historical moments of the system by introducing a time-series feature extraction model. This fully explores the uncertain trend patterns in the data, avoids the policy fluctuation problem caused by neglecting the temporal correlation in traditional scheduling methods, dynamically normalizes and updates the real-time state of the system, reduces the original fluctuation range of training data from ±15% to within ±5%, and completely avoids the policy and value function collapse problem that may be caused by traditional batch normalization with fixed statistics. This achieves a better energy management strategy and shows significant advantages in terms of system operating cost error control and real-time performance.

[0069] To fully verify the effectiveness, economy, and stability of the method provided in this application, simulation experiments were conducted based on the method provided in this application, using an industrial park covering an area of ​​3km².2 The study focuses on a comprehensive energy system comprising 20 production plants and 5 office buildings, with a peak electrical load of 1200kW and a peak thermal load of 800kW. Simulation experiments are conducted using real historical data. Through horizontal comparison with traditional optimization algorithms and conventional PPO algorithms, the technical advantages of the method provided in this application are fully verified.

[0070] In the basic data sourcing and processing phase, the basic data used in the experiment were all derived from one year of historical operating data of the park and publicly available data from power operators, specifically including: (1) Renewable energy data: Distributed photovoltaic (PV) power output data with an installed capacity of 500kW and a time resolution of 1h, covering different weather scenarios such as sunny, cloudy, and rainy. Backup distributed wind power output data with an installed capacity of 300kW is used for extreme scenario verification.

[0071] (2) Load data: Electrical load includes production equipment load, office air conditioning load, lighting load, and heat load includes production heat load, winter heating load, and domestic hot water load, with a time resolution of 1 hour. (3) Energy price data: time-of-use electricity price and natural gas unit price. (4) Environmental data: outdoor temperature, light intensity and wind speed data, used to correlate PV output with heat load fluctuation patterns.

[0072] The data was divided into a training set (8760h × 70%) and a test set (8760h × 30%) in a 7:3 ratio. Seven typical days (including weekdays, weekends, and extreme weather days) were randomly selected for the test set to ensure the universality of the verification results. The core equipment parameters of the park's integrated electric-thermal energy system are shown in Table 1, all of which refer to the selection standards of actual projects.

[0073] Table 1

[0074] In simulation experiments, the model structure and training parameters of the method provided in this application were determined through grid search optimization. The temporal feature extraction model adopts a 3-layer LSTM network with an input dimension of 24 steps × 7 classes of state variables, 64 hidden layer neurons, and an output dimension of 24 (temporal feature vector). The activation function is ReLU, the optimizer is Adam, the learning rate is 3 × 10⁻⁴, and the training objective is to minimize the 1-step prediction error of PV output and load with a validation set error ≤ 5%. The decision optimization model adopts a 3-layer fully connected network for both the policy network and the value network, with an input dimension of LSTM output features of 24 dimensions + real data. The time state is 7-dimensional, and the output dimensions are 7-dimensional action probability distribution, action variables, and 1-dimensional state value; the learning rate is 3×10⁻⁴ policy network and 1×10⁻³ value network; the discount factor γ is 0.98 to balance short-term and long-term rewards; the pruning coefficient ε is 0.2 to limit the policy update amplitude; the batch size is 64 empirical trajectory sampling batches; the number of training iterations is 1000 steps to ensure convergence stability; the reward function parameters are cost weight w1=0.01, carbon emission weight w2=0.005, comfort weight w3=5, and constraint penalty weight w4=100, and the optimal combination is determined through robustness analysis.

[0075] Figure 2 This paper compares the training convergence curves of the model in the method provided in this application with those of the conventional PPO algorithm, using average reward value as the evaluation metric. In the initial stage (0-200 steps), both algorithms suffer from high constraint violation penalties and low average reward values ​​due to insufficient agent exploration of the environment. As the number of training steps increases, the agent gradually accumulates experience, and the reward value continuously rises and tends to stabilize. The policy optimization model in the method provided in this application reaches convergence around 450 steps, with a stable average reward value of -210±5 and a fluctuation range of only 2.38%. The conventional PPO algorithm requires more than 700 steps to converge, with a stable average reward value of -250±12 and a fluctuation range of 4.8%. The core reason for the difference in convergence performance lies in the fact that the policy optimization model in the method provided in this application captures the temporal correlation between PV output and load through LSTM, while dynamic state normalization reduces the training data fluctuation range from ±15% to ±5%, avoiding the policy collapse problem caused by data distribution shift in PPO, and significantly improving convergence speed and stability.

[0076] For example Figure 3 Taking the PV output and load data of a typical day (working day) shown in the figure as an example, the power optimization scheduling results of the method provided in this application are as follows: Figure 4 , Figure 5 As shown.

[0077] The results of power dispatch are as follows Figure 4 As shown, Figure 4 The Chinese translations of the English terms are shown in Table 2.

[0078] Table 2

[0079] from Figure 4 As can be seen, the method provided in this application prioritizes the use of PV output to meet the electrical load, while controlling BESS discharge to reduce the purchase of electricity from the grid; at night, the algorithm controls BESS charging, while utilizing the low-load operation of the CHP unit to reduce the cost of peak electricity consumption the next day.

[0080] Thermal power scheduling results are as follows Figure 5 As shown, Figure 5 The vertical axis represents thermal power, and the horizontal axis represents time (hours). Figure 5 The Chinese translations of the other English terms are shown in Table 3.

[0081] Table 3

[0082] from Figure 5 As can be seen, the method provided in this application prioritizes the use of CHP and GB heat generation in the morning, with EB serving only as a backup, thus avoiding EB power consumption during periods of high electricity prices. At midday, the algorithm reduces GB output and utilizes the heat storage characteristics of the heating network to shift some of the heat load to off-peak electricity prices, thereby reducing natural gas consumption. Figure 6 The paper presents the system operating costs of the method provided in this application on seven typical days, as well as the cost error relative to MILP. The results show that the economic advantage of the method provided in this application stems from the accurate trend prediction of PV output by LSTM, which reduces power curtailment losses, and the improvement of policy stability by dynamic normalization, which reduces constraint violation penalties.

[0083] The online decision-making time for each method is shown in Table 4. The method provided in this application fully meets the real-time scheduling requirements of the park.

[0084] Table 4

[0085] To verify the impact of four key parameters (w1, w2, w3, w4) in the objective function on the algorithm performance, a robustness experiment was conducted using the controlled variable method. The results are as follows: Figure 7 As shown: (1) Cost weight w1: When w1=0.001, the cost optimization sensitivity is insufficient and the cost error rises to 2.91%; when w1=0.1, the excessive focus on cost leads to an increase of 4.2% in carbon emissions and a temperature deviation of 1.8℃; when w1=0.01, the cost error of 1.27% is optimal and the multi-objective balance is optimal.

[0086] (2) Carbon emission weight w2: When w2=0.001, carbon emissions increase by 6.8%; when w2=0.01, the cost error rises to 1.89%; when w2=0.005, carbon emissions decrease by 3.14% and costs are controllable. (3) Comfort weight w3: When w3=2, the temperature deviation exceeds 2.2℃ and is outside the comfort range; when w3=8, the cost error rises to 1.65%; when w3=5, the temperature deviation is 1.3℃ and the cost is stable. (4) Constraint penalty weight w4: When w4=50, the constraint violation rate rises to 4.1% and the cost error is 2.38%; when w4=150, the reward fluctuation range reaches 0.72%; when w4=100, the constraint violation rate is 1.2% and the fluctuation is minimal. Experiments show that the method provided in this application is highly robust to parameter variations. Under the optimal combination of w1=0.01, w2=0.005, w3=5, and w4=100, it can achieve a synergistic effect of three-dimensional objectives: economy, environmental friendliness, and comfort. The reward fluctuation results were obtained through statistical analysis during the training process, and their trend is consistent with the system operating cost error.

[0087] The method provided in this application reduces the convergence steps of the strategy optimization model by 35.7% and the fluctuation amplitude by 52.1% compared to conventional PPO, demonstrating a significant synergistic effect between dynamic normalization and time series extraction. The average operating cost is only 1.29% higher than the theoretically optimal MILP, and 3.65%, 5.12%, and 4.87% lower than conventional PPO, PSO, and SP, respectively. The online decision-making time is 28ms, meeting the engineering requirement of ≤100ms for real-time scheduling in industrial parks, far superior to traditional algorithms. The algorithm maintains stable performance even when the objective function parameters change over a large range, adapting to the personalized needs of different industrial parks. This method can be directly integrated into the park's energy management system (EMS), providing a reliable optimized scheduling scheme for integrated electric-thermal energy systems in parks with a high proportion of renewable energy access, and possesses significant engineering application value.

[0088] The following describes the integrated energy system scheduling device for parks based on time-series feature extraction and strategy optimization provided in this application. The integrated energy system scheduling device for parks based on time-series feature extraction and strategy optimization described below can be referred to in correspondence with the integrated energy system scheduling method for parks based on time-series feature extraction and strategy optimization described above. Figure 8 As shown, the integrated energy system scheduling device for parks based on time-series feature extraction and strategy optimization provided in this application includes the following modules: The historical data acquisition module 810 is used to acquire park energy status data at multiple historical moments. The park energy status data reflects the output of clean energy in the park and the energy load in the park. The energy load includes electrical load and heat load. The time-series feature extraction module 820 is used to input the energy status data of the park at multiple historical moments into the trained time-series feature extraction model to obtain the time-series features output by the time-series feature extraction model. The scheduling strategy output module 830 is used to input input data into the strategy optimization model. Based on the output data of the strategy optimization model, the energy system scheduling strategy is determined. The input data includes time series characteristics and the current energy status data of the park. The energy system scheduling strategy includes the demand response execution amount and equipment adjustment amount in the park's energy system.

[0089] Figure 9 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 9 As shown, the electronic device may include: a processor 910, a communication interface 920, a memory 930, and a communication bus 940. The processor 910, communication interface 920, and memory 930 communicate with each other via the communication bus 940. The processor 910 can call logical instructions in the memory 930 to execute a park integrated energy system scheduling method based on time-series feature extraction and strategy optimization. This method includes: acquiring park energy status data at multiple historical moments, reflecting the output of clean energy and the energy load in the park, including electrical and thermal loads; inputting the park energy status data at multiple historical moments into a trained time-series feature extraction model to obtain the time-series features output by the model; inputting the input data into a strategy optimization model, and determining an energy system scheduling strategy based on the output data of the strategy optimization model. The input data includes the time-series features and the current park energy status data. The energy system scheduling strategy includes the demand response execution amount and equipment adjustment amount in the park energy system.

[0090] Furthermore, the logical instructions in the aforementioned memory 930 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0091] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the integrated energy system scheduling method for parks based on time-series feature extraction and strategy optimization provided by the above methods. The method includes: acquiring park energy status data at multiple historical moments, the park energy status data reflecting the output of clean energy and the energy load in the park, the energy load including electrical load and thermal load; inputting the park energy status data at multiple historical moments into a trained time-series feature extraction model to obtain the time-series features output by the time-series feature extraction model; inputting the input data into a strategy optimization model, and determining an energy system scheduling strategy based on the output data of the strategy optimization model, the input data including the time-series features and the park energy status data at the current moment, the energy system scheduling strategy including the demand response execution amount and equipment adjustment amount in the park energy system.

[0092] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the above-described method for scheduling a comprehensive energy system in a park based on time-series feature extraction and strategy optimization. The method includes: acquiring park energy status data at multiple historical moments, the park energy status data reflecting the output of clean energy and the energy load in the park, the energy load including electrical load and thermal load; inputting the park energy status data at multiple historical moments into a trained time-series feature extraction model to obtain the time-series features output by the time-series feature extraction model; inputting the input data into a strategy optimization model, and determining an energy system scheduling strategy based on the output data of the strategy optimization model, the input data including the time-series features and the park energy status data at the current moment, the energy system scheduling strategy including the demand response execution amount and equipment adjustment amount in the park energy system.

[0093] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0094] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for scheduling a comprehensive energy system in a park based on time-series feature extraction and strategy optimization, characterized in that, The method includes: The park's energy status data at multiple historical moments are acquired. The park's energy status data reflects the output of clean energy and the energy load in the park, including electrical load and thermal load. The energy status data of the park at the multiple historical moments are input into the trained time series feature extraction model to obtain the time series features output by the time series feature extraction model. The input data is fed into the strategy optimization model. Based on the output data of the strategy optimization model, the energy system scheduling strategy is determined. The input data includes the time-series characteristics and the current energy status data of the park. The energy system scheduling strategy includes the demand response execution amount and equipment adjustment amount in the park's energy system.

2. The method for scheduling a comprehensive energy system in a park based on time-series feature extraction and strategy optimization according to claim 1, characterized in that, The park's energy status data includes the power of clean energy, electrical load, heat load, electricity price, state of charge of energy storage systems, indoor temperature, and outdoor temperature.

3. The method for scheduling a comprehensive energy system in a park based on time-series feature extraction and strategy optimization according to claim 1, characterized in that, The training process of the policy optimization model includes: The sample time-series features are input into the strategy optimization model, and the sample energy system scheduling strategy is determined based on the sample output data of the strategy optimization model. Based on the state transition results corresponding to the sample energy system scheduling strategy, the evaluation scores of the sample energy system scheduling strategy on multiple indicators are determined. The evaluation scores of the multiple indicators are weighted and fused to obtain the objective function value; The policy optimization model is updated based on the objective function value.

4. The method for scheduling a comprehensive energy system in a park based on time-series feature extraction and strategy optimization according to claim 3, characterized in that, The indicators include a first indicator, which includes the park's energy cost, carbon emissions, and temperature deviation.

5. The method for scheduling a comprehensive energy system in a park based on time-series feature extraction and strategy optimization according to claim 3, characterized in that, The indicator also includes a second indicator, which reflects the degree to which the sample energy system scheduling strategy violates the constraints of the energy system. The constraints of the energy system include comfort constraints, electrical energy supply and demand balance constraints, thermal energy supply and demand balance constraints, and operational constraints of multiple devices in the energy system.

6. The method for scheduling a comprehensive energy system in a park based on time-series feature extraction and strategy optimization according to claim 1, characterized in that, The time-series feature extraction model is trained based on multiple sets of training data. Each set of training data includes sample park energy status data and the state prediction label corresponding to the sample park energy status data. The training process of the time-series feature extraction model includes: Obtain energy status data of the sample park, input the energy status data of the sample park into the time series feature extraction model, and obtain the sample time series features output by the time series feature extraction model; The temporal features of the sample are input into the prediction model to obtain the state prediction result output by the prediction model; Based on the state prediction results and the state prediction labels corresponding to the energy state data of the sample park, the training loss is determined. The temporal feature extraction model is updated based on the training loss. Before inputting the energy status data of the sample park into the time-series feature extraction model, the process includes: performing dynamic normalization processing on the energy status data of the sample park.

7. A scheduling device for a park integrated energy system based on time-series feature extraction and strategy optimization, characterized in that, The device includes: The historical data acquisition module is used to acquire park energy status data at multiple historical moments. The park energy status data reflects the output of clean energy in the park and the energy load in the park. The energy load includes electrical load and thermal load. The time-series feature extraction module is used to input the park energy status data of the multiple historical moments into the trained time-series feature extraction model to obtain the time-series features output by the time-series feature extraction model. The scheduling strategy output module is used to input input data into the strategy optimization model, and determine the energy system scheduling strategy based on the output data of the strategy optimization model. The input data includes the time series characteristics and the current energy status data of the park. The energy system scheduling strategy includes the demand response execution amount and equipment adjustment amount in the park energy system.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the integrated energy system scheduling method for parks based on time-series feature extraction and strategy optimization as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the integrated energy system scheduling method for parks based on time-series feature extraction and strategy optimization as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the integrated energy system scheduling method for parks based on time-series feature extraction and strategy optimization as described in any one of claims 1 to 6.

Images