A comprehensive energy optimization scheduling method for industrial parks based on artificial intelligence

Abstract

The present invention discloses an artificial intelligence-based integrated energy optimization and scheduling method for a park, which relates to the technical field of energy scheduling. The present invention adopts a Temporal Fusion Transformer model to perform time series modeling on the electric, thermal, cooling loads and wind and solar power generation output in the park, captures the load change trend, generates a probability distribution of prediction errors, dynamically adapts to the load characteristics and wind and solar power generation volatility in the park, and significantly reduces the prediction error; in addition, combined with real-time monitoring and event-driven mechanisms, it quickly responds to abnormal loads or equipment status, dynamically adjusts energy storage discharge, starts backup equipment, or modifies load plans to ensure stable system operation; real-time energy evaluation and trading further efficiently utilize surplus energy, and can quickly make up for energy gaps, enhancing the flexibility and economy of the park's integrated energy management.

Description

Technical Field

[0001] The present invention relates to the field of energy scheduling technology, and in particular to an artificial intelligence-based integrated energy optimization scheduling method for a park. Background Art

[0002] With the diversified development of energy demand in industrial parks, energy systems have gradually integrated multiple energy forms such as electricity, heat, cooling, and gas. The optimized scheduling has become a key means to improve energy utilization efficiency and reduce operating costs. However, the energy load of industrial parks is highly dynamic and easily affected by various factors such as season, weather, and production activities. In addition, the volatility of renewable energy such as wind and solar energy further increases the complexity of scheduling. How to make full use of renewable energy and optimize energy flow during periods of energy surplus or peak load has become a core problem facing the current integrated energy scheduling.

[0003] Traditional campus energy scheduling methods primarily employ linear programming (LP), mixed integer linear programming (MILP), or heuristic algorithms to optimize for reducing operating costs, balancing loads, and improving renewable energy utilization. These methods perform offline forecasts of historical load data and renewable energy output power to develop time-segmented scheduling plans. At the same time, they utilize peak-shaving and valley-filling strategies to guide energy storage facilities to store energy during off-peak periods and release energy during peak periods, thereby smoothing out load fluctuations. However, these solutions rely too heavily on fixed load patterns and static optimization, lacking the ability to cope with sudden load fluctuations. Furthermore, the energy trading mechanism is relatively independent of internal scheduling, making it difficult to achieve real-time and efficient energy sharing among multiple entities. Therefore, a more dynamic and flexible integrated campus energy optimization scheduling method is urgently needed to address these issues. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] The present invention provides an artificial intelligence-based integrated energy optimization and scheduling method for a park to solve the problem that traditional energy scheduling solutions rely heavily on load forecasting and are difficult to deal with real-time scheduling deviations caused by forecast errors; peak shaving and valley filling strategies rely too much on energy storage facilities and cannot fully utilize other load resources.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] The embodiment of the present invention provides a park comprehensive energy optimization scheduling method based on artificial intelligence, which includes:

[0008] Step S1: Within the park, IoT sensors and edge computing devices are used to monitor the load data and operating status of energy equipment in real time. The Temporal Fusion Transformer model, a deep learning model with time-domain fusion, is used to predict the changing trends of electricity, heating, and cooling loads within the park, as well as the output fluctuations of wind and solar power generation. A probability distribution of the prediction error is also generated.

[0009] Step S2: Based on the load forecast results of step S1 and the operating status of the equipment in the park, an optimization model is constructed to generate a scheduling strategy including the energy storage charging and discharging sequence and the electric refrigeration machine startup plan;

[0010] Step S3: Based on the scheduling strategy, an event-driven mechanism is introduced to perform real-time response;

[0011] Step S4: After the real-time response is completed, the adjusted park load data and equipment operating status are used as input to evaluate the surplus energy or energy gap; based on the evaluation results, the energy in the park is traded through the blockchain-based power trading platform.

[0012] As a preferred solution of the artificial intelligence-based integrated energy optimization and scheduling method for a park described in the present invention, the step of using the temporal fusion deep learning Temporal Fusion Transformer model to predict the changing trends of electricity, heat, and cooling loads in the park and the output fluctuations of wind and solar power generation is as follows:

[0013] The load data, historical data of wind and solar power generation output and external characteristics are organized into a time series input matrix. The input matrix is: Where X is the input multivariate time series matrix, is the i-th eigenvalue at time t, T is the time step, and n is the feature dimension;

[0014] Define the target variable and the definition formula is: Where Y is the target output sequence, including the electrical load heat load Cooling load and wind and solar power generation

[0015] The data matrix X is divided into a historical window and a prediction window and fed into the TFT model. The historical window is represents the input feature value from time point tL to t, where L is the length of the historical window and the prediction window is the target value used to predict the future time step F.

[0016] As a preferred solution of the artificial intelligence-based park integrated energy optimization scheduling method described in the present invention, the architecture of the temporal fusion deep learning Temporal Fusion Transformer model includes an encoder part and a decoder part.

[0017] Encoder part: The encoder extracts the key features of the input sequence through a multi-head attention mechanism. The extraction formula is:

[0018] H enc =Encoder(X; Θ enc ),

[0019] Among them, H enc is the feature representation matrix output by the encoder,

[0020] Θ enc is the learnable parameter of the encoder;

[0021] Decoder part: The decoder combines historical sequences and external features to generate prediction values. The generation formula is:

[0022] H dec =Decoder(H enc ,Z;Θ dec ),

[0023] Among them, H dec is the feature representation matrix generated by the decoder, Z is the external feature,

[0024] Θ dec are the learnable parameters of the decoder.

[0025] As a preferred solution of the artificial intelligence-based integrated energy optimization and scheduling method for a park described in the present invention, the step of using the temporal fusion deep learning Temporal Fusion Transformer model to predict the change trend of electricity, heat, and cooling loads in the park and the output fluctuation of wind and solar power generation also includes:

[0026] The predicted value is mapped to the target variable through the fully connected layer. The mapping formula is:

[0027]

[0028] in, is the predicted value of the jth target at time t+f, W h is the weight matrix of the fully connected layer, b h is the bias vector of the fully connected layer;

[0029] The normal distribution function is used to model the prediction error. The modeling formula is:

[0030]

[0031] in,

[0032] is the conditional probability between the target value and the predicted value, is the variance of the prediction error of target j;

[0033] By maximizing the likelihood function, the parameters of the model are optimized. The optimization formula is:

[0034]

[0035] TFT model parameters are optimized by back propagation and gradient descent so that the predicted value Close to the true value The final output is:

[0036] Future time step predictions of electricity, heating, cooling loads, and wind and solar power generation

[0037] and the corresponding prediction error probability distribution parameters.

[0038] As a preferred solution of the artificial intelligence-based integrated energy optimization scheduling method for a park described in the present invention, the steps of constructing an optimization model based on the load forecast result of step S1 and the operating status of the equipment in the park to generate a scheduling strategy including the energy storage charging and discharging sequence and the electric refrigeration machine startup plan are as follows:

[0039] The objective function is defined as minimizing the total operating cost of the park while satisfying equipment operation constraints and load balance. The operating cost includes electricity purchase cost, equipment operation cost and penalty for insufficient utilization of renewable energy.

[0040] The objective function is:

[0041]

[0042] Among them, F is the total operating cost, T is the total number of optimization periods, C grid,t is the electricity purchase cost in period t, C op,t is the operating cost of the equipment in period t, P penalty,t is the penalty for not fully utilizing renewable energy in period t;

[0043] Calculate the cost of purchasing electricity using the following formula:

[0044]

[0045] in,

[0046] p grid,t is the electricity purchase price in period t; is the power purchased from the grid during period t;

[0047] Calculate the equipment operating cost using the following formula:

[0048]

[0049] in,

[0050] c charge is the charging cost coefficient of the energy storage equipment, c discharge is the discharge cost coefficient of the energy storage equipment, is the charging power of the energy storage device in time period t, is the discharge power of the energy storage device in time period t, c cooling is the unit power operating cost of the electric refrigerator, P cool,t is the operating power of the electric refrigerator in time period t;

[0051] A penalty coefficient is set for unused renewable energy. The penalty coefficient setting formula is:

[0052]

[0053] Among them, p penalty is the penalty coefficient for not utilizing renewable energy, P ren,t is the renewable energy generation power in period t, is the renewable energy power actually used during period t.

[0054] As a preferred solution of the artificial intelligence-based integrated energy optimization scheduling method for a park described in the present invention, the step of constructing an optimization model based on the load forecast result of step S1 and the operating status of the equipment in the park to generate a scheduling strategy including energy storage charging and discharging timing and electric refrigeration machine startup plan also includes:

[0055] Add constraints to the optimization model, including,

[0056] Power balance constraint, the constraint formula is:

[0057]

[0058] Among them, P load,t is the total load of the park in period t,

[0059] Energy storage charging and discharging constraints, the constraint formula is:

[0060]

[0061] in, is the maximum charging and discharging power of the energy storage device,

[0062] Energy storage energy limit, the constraint formula is:

[0063]

[0064] Among them, E bat,t is the remaining energy of the energy storage device in time period t, is the minimum and maximum energy of the energy storage device, η ch ,η dis is the charge and discharge efficiency, Δt is the time step,

[0065] The operating constraints of the electric refrigerator are as follows:

[0066] P cool,t =L cool,t ,

[0067] Among them, L cool,t is the cooling load demand in period t;

[0068] The second generation of non-dominated sorting genetic algorithm NSGAII and distributed gradient descent algorithm are used to solve the optimal value of the objective function min F while satisfying the constraints. The output includes:

[0069] The charging and discharging power of the energy storage device in time period t The operating power P of the electric refrigerator in time period t cool,t .

[0070] As a preferred solution of the park comprehensive energy optimization scheduling method based on artificial intelligence described in the present invention, wherein: the event-driven mechanism is,

[0071] When IoT sensors or edge computing devices detect abnormalities in the operating status of loads or equipment, an event-driven mechanism is triggered to make adjustments, including dynamically starting backup equipment, adjusting energy storage discharge power, or modifying load operation plans.

[0072] As a preferred solution of the artificial intelligence-based park comprehensive energy optimization scheduling method described in the present invention, the step of introducing an event-driven mechanism for real-time response based on the scheduling strategy is as follows:

[0073] The event-driven mechanism uses real-time monitoring data from IoT sensors and edge computing devices as input. When abnormal load or equipment operating status is detected, it dynamically adjusts the energy storage discharge power, activates backup equipment, or modifies the load operation plan.

[0074] The variables in the park are collected in real time through IoT sensors, and the collected data set is expressed as:

[0075]

[0076] Among them, M t is the equipment operation status and load information set of time period t, P load,t is the current park load, E bat,t is the remaining energy of the energy storage device, is the renewable energy power currently used, S dev,t Indicates the operating status of the equipment;

[0077] Define anomaly detection rules, including load surge anomalies and insufficient energy storage capacity,

[0078] Abnormal load surge:

[0079] δP load,t =P load,t -P load,t-1 ,if Then trigger the event, where is the abnormal threshold of load change;

[0080] Abnormal insufficient energy storage:

[0081] if, Then trigger the event, where Energy threshold for energy storage alarm;

[0082] When an anomaly is detected, an event-driven mechanism is triggered to adjust the scheduling strategy.

[0083] As a preferred solution of the park comprehensive energy optimization scheduling method based on artificial intelligence described in the present invention, the scheduling strategy includes:

[0084] Dynamically adjust energy storage discharge power:

[0085] When the load suddenly increases, When , energy is released through the energy storage system, and the release adjustment formula is:

[0086]

[0087] in, is the adjusted energy storage discharge power, Δt is the time step, is the maximum discharge power of the energy storage device;

[0088] Start the backup device:

[0089] When the stored energy is insufficient, When renewable energy cannot meet the load, the backup equipment is started and the power of the backup equipment is adjusted. The adjustment formula is:

[0090]

[0091] Among them, Pbackup,t is the output power of the standby equipment, is the maximum output power of the standby equipment,

[0092] To purchase electricity from the grid;

[0093] Modify the load operation plan:

[0094] When the load exceeds the total energy supply capacity of the equipment, load reduction is performed. The reduction formula is:

[0095]

[0096] in, is the adjusted total load, P noncritical,t is the reduction amount of non-critical load in period t;

[0097] In step S3, after the dynamic adjustment is completed, the scheduling strategy is updated, including the energy storage power, the start and stop status of the backup equipment, and the load reduction plan, and is updated to:

[0098]

[0099] in, is the adjusted equipment operation status set;

[0100] The adjusted operation plan The data is sent to the device, and the edge computing device provides real-time feedback on the execution results and continues to monitor the operating status.

[0101] As a preferred solution of the artificial intelligence-based park comprehensive energy optimization scheduling method described in the present invention, the step of using the adjusted park load data and equipment operating status as input to evaluate the surplus energy or energy gap is as follows:

[0102] The adjusted key data are summarized into a state vector, which is expressed as:

[0103]

[0104] in, is the total load of the park after adjustment during period t, is the adjusted energy storage discharge power, P backup,t is the output power of the standby equipment, is the actual renewable energy power used, E bat,t The remaining energy after adjustment of the energy storage device;

[0105] Calculate the real-time total energy demand using the following formula:

[0106]

[0107] Among them, P total,t is the real-time total energy demand in period t, P reserve,t For spare capacity requirements;

[0108] Calculate the spare capacity requirement using the following formula:

[0109]

[0110] Where α is the safety factor,

[0111] is the standard deviation of the adjusted afterload,

[0112] Calculate the real-time total available energy, which includes:

[0113] Renewable energy

[0114]

[0115] Among them, P ren,t is the total power generation of renewable energy in period t,

[0116] Energy storage output

[0117]

[0118] Backup device output

[0119]

[0120] Calculate the total available energy P avail,t , the calculation formula is:

[0121]

[0122] If the total available energy exceeds the total demand, it is considered as surplus energy. The surplus energy calculation formula is:

[0123] P surplus,t =P avail,t -P total,t ,

[0124] Among them, P surplus,t is the residual energy in time period t,

[0125] Residual energy ratio R surplus,t The calculation formula is:

[0126]

[0127] If the total demand exceeds the total available energy, it is considered that there is an energy gap. The energy gap calculation formula is:

[0128] P deficit,t =P total,t -P avail,t ,

[0129] Among them, P deficit,t is the energy gap in time period t,

[0130] Gap degree R deficit,t The calculation formula is:

[0131]

[0132] The beneficial effects of the present invention are as follows: the present invention adopts the Temporal Fusion Transformer model to perform time series modeling of the electricity, heat, cooling loads and wind and solar power generation output in the park, captures the load change trend, generates the probability distribution of the prediction error, dynamically adapts to the load characteristics and wind and solar power generation volatility in the park, and significantly reduces the prediction error; in addition, combined with real-time monitoring and event-driven mechanisms, it quickly responds to abnormal load or equipment status, dynamically adjusts energy storage discharge, starts backup equipment or modifies load plans to ensure stable operation of the system; real-time energy evaluation and trading further make efficient use of surplus energy, and can quickly make up for energy gaps, enhancing the flexibility and economy of the park's comprehensive energy management. BRIEF DESCRIPTION OF THE DRAWINGS

[0133] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0134] Figure 1 This is a schematic flow chart of the park comprehensive energy optimization scheduling method in Example 1. DETAILED DESCRIPTION

[0135] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0136] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0137] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0138] Example 1, with reference to Figure 1 This embodiment provides an artificial intelligence-based integrated energy optimization and scheduling method for a park, including:

[0139] Step S1: Within the park, IoT sensors and edge computing devices are used to monitor the load data and operating status of energy equipment in real time. The Temporal Fusion Transformer model, a deep learning model with time-domain fusion, is used to predict the changing trends of electricity, heating, and cooling loads within the park, as well as the output fluctuations of wind and solar power generation. A probability distribution of the prediction error is also generated.

[0140] The steps for using the Temporal Fusion Transformer model to predict the changing trends of electricity, heating, and cooling loads in the park and the output fluctuations of wind and solar power generation are as follows:

[0141] The load data, historical data of wind and solar power generation output and external characteristics are organized into a time series input matrix. The input matrix is: Where X is the input multivariate time series matrix, is the i-th eigenvalue at time t, T is the time step, and n is the feature dimension;

[0142] Define the target variable and the definition formula is: Where Y is the target output sequence, including the electrical load heat load Cooling load and wind and solar power generation

[0143] The data matrix X is divided into a historical window and a prediction window and fed into the TFT model. The historical window is represents the input feature value from time point tL to t, where L is the length of the historical window and the prediction window is the target value used to predict the future time step F.

[0144] The architecture of the temporal fusion deep learning TemporalFusionTransformer model includes an encoder part and a decoder part.

[0145] Encoder part: The encoder extracts the key features of the input sequence through a multi-head attention mechanism. The extraction formula is:

[0146] H enc =Encoder(X; Θ enc ),

[0147] Among them, H enc is the feature representation matrix output by the encoder,

[0148] Θ enc is the learnable parameter of the encoder;

[0149] Decoder part: The decoder combines historical sequences and external features to generate prediction values. The generation formula is:

[0150] H dec =Decoder(H enc ,Z;Θ dec ),

[0151] Among them, H dec is the feature representation matrix generated by the decoder, Z is the external feature,

[0152] Θ dec are the learnable parameters of the decoder;

[0153] The steps of using the TemporalFusion Transformer model to predict the changing trends of electricity, heating, and cooling loads in the park and the output fluctuations of wind and solar power generation also include:

[0154] The predicted value is mapped to the target variable through the fully connected layer. The mapping formula is:

[0155]

[0156] in, is the predicted value of the jth target at time t+f, W h is the weight matrix of the fully connected layer, b h is the bias vector of the fully connected layer;

[0157] The normal distribution function is used to model the prediction error. The modeling formula is:

[0158]

[0159] in,

[0160] is the conditional probability between the target value and the predicted value, is the variance of the prediction error of target j;

[0161] By maximizing the likelihood function, the parameters of the model are optimized. The optimization formula is:

[0162]

[0163] TFT model parameters are optimized by back propagation and gradient descent so that the predicted value Close to the true value The final output is:

[0164] Future time step predictions of electricity, heating, cooling loads, and wind and solar power generation

[0165] and the corresponding prediction error probability distribution parameters;

[0166] Specifically, through the TFT model, combined with historical data and external characteristics, the load change trend and wind and solar power generation output of the park's energy system are predicted, and the probability distribution of prediction errors is generated at the same time, providing robust support for optimized scheduling; improving the real-time scheduling and better adapting to the dynamic needs of the park.

[0167] Step S2: Based on the load forecast results of step S1 and the operating status of the equipment in the park, an optimization model is constructed to generate a scheduling strategy including the energy storage charging and discharging sequence and the electric refrigeration machine startup plan;

[0168] Based on the load forecast results of step S1 and the operating status of the equipment in the park, the steps of constructing an optimization model and generating a scheduling strategy including the energy storage charging and discharging sequence and the electric refrigeration machine start-up plan are as follows:

[0169] The objective function is defined as minimizing the total operating cost of the park while satisfying equipment operation constraints and load balance. The operating cost includes electricity purchase cost, equipment operation cost and penalty for insufficient utilization of renewable energy.

[0170] The objective function is:

[0171]

[0172] Among them, F is the total operating cost, T is the total number of optimization periods, C grid,t is the electricity purchase cost in period t, C op,t is the operating cost of the equipment in period t, P penalty,t is the penalty for not fully utilizing renewable energy in period t;

[0173] Calculate the cost of purchasing electricity using the following formula:

[0174]

[0175] in,

[0176] p grid,t is the electricity purchase price in period t; is the power purchased from the grid during period t;

[0177] Calculate the equipment operating cost using the following formula:

[0178]

[0179] in,

[0180] c charge is the charging cost coefficient of the energy storage equipment, c discharge is the discharge cost coefficient of the energy storage equipment, is the charging power of the energy storage device in time period t, is the discharge power of the energy storage device in time period t, c cooling is the unit power operating cost of the electric refrigerator, P cool,t is the operating power of the electric refrigerator in time period t;

[0181] A penalty coefficient is set for unused renewable energy. The penalty coefficient setting formula is:

[0182]

[0183] Among them, p penalty is the penalty coefficient for not utilizing renewable energy, P ren,t is the renewable energy generation power in period t, is the renewable energy power actually used in period t;

[0184] Based on the load forecast results of step S1 and the operating status of the equipment in the park, the steps of constructing an optimization model and generating a scheduling strategy including energy storage charging and discharging timing and electric refrigeration machine startup plan also include:

[0185] Add constraints to the optimization model, including,

[0186] Power balance constraint, the constraint formula is:

[0187]

[0188] Among them, P load,t is the total load of the park in period t,

[0189] Energy storage charging and discharging constraints, the constraint formula is:

[0190]

[0191] in, is the maximum charging and discharging power of the energy storage device,

[0192] Energy storage energy limit, the constraint formula is:

[0193]

[0194] Among them, E bat,t is the remaining energy of the energy storage device in time period t, is the minimum and maximum energy of the energy storage device, η ch ,η dis is the charge and discharge efficiency, Δt is the time step,

[0195] The operating constraints of the electric refrigerator are as follows:

[0196] P cool,t =L cool,t ,

[0197] Among them, L cool,t is the cooling load demand in period t;

[0198] The second generation of non-dominated sorting genetic algorithm NSGAII and distributed gradient descent algorithm are used to solve the optimal value of the objective function min F while satisfying the constraints. The output includes:

[0199] The charging and discharging power of the energy storage device in time period t The operating power P of the electric refrigerator in time period t cool,t ;

[0200] Specifically, through a dynamic optimization model based on load forecast results, the energy storage charging and discharging timing and electric refrigeration machine start-up plan are generated to minimize costs and maximize the utilization of renewable energy.

[0201] Step S3: Based on the scheduling strategy, an event-driven mechanism is introduced to perform real-time response;

[0202] The event-driven mechanism is,

[0203] When IoT sensors or edge computing devices detect abnormal load or equipment operating status, event-driven mechanisms are triggered to make adjustments, including dynamically starting backup equipment, adjusting energy storage discharge power, or modifying load operation plans.

[0204] Based on the scheduling strategy, the steps to introduce the event-driven mechanism for real-time response are as follows:

[0205] The event-driven mechanism uses real-time monitoring data from IoT sensors and edge computing devices as input. When abnormal load or equipment operating status is detected, it dynamically adjusts the energy storage discharge power, activates backup equipment, or modifies the load operation plan.

[0206] The variables in the park are collected in real time through IoT sensors, and the collected data set is expressed as:

[0207]

[0208] Among them, Mt is the equipment operation status and load information set of time period t, P load,t is the current park load, E bat,t is the remaining energy of the energy storage device, is the renewable energy power currently used, S dev,t Indicates the operating status of the equipment;

[0209] Define anomaly detection rules, including load surge anomalies and insufficient energy storage capacity,

[0210] Abnormal load surge:

[0211] δP load,t =P load,t -P load,t-1 ,if Then trigger the event, where is the abnormal threshold of load change;

[0212] Abnormal insufficient energy storage:

[0213] if, Then trigger the event, where Energy threshold for energy storage alarm;

[0214] When an anomaly is detected, the event-driven mechanism is triggered to adjust the scheduling strategy;

[0215] Scheduling strategies include:

[0216] Dynamically adjust energy storage discharge power:

[0217] When the load suddenly increases, When , energy is released through the energy storage system, and the release adjustment formula is:

[0218]

[0219] in, is the adjusted energy storage discharge power, Δt is the time step, is the maximum discharge power of the energy storage device;

[0220] Start the backup device:

[0221] When the stored energy is insufficient, When renewable energy cannot meet the load, the backup equipment is started and the power of the backup equipment is adjusted. The adjustment formula is:

[0222]

[0223] Among them, P backup,t is the output power of the standby equipment, is the maximum output power of the standby equipment,

[0224] To purchase electricity from the grid;

[0225] Modify the load operation plan:

[0226] When the load exceeds the total energy supply capacity of the equipment, load reduction is performed. The reduction formula is:

[0227]

[0228] in, is the adjusted total load, P noncritical,t is the reduction amount of non-critical load in period t;

[0229] In step S3, after the dynamic adjustment is completed, the scheduling strategy is updated, including the energy storage power, the start and stop status of the backup equipment, and the load reduction plan, and is updated to:

[0230]

[0231] in, is the adjusted equipment operation status set;

[0232] The adjusted operation plan The data is sent to the device, and the edge computing device provides real-time feedback on the execution results and continues to monitor the operating status.

[0233] Specifically, the event-driven mechanism monitors the load and equipment status within the park in real time, and triggers dynamic adjustment strategies after detecting anomalies, including energy storage power adjustment, backup equipment startup, and load reduction, to effectively respond to load fluctuations and operational anomalies.

[0234] Step S4: After the real-time response is completed, the adjusted park load data and equipment operating status are used as input to evaluate the surplus energy or energy gap; based on the evaluation results, energy within the park is traded through the blockchain-based power trading platform;

[0235] Taking the adjusted park load data and equipment operating status as input, the steps to evaluate the surplus energy or energy gap are as follows:

[0236] The adjusted key data are summarized into a state vector, which is expressed as:

[0237]

[0238] in, is the total load of the park after adjustment during period t, is the adjusted energy storage discharge power, P backup,t is the output power of the standby equipment, is the actual renewable energy power used, E bat,tThe remaining energy after adjustment of the energy storage device;

[0239] Calculate the real-time total energy demand using the following formula:

[0240]

[0241] Among them, P total,t is the real-time total energy demand in period t, P reserve,t For spare capacity requirements;

[0242] Calculate the spare capacity requirement using the following formula:

[0243]

[0244] Where α is the safety factor,

[0245] is the standard deviation of the adjusted afterload,

[0246] Calculate the real-time total available energy, which includes:

[0247] Renewable energy

[0248]

[0249] Among them, P ren,t is the total power generation of renewable energy in period t,

[0250] Energy storage output

[0251]

[0252] Backup device output

[0253]

[0254] Calculate the total available energy P avail,t , the calculation formula is:

[0255]

[0256] If the total available energy exceeds the total demand, it is considered as surplus energy. The surplus energy calculation formula is:

[0257] P surplus,t =P avail,t -P total,t ,

[0258] Among them, P surplus,t is the residual energy in time period t,

[0259] Residual energy ratio R surplus,tThe calculation formula is:

[0260]

[0261] If the total demand exceeds the total available energy, it is considered that there is an energy gap. The energy gap calculation formula is:

[0262] P deficit,t =P total,t -P avail,t ,

[0263] Among them, P deficit,t is the energy gap in time period t,

[0264] Gap degree R deficit,t The calculation formula is:

[0265]

[0266] Specifically, by evaluating the adjusted park load and equipment operating status in real time, the balance between available energy and demand is calculated, and the surplus energy or energy gap is quantified respectively.

[0267] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A comprehensive energy optimization and scheduling method for a park based on artificial intelligence, characterized by: include, Step S1: Within the park, IoT sensors and edge computing devices are used to monitor the load data and operating status of energy equipment in real time. The Temporal Fusion Transformer model, a deep learning model with time-domain fusion, is used to predict the changing trends of electricity, heating, and cooling loads within the park, as well as the output fluctuations of wind and solar power generation. A probability distribution of the prediction error is also generated. Step S2: Based on the load forecast results of step S1 and the operating status of the equipment in the park, an optimization model is constructed to generate a scheduling strategy including the energy storage charging and discharging sequence and the electric refrigeration machine startup plan; Step S3: Based on the scheduling strategy, an event-driven mechanism is introduced to perform real-time response; Step S4: After the real-time response is completed, the adjusted park load data and equipment operating status are used as input to evaluate the surplus energy or energy gap; Based on the evaluation results, energy within the park will be traded through a blockchain-based power trading platform; The steps of using the Temporal Fusion Transformer model to predict the load trends of electricity, heat and cooling loads in the park and the output fluctuations of wind and solar power generation include: The predicted value is mapped to the target variable through the fully connected layer. The mapping formula is: in, is the predicted value of the jth target at time t+f, W h is the weight matrix of the fully connected layer, H dec The feature representation matrix generated by the decoder, b h is the bias vector of the fully connected layer; The normal distribution function is used to model the prediction error. The modeling formula is: in, is the conditional probability between the target value and the predicted value, is the variance of the prediction error of target j; By maximizing the likelihood function, the parameters of the model are optimized. The optimization formula is: TFT model parameters are optimized by back propagation and gradient descent so that the predicted value Close to the true value The final output is: Future time step predictions of electricity, heating, cooling loads, and wind and solar power generation and the corresponding prediction error probability distribution parameters; The steps of constructing an optimization model based on the load forecast result of step S1 and the operating status of the equipment in the park to generate a scheduling strategy including the energy storage charging and discharging sequence and the electric refrigeration machine startup plan are as follows: The objective function is defined as minimizing the total operating cost of the park while satisfying equipment operation constraints and load balance. The operating cost includes electricity purchase cost, equipment operation cost and penalty for insufficient utilization of renewable energy. The objective function is: Among them, F is the total operating cost, T opt is the total number of optimization periods, C grid,t is the electricity purchase cost in period t, C op,t is the operating cost of the equipment in period t, P penalty,t is the penalty for not fully utilizing renewable energy in period t; Calculate the cost of purchasing electricity using the following formula: Among them, p grid,t is the electricity purchase price in period t; is the power purchased from the grid during period t; Calculate the equipment operating cost using the following formula: Among them, c charge is the charging cost coefficient of the energy storage equipment, c discharge is the discharge cost coefficient of the energy storage equipment, is the charging power of the energy storage device in time period t, is the discharge power of the energy storage device in time period t, c cooling is the unit power operating cost of the electric refrigerator, P cool,t is the operating power of the electric refrigerator in time period t; A penalty coefficient is set for unused renewable energy. The penalty coefficient setting formula is: Among them, p penalty is the penalty coefficient for not utilizing renewable energy, P ren,t is the renewable energy generation power in period t, is the renewable energy power actually used during period t.

2. The method for optimizing and scheduling integrated energy resources in a park based on artificial intelligence according to claim 1, wherein: The steps of using the TemporalFusion Transformer model to predict the load changes of the electricity, heating and cooling loads in the park and the output fluctuations of wind and solar power generation are as follows: The load data, historical data of wind and solar power generation output and external characteristics are organized into a time series input matrix. The input matrix is: Where X is the input multivariate time series matrix, is the i-th eigenvalue at time t, T is the time step, and n is the feature dimension; Define the target variable and the definition formula is: Where Y is the target output sequence, including the electrical load heat load Cooling load and wind and solar power generation The data matrix X is divided into a historical window and a prediction window and fed into the TFT model. The historical window is represents the input feature value from time point tL to t, where L is the length of the historical window and the prediction window is the target value used to predict the future time step F.

3. The method for optimizing and scheduling integrated energy resources in a park based on artificial intelligence according to claim 2, wherein: The architecture of the temporal fusion deep learning TemporalFusion Transformer model includes an encoder part and a decoder part. Encoder part: The encoder extracts the key features of the input sequence through a multi-head attention mechanism. The extraction formula is: H enc =Encoder(X;Θ enc ), Among them, H enc is the feature representation matrix output by the encoder, Θ enc is the learnable parameter of the encoder; Decoder part: The decoder combines historical sequences and external features to generate prediction values. The generation formula is: H dec =Decoder(H enc ,Z;Θ dec ), Among them, H dec is the feature representation matrix generated by the decoder, Z is the external feature, Θ dec are the learnable parameters of the decoder.

4. The method for optimizing and scheduling integrated energy resources in a park based on artificial intelligence according to claim 3, wherein: The step of constructing an optimization model based on the load forecast result of step S1 and the operating status of the equipment in the park to generate a scheduling strategy including the energy storage charging and discharging sequence and the electric refrigeration machine startup plan also includes: Add constraints to the optimization model, including, Power balance constraint, the constraint formula is: Among them, P load,t is the total load of the park in period t, Energy storage charging and discharging constraints, the constraint formula is: in, is the maximum charging and discharging power of the energy storage device, Energy storage energy limit, the constraint formula is: Among them, E bat,t is the remaining energy of the energy storage device in time period t, is the minimum and maximum energy of the energy storage device, η ch ,η dis is the charge and discharge efficiency, Δt is the time step, The operating constraints of the electric refrigerator are as follows: P cool,t =L cool,t , Among them, L cool,t is the cooling load demand in period t; The second generation of non-dominated sorting genetic algorithm NSGAII and distributed gradient descent algorithm are used to solve the optimal value of the objective function minF while satisfying the constraints. The output includes: The charging and discharging power of the energy storage device in time period t The operating power P of the electric refrigerator in time period t cool,t .

5. The method for optimizing and scheduling integrated energy resources in a park based on artificial intelligence according to claim 4, characterized in that: The event-driven mechanism is: When IoT sensors or edge computing devices detect abnormalities in the operating status of loads or equipment, an event-driven mechanism is triggered to make adjustments, including dynamically starting backup equipment, adjusting energy storage discharge power, or modifying load operation plans.

6. The method for optimizing and scheduling integrated energy resources in a park based on artificial intelligence according to claim 5, characterized in that: The steps of introducing an event-driven mechanism for real-time response based on the scheduling strategy are: The variables in the park are collected in real time through IoT sensors, and the collected data set is expressed as: Among them, M t is the equipment operation status and load information set of time period t, P load,t is the current park load, E bat,t is the remaining energy of the energy storage device, is the renewable energy power actually used currently, S dev,t Indicates the operating status of the equipment; Define anomaly detection rules, including load surge anomalies and insufficient energy storage capacity, Abnormal load surge: δP load,t =P load,t -P load,t-1 ,if Then trigger the event, where is the abnormal threshold of load change; Abnormal insufficient energy storage: if, Then trigger the event, where Energy threshold for energy storage alarm; When an anomaly is detected, an event-driven mechanism is triggered to adjust the scheduling strategy.

7. The method for optimizing and scheduling integrated energy resources in a park based on artificial intelligence according to claim 6, characterized in that: The scheduling strategy includes: Dynamically adjust energy storage discharge power: When the load suddenly increases, When , energy is released through the energy storage system, and the release adjustment formula is: in, is the adjusted energy storage discharge power, Δt is the time step, is the maximum discharge power of the energy storage device; Start the backup device: When the stored energy is insufficient, When renewable energy cannot meet the load, the backup equipment is started and the power of the backup equipment is adjusted. The adjustment formula is: Among them, P backup,t is the output power of the standby equipment, is the maximum output power of the standby equipment, To purchase electricity from the grid; Modify the load operation plan: When the load exceeds the total energy supply capacity of the equipment, load reduction is performed. The reduction formula is: in, is the adjusted total load, P noncritical,t is the reduction amount of non-critical load in period t; In step S3, after the dynamic adjustment is completed, the scheduling strategy is updated, including the energy storage power, the start and stop status of the backup equipment, and the load reduction plan, and is updated to: in, is the adjusted equipment operation status set; The adjusted operation plan The data is sent to the device, and the edge computing device provides real-time feedback on the execution results and continues to monitor the operating status.

8. The method for optimizing and scheduling integrated energy resources in a park based on artificial intelligence according to claim 7, characterized in that: The adjusted park load data and equipment operating status are used as input to evaluate the residual The steps to fill the energy gap are: The adjusted key data are summarized into a state vector, which is expressed as: in, is the total load of the park after adjustment during period t, is the adjusted energy storage discharge power, P backup,t is the output power of the standby equipment, is the actual renewable energy power used, E bat,t The remaining energy after adjustment of the energy storage device; Calculate the real-time total energy demand using the following formula: Among them, P total,t is the real-time total energy demand in period t, P reserve,t is the spare capacity requirement; the spare capacity requirement is calculated using the following formula: Where α is the safety factor, is the standard deviation of the adjusted afterload, Calculate the real-time total available energy, which includes: Renewable energy Among them, P ren,t is the total power generation of renewable energy in period t, Energy storage output Backup device output Calculate the total available energy P avail,t , the calculation formula is: If the total available energy exceeds the total demand, it is considered as surplus energy. The surplus energy calculation formula is: P surplus,t =P avail,t -P total,t , Among them, P surplus,t is the residual energy in time period t, Residual energy ratio R surplus,t The calculation formula is: If the total demand exceeds the total available energy, it is considered that there is an energy gap. The energy gap calculation formula is: P deficit,t =P total,t -P avail,t , Among them, P deficit,t is the energy gap in time period t, Gap degree R deficit,t The calculation formula is:

Images