A park energy efficient scheduling calculation method fusing digital twinning

Abstract

The application discloses a kind of park energy efficient scheduling calculation methods of fusion digital twin, it is related to digital twin and park energy scheduling technical field, including obtaining the continuous running state sequence of park energy physical system in historical time period and real-time load fluctuation sequence, wherein the park energy physical system includes multiple energy equipment, according to continuous running state sequence and real-time load fluctuation sequence, using long short-term memory network extracts the time-varying operating characteristics of park energy physical system under different time dimensions, and obtains feature set;The park energy efficient scheduling calculation method of fusion digital twin improves the accuracy of supply-demand gap prediction and scheduling decision, is conducive to reducing energy operation cost, improving energy utilization efficiency, reducing supply-demand imbalance risk, and improving the reliability and efficiency of park energy scheduling.

Description

Technical Field

[0001] This invention relates to the field of digital twin and park energy scheduling technology, specifically to a park energy efficient scheduling calculation method that integrates digital twins. Background Technology

[0002] Energy dispatching is a crucial aspect of park energy management, directly impacting operating costs, energy efficiency, and supply-demand balance. As park energy structures become increasingly complex, their physical energy systems typically include various energy devices such as photovoltaics, energy storage, combined cooling, heating, and power (CCHP), air conditioning units, charging facilities, and building loads. These different devices involve the conversion, transmission, and coupling of various energy forms, including electricity, heat, and cooling. Therefore, achieving efficient and reliable energy dispatching under the condition of coordinated operation of multiple energy devices has become a critical issue in the operation and management of park energy systems.

[0003] Existing energy dispatching methods in industrial parks typically rely on human experience, historical statistical patterns, or simplified mathematical models for decision-making. These methods generally idealize the operating status of energy equipment, load changes, and energy flow relationships, making it difficult to accurately represent the dynamic changes of the park's energy physical system during continuous operation. Especially when park load fluctuates in real time due to factors such as weather, production plans, building usage status, and charging demand, traditional dispatching models struggle to capture load change patterns and equipment response characteristics across different time dimensions, leading to discrepancies between dispatching results and actual operational needs. Digital twin technology, by constructing a virtual model corresponding to the park's energy physical system, can map, simulate, and predict the operating status of the park's energy system, providing a new technical path for energy dispatching. However, existing digital twin modeling methods largely rely on manual modeling and empirical parameter setting, resulting in lengthy model construction processes and insufficient representation of multi-energy coupling relationships between different energy devices. In scenarios where multiple energy devices operate simultaneously, the energy flow relationships, operational constraints, and mutual influences between devices are complex. If the virtual model cannot accurately reflect these coupling relationships, inconsistencies between simulation results and the actual operating status of the park can easily arise.

[0004] Furthermore, the park's energy physical system exhibits significant nonlinearity, time-varying characteristics, and uncertainty. Taking the coordinated operation of energy storage, photovoltaic (PV) equipment, and combined cooling, heating, and power (CCHP) equipment as an example, the energy storage charging and discharging efficiency, PV output variations, CCHP load demand, and equipment operating efficiency all change with time and operating conditions. If existing digital twin models lack effective mechanisms for identifying and correcting model biases during operation, the virtual model may deviate from the real physical system over a long period, thus affecting the accuracy of subsequent supply-demand imbalance predictions and dispatch instructions. Moreover, park energy dispatch not only needs to meet supply-demand balance but also needs to consider goals such as reducing energy operating costs, improving energy utilization efficiency, and mitigating supply-demand imbalances. Existing dispatch methods typically lack sufficient simulation capabilities for future operating conditions, making it difficult to predict potential supply-demand imbalance changes under different dispatch strategies. When dispatch instructions fail to fully consider future load fluctuations and equipment operating status changes, problems such as energy waste, inefficient equipment operation, or localized power shortages may occur. Summary of the Invention

[0005] The purpose of this invention is to provide a computational method for efficient energy scheduling in industrial parks that integrates digital twins, thereby solving the problems existing in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a computational method for efficient energy scheduling in a park integrating digital twins, comprising: S1, acquiring the continuous operating state sequence and real-time load fluctuation sequence of the park's energy physical system, wherein the park's energy physical system includes multiple energy devices; S2, extracting the time-varying operating characteristics of the park's energy physical system to obtain a feature set; S3, identifying the multi-energy coupling relationship between different energy devices based on the feature set to obtain a multi-energy coupling feature matrix; S4, constructing an initial digital twin virtual model based on the multi-energy coupling feature matrix, the operating states of multiple energy devices, and the energy flow relationship between multiple energy devices; S5, acquiring the prediction sequence output by the initial digital twin virtual model within the current time window, comparing the prediction sequence with the actual operating sequence of the park's energy physical system within the current time window, and determining the model bias. S6. If the model deviation value is greater than the preset deviation threshold, the coupling weights of the initial digital twin virtual model are corrected to obtain the target digital twin virtual model; if the model deviation value is not greater than the preset deviation threshold, the model stability is verified, and the initial digital twin virtual model that passes the verification is determined as the target digital twin virtual model; S7. The future operating state of the park's energy physical system is simulated using the target digital twin virtual model to obtain the expected supply and demand difference sequence and generate the initial scheduling instruction; S8. The extreme value of the supply and demand difference corresponding to the initial scheduling instruction is determined; if the extreme value of the supply and demand difference is less than the preset imbalance threshold, the initial scheduling instruction is determined as the target scheduling instruction; S9. If the extreme value of the supply and demand difference is not less than the preset imbalance threshold, the initial scheduling instruction is corrected according to the expected supply and demand difference sequence and the efficient scheduling objective to obtain the target scheduling instruction.

[0007] Preferably, step S1 includes obtaining equipment operation logs to generate a continuous operating state sequence and collecting real-time load data to obtain a real-time load fluctuation sequence; aligning the continuous operating state sequence and the real-time load fluctuation sequence to obtain a state fluctuation mapping relationship; constructing an energy operation map based on the state fluctuation mapping relationship; removing isolated nodes from the energy operation map if any exist, and obtaining an updated energy operation map; and obtaining the continuous operating state sequence and real-time load fluctuation sequence of the park's energy physical system within a historical time period through the updated energy operation map.

[0008] Preferably, step S2 includes acquiring a continuous operating state sequence and a real-time load fluctuation sequence; performing synchronization alignment processing on the continuous operating state sequence and the real-time load fluctuation sequence to obtain a synchronized operating state sequence and a synchronized load fluctuation sequence; performing sequence truncation based on the synchronized operating state sequence and the synchronized load fluctuation sequence to obtain a multidimensional state subsequence and a multidimensional load subsequence; and extracting time-varying operating features from the multidimensional state subsequence and the multidimensional load subsequence using a long short-term memory network to obtain a feature set.

[0009] Preferably, step S3 includes acquiring a feature set to construct an initial feature vector; receiving the initial feature vector through a random forest model to obtain a feature contribution sequence; calculating the coupling weights between different energy devices based on the feature contribution sequence; determining the multi-energy coupling relationship between different energy devices if the coupling weights meet preset matching conditions; and performing matrix processing on the multi-energy coupling relationship to obtain a multi-energy coupling feature matrix.

[0010] Preferably, step S4 includes acquiring a multi-energy coupling feature matrix and the operating status of various energy devices. The multi-energy coupling feature matrix is ​​used to extract coupling weights, and the operating status is used to extract state mapping vectors. Energy flow relationships are determined based on the coupling weights and the state mapping vectors. It is then determined whether the coupling weights exceed a preset threshold. If the coupling weights exceed the preset threshold, flow nodes corresponding to the energy flow relationships are generated. A topology connection graph is constructed based on the flow nodes and energy flow relationships. The topology connection graph is used to generate virtual entities and mapping edges. An initial digital twin virtual model corresponding to the park's energy physical system is constructed using the virtual entities and mapping edges.

[0011] Preferably, step S5 includes acquiring the predicted sequence and the actual running sequence of the initial digital twin virtual model; processing the predicted sequence and the actual running sequence using a dynamic time warping algorithm to obtain a sequence alignment matrix; generating an error distribution vector based on the sequence alignment matrix; determining whether the error distribution vector exceeds a preset threshold; if the error distribution vector exceeds the preset threshold, extracting the cumulative deviation; and using the cumulative deviation to determine the model deviation value.

[0012] Preferably, step S6 includes determining whether the model deviation value is greater than a preset deviation threshold; if the model deviation value is greater than the preset deviation threshold, generating a weight update vector based on the energy conversion efficiency and load response gradient in the multi-energy coupling feature matrix; correcting the coupling weights of the initial digital twin virtual model through the weight update vector to obtain the target digital twin virtual model; if the model deviation value is not greater than the preset deviation threshold, determining the initial digital twin virtual model that has passed the verification using the multi-energy coupling feature matrix as the target digital twin virtual model.

[0013] Preferably, step S7 includes: simulating the park's energy physical system using a target digital twin virtual model to obtain the future operating state; extracting predicted quantities based on the future operating state to obtain an expected supply-demand gap sequence; determining whether the maximum gap value of the expected supply-demand gap sequence is greater than an imbalance threshold; if the maximum gap value is greater than the imbalance threshold, constructing an instruction optimization space based on the expected supply-demand gap sequence and the efficient scheduling objective; and generating initial scheduling instructions by searching in the instruction optimization space. The efficient scheduling objective includes at least one of reducing energy operating costs, improving energy utilization efficiency, and reducing the degree of supply-demand imbalance.

[0014] Preferably, step S8 includes parsing the expected energy supply sequence and the expected energy demand sequence according to the initial scheduling instruction; performing time-slice alignment processing on the expected energy supply sequence and the expected energy demand sequence to obtain the supply-demand difference fluctuation range; performing extreme value search on the supply-demand difference fluctuation range and removing outliers to obtain the extreme value of the supply-demand difference; determining whether the extreme value of the supply-demand difference is less than a preset imbalance threshold; if the extreme value of the supply-demand difference is less than the preset imbalance threshold, then the initial scheduling instruction is determined as the target scheduling instruction.

[0015] Preferably, step S9 includes: if the extreme value of the supply-demand difference is not less than a preset imbalance threshold, parsing the initial scheduling instruction according to the efficient scheduling objective to obtain equipment response delay and energy conversion loss; constructing an instruction correction vector based on the equipment response delay and energy conversion loss; processing the initial scheduling instruction using the instruction correction vector to obtain the correction magnitude; updating the initial scheduling instruction according to the correction magnitude to obtain the target scheduling instruction; and outputting the target scheduling instruction to control various energy devices in the park's energy physical system to perform park energy scheduling.

[0016] As can be seen from the above technical solution, the present invention has the following beneficial effects: This computational method for efficient energy dispatching in industrial parks, integrating digital twins, acquires the continuous operating state sequence and real-time load fluctuation sequence of the park's energy physical system. It then utilizes a long short-term memory network to extract time-varying operating characteristics across different time dimensions, enabling a more accurate reflection of the dynamic changes in the park's energy system during continuous operation. Simultaneously, it identifies multi-energy coupling relationships between different energy devices using random forests and constructs a digital twin virtual model by combining the multi-energy coupling feature matrix, device operating status, and energy flow relationships. This enhances the virtual model's ability to express complex coupling relationships within the park's energy physical system. Furthermore, by comparing the predicted sequence with the actual operating sequence, it determines the model deviation value and corrects or verifies the stability of the digital twin virtual model based on this deviation, allowing the target digital twin virtual model to better adapt to the nonlinearity, time-varying nature, and uncertainty of the park's energy system. Building upon this, it simulates future operating states using the target digital twin virtual model to obtain the expected supply-demand imbalance sequence. Combined with the efficient dispatching target, it generates and corrects dispatching instructions, thereby improving the accuracy of supply-demand imbalance prediction and dispatching decisions. This approach helps reduce energy operating costs, improve energy utilization efficiency, mitigate the risk of supply-demand imbalance, and enhance the reliability and efficiency of energy dispatching in the park. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a timing diagram of the present invention; Figure 3This is a schematic diagram illustrating the construction of the digital twin virtual model of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1: like Figure 1 and Figure 3 As shown, the present invention provides a technical solution: a computational method for efficient energy scheduling in a park that integrates digital twins, comprising: S1. Obtain the continuous operating status sequence and real-time load fluctuation sequence of the park's energy physical system within a historical time period, wherein the park's energy physical system includes various energy devices; S2. Based on the continuous operating state sequence and the real-time load fluctuation sequence, a long short-term memory network is used to extract the time-varying operating characteristics of the park's energy physical system under different time dimensions to obtain a feature set. S3. For the feature set, identify the multi-energy coupling relationship between different energy devices through random forest and obtain the multi-energy coupling feature matrix; S4. Based on the multi-energy coupling characteristic matrix, the operating status of various energy devices, and the energy flow relationship between various energy devices, construct an initial digital twin virtual model corresponding to the park's energy physical system; S5. Obtain the prediction sequence output by the initial digital twin virtual model within the current time window, compare the prediction sequence with the actual operation sequence of the park's energy physics system within the current time window, and determine the model deviation value. S6. If the model deviation value is greater than the preset deviation threshold, the coupling weight of the initial digital twin virtual model is corrected based on the model deviation value and the multi-energy coupling feature matrix to obtain the target digital twin virtual model; if the model deviation value is not greater than the preset deviation threshold, the model stability is verified based on the multi-energy coupling feature matrix, and the initial digital twin virtual model that passes the verification is determined as the target digital twin virtual model. S7. Simulate the future operating state of the park's energy physical system using a target digital twin virtual model to obtain the expected supply and demand gap sequence; generate initial scheduling instructions based on the expected supply and demand gap sequence and preset efficient scheduling objectives, wherein the efficient scheduling objectives include at least one of reducing energy operating costs, improving energy utilization efficiency, and reducing the degree of supply and demand imbalance; S8. Determine the extreme value of the supply-demand gap corresponding to the initial scheduling instruction; if the extreme value of the supply-demand gap is less than the preset imbalance threshold, then determine the initial scheduling instruction as the target scheduling instruction. S9. If the extreme value of the supply-demand gap is not less than the preset imbalance threshold, the initial scheduling instruction is modified according to the expected supply-demand gap sequence and the efficient scheduling target to obtain the target scheduling instruction. The target scheduling instruction is then output to control various energy devices in the park's energy physical system to perform park energy scheduling.

[0020] In this embodiment, the park's energy physical system includes two or more types of energy systems, such as electrical energy systems, thermal energy systems, cold energy systems, gas systems, and energy storage systems, and includes a set of energy equipment corresponding to the above-mentioned energy systems; the set of energy equipment may include photovoltaic power generation equipment, wind power generation equipment, gas turbines, gas boilers, waste heat recovery equipment, chillers, heat pump units, battery energy storage equipment, cold storage equipment, charging piles, transformers, distribution cabinets, heat exchangers, smart meters, heat meters, cold energy meters, and gas flow meters.

[0021] In this system, each energy device forms an energy flow relationship through energy supply, energy consumption, energy conversion, energy storage, and energy release during operation. For example, a gas turbine outputs electrical energy and generates waste heat, which is then fed into a heating or cooling branch via a waste heat recovery device. Battery energy storage devices receive electrical energy during a set charging period and supply energy to the electrical load during a set discharging period. Chillers consume electrical energy and output cooling capacity to the cooling load side. Therefore, this implementation method achieves the scheduling calculation of the park's energy physical system through progressive processing between continuous operating state sequences, real-time load fluctuation sequences, time-varying operating characteristics, energy coupling characteristic matrices, and digital twin virtual models.

[0022] Specifically, the continuous operating status sequence is collected by sensors, metering devices, equipment controllers, or energy management platforms deployed in the park's energy physics system.

[0023] Specifically, the continuous operating status sequence includes the input power, output power, start / stop status, load rate, operating efficiency, energy storage state of charge, supply water temperature, return water temperature, gas consumption, equipment alarm status, and environmental parameters of each energy device within a historical time period; the real-time load fluctuation sequence includes the change data of electrical load, heat load, cooling load, gas load, charging load, or comprehensive energy load within the current sampling period; the historical time period can be set to 24 hours, 72 hours, or 168 hours prior to the current time, and the sampling interval can be set to 1 minute, 5 minutes, or 15 minutes; the collected data is aligned according to the timestamp and subjected to missing value imputation, outlier removal, and normalization processing to ensure that the continuous operating status sequence and the real-time load fluctuation sequence meet the data format requirements of the subsequent model input.

[0024] Furthermore, the continuous operating state sequence and real-time load fluctuation sequence are input into the long short-term memory network, which extracts the time-varying operating characteristics of the park's energy physical system under different time dimensions.

[0025] The Long Short-Term Memory (LSTM) network selectively retains, updates, and outputs historical information through input, forget, and output gates, thereby characterizing the long-term dependencies and short-term fluctuations in the park's energy system operation data. For the park's energy physical system, the electrical load is affected by production scheduling, office hours, charging behavior, and external electricity prices; the cooling load is affected by outdoor temperature, building usage status, and the operating mode of cooling equipment; and the heat load is affected by heating demand, heat source equipment output, and waste heat recovery efficiency. The LSTM network extracts minute-level fluctuation characteristics, hourly-level change characteristics, and intraday cycle characteristics, respectively, and forms a feature set that includes load trend characteristics, equipment output change characteristics, energy storage charging and discharging characteristics, energy conversion efficiency characteristics, and environmental disturbance characteristics.

[0026] In this embodiment, a random forest is used to identify the energy coupling relationship between different energy devices for the feature set. The random forest consists of a set number of decision trees. Each decision tree is trained based on samples and features in the feature set, and the degree of influence of different features on the target variable is determined by the comprehensive output of each decision tree.

[0027] In practical implementation, the output power, input energy consumption, operating efficiency, load response or energy storage status of an energy device are taken as target variables, and the operating status characteristics, load fluctuation characteristics and environmental characteristics of other energy devices are taken as input variables. The correlation contribution between energy devices is calculated. An energy coupling feature matrix is ​​constructed based on the correlation contribution. The elements in the matrix represent the influence intensity of one energy device or one energy form on another energy device or another energy form.

[0028] For example, the impact of gas turbine output power on the heating capacity of waste heat recovery equipment, the impact of chiller unit power consumption on peak electrical load, and the impact of battery discharge power on external power purchase are all characterized by the corresponding elements in the energy coupling characteristic matrix.

[0029] Furthermore, based on the energy coupling characteristic matrix, the operating status of the energy equipment set, and the energy flow relationship between the energy equipment, an initial digital twin virtual model corresponding to the park's energy physical system is constructed.

[0030] Specifically, the initial digital twin virtual model includes an equipment state sub-model, an energy flow rotor model, a load forecasting sub-model, a constraint verification sub-model, and a scheduling simulation sub-model. The equipment state sub-model is used to characterize the start-stop status, rated power, minimum operating power, ramp rate, operating efficiency, and safe operating boundary of each energy device. The energy flow rotor model is used to characterize the input, output, conversion, transmission, and storage relationships of electrical energy, thermal energy, cold energy, gas energy, and energy storage among different devices. The load forecasting sub-model is used to predict future load changes based on time-varying operating characteristics. The constraint verification sub-model is used to determine whether the equipment output, energy storage state of charge, energy supply capacity, and energy balance meet the operating constraints. The scheduling simulation sub-model is used to simulate the supply and demand matching state under different scheduling commands.

[0031] Among them, the energy coupling feature matrix serves as the coupling weight in the initial digital twin virtual model, enabling the virtual model to reflect the correlation between equipment, energy forms, and time scales in the park's energy physical system.

[0032] In this embodiment, the predicted sequence output by the initial digital twin virtual model within the current time window is obtained, and the predicted sequence is compared with the actual operating sequence of the park's energy physics system within the current time window.

[0033] The current time window can be set to 15 minutes, 30 minutes, or 60 minutes; the forecast sequence includes forecast electrical load, forecast thermal load, forecast cooling load, forecast gas load, forecast equipment output, forecast energy storage state of charge, and forecast external energy purchase.

[0034] The actual operating sequence is acquired in real time by on-site data acquisition equipment or energy management platform in the park; the model deviation value can be calculated using mean square error, mean absolute error, maximum absolute deviation, normalized error or weighted comprehensive error; when electricity, heat, cold energy and gas energy are involved, the deviation under each energy dimension is calculated separately, and after setting weights according to energy cost, energy supply security level or load importance, the weighted sum is performed to obtain the model deviation value used to characterize the overall deviation of the model.

[0035] Furthermore, the model deviation value is compared with the preset deviation threshold. When the model deviation value is greater than the preset deviation threshold, it indicates that the operating state of the initial digital twin virtual model deviates from the allowable range from the energy physical system of the park. At this time, the coupling weight in the initial digital twin virtual model is corrected based on the model deviation value and the energy coupling feature matrix.

[0036] Specifically, the corresponding coupling weights are adjusted based on the energy type and equipment object causing the deviation. For example, when the predicted electrical load is less than the actual electrical load and the actual power consumption of the chiller unit increases, the coupling weight of the cooling load change on the electrical load prediction is increased; when the actual discharge capacity of the energy storage device is less than the model prediction value, the weight relationship between the energy storage state of charge and the available discharge power is corrected; when the waste heat recovery is less than the predicted value, the coupling weight between the gas turbine output and the waste heat supply capacity is decreased. The corrected digital twin virtual model serves as the target digital twin virtual model.

[0037] In addition, when the model deviation value is not greater than the preset deviation threshold, it means that the prediction result of the initial digital twin virtual model is within the allowable deviation range; at this time, the stability of the model is further verified based on the energy coupling feature matrix.

[0038] Specifically, stability verification includes determining whether there are abnormal abrupt changes in the key coupling weights in the energy coupling feature matrix, whether the coupling relationship between different energy devices satisfies the energy conservation constraint, whether the prediction deviation within a continuous time window remains within a preset stable range, and whether the operating boundary of the key energy device is consistent with the actual state. If the verification passes, the initial digital twin virtual model is determined as the target digital twin virtual model. Through the above processing, the target digital twin virtual model can track the current state of the physical system and update parameters according to changes in energy coupling relationships.

[0039] Furthermore, the future operating status of the park's energy physical system is simulated using a target digital twin virtual model to obtain the expected supply and demand gap sequence.

[0040] Specifically, the future operating status includes load demand, renewable energy output, conventional energy equipment output, energy storage charging and discharging status, external energy purchases, and adjustable load response within the future dispatch cycle. The expected supply-demand gap sequence represents the difference between energy supply and energy demand at each future moment; a positive difference indicates a surplus in energy supply, while a negative difference indicates a supply deficit. Initial dispatch instructions are generated based on the expected supply-demand gap sequence and preset dispatch objectives. These preset dispatch objectives include one or more of the following: reducing energy operating costs, enhancing energy utilization efficiency, and reducing the degree of supply-demand imbalance.

[0041] For example, to reduce energy operating costs, priority is given to using photovoltaic power generation, wind power generation, or energy storage power, while controlling external power purchases; to enhance energy efficiency, priority is given to activating combined cooling, heating, and power (CCHP) equipment, heat pump units, or waste heat recovery equipment that meet the set energy efficiency requirements; to reduce the degree of supply and demand imbalance, the supply and demand difference at each scheduling time is reduced through energy storage charging and discharging, standby equipment start-up and shutdown, and adjustable load transfer; initial scheduling instructions include energy equipment start-up and shutdown instructions, power setpoints, energy storage charging and discharging power, cold and heat source output allocation, external power purchase plans, external gas purchase plans, load reduction amounts, or adjustable load transfer amounts.

[0042] In this embodiment, the extreme value of the supply-demand gap corresponding to the initial scheduling instruction is determined; the extreme value of the supply-demand gap is the maximum value of the absolute value of the supply-demand gap at each time in the expected supply-demand gap sequence, or it is determined as the maximum supply surplus value or the maximum supply gap value according to the actual control requirements.

[0043] If the extreme value of the supply-demand difference is less than the preset imbalance threshold, it means that the initial dispatch instruction can keep the park's energy physical system within the allowable supply-demand balance range in the future dispatch cycle. Therefore, the initial dispatch instruction is determined as the target dispatch instruction. If the extreme value of the supply-demand difference is not less than the preset imbalance threshold, it means that the initial dispatch instruction may lead to an energy surplus or an energy shortage. In this case, the initial dispatch instruction is modified according to the expected supply-demand difference sequence and the preset dispatch target.

[0044] For example, when the expected supply-demand gap sequence indicates a supply gap in a future period, the system increases the energy storage discharge power, activates backup energy equipment, increases external energy purchases, or implements adjustable load reduction; when the expected supply-demand gap sequence indicates a supply surplus in a future period, the system increases the energy storage charging power, transfers adjustable loads, reduces the output of designated equipment, or increases the local consumption of renewable energy. After correction, the target dispatch instruction is obtained and output to the energy equipment set in the park's energy physical system to control the energy equipment set to execute park energy dispatch.

[0045] To verify the effect of the proposed method on improving the energy dispatching efficiency of the park, an offline playback experiment was conducted using historical operating data from a comprehensive park. The park includes photovoltaic power generation equipment, gas boilers, chillers, heat pump units, battery energy storage equipment, charging piles, and external power purchase interfaces. The experiment used data recorded by the park's energy management platform over three consecutive working days, with a sampling interval of 5 minutes, resulting in 288 sets of sampling data per day.

[0046] The experimental data includes the start-stop status, input power, output power, operating efficiency, energy storage state of charge, electrical load, thermal load, cooling load, external power purchase, and equipment execution feedback of each energy device. During the experiment, the data from the first two days were used as model building and parameter determination data, and the data from the third day were used as playback verification data.

[0047] The experiment sets up two scheduling methods for comparison: Group 1 is the control method, which adopts a fixed rule scheduling method. This method starts and stops equipment according to a preset time period. When the load increases, it prioritizes increasing the external power purchase and reduces the output of conventional equipment when the load decreases. It does not build a digital twin virtual model, nor does it modify the scheduling instructions according to the multi-energy coupling relationship.

[0048] The second group is the method of this application. This method first obtains the continuous operating state sequence and the real-time load fluctuation sequence, uses the long short-term memory network to extract time-varying operating features, and then uses random forest to identify the multi-energy coupling relationship between energy devices to construct a target digital twin virtual model. Subsequently, the target digital twin virtual model is used to deduce the future operating state, generate the expected supply and demand difference sequence, and confirm or correct the initial scheduling instructions according to the extreme value of the supply and demand difference.

[0049] During the replay verification on the third day, the two methods used the same historical load data, the same equipment capacity constraints, the same external electricity purchase price, and the same energy storage capacity constraints. They generated a dispatch instruction every 30 minutes and input the dispatch instruction into the same park energy simulation environment to record equipment output, electricity purchase, energy storage charging and discharging, and supply and demand difference.

[0050] Evaluation indicators: 1. Average supply-demand gap: At each sampling time, the absolute value of the difference between energy supply and energy demand is calculated, and then the average value of the absolute values ​​of the difference is taken over all 288 sampling times on the third day. This indicator is used to reflect the degree of supply and demand matching during the entire day's scheduling process.

[0051] 2. Maximum supply-demand gap: Select the value with the largest absolute value of the supply-demand gap from all 288 sampling times on day 3. This indicator is used to reflect the maximum supply-demand deviation that occurs during the scheduling process.

[0052] 3. Daily Energy Purchase Cost: This is calculated by calculating the externally purchased electricity and gas volume on the third day, multiplying each by the internally set electricity and gas purchase prices for the park, and then summing them up to obtain the daily energy purchase cost. This indicator reflects the park's energy operating costs.

[0053] 4. Effective utilization times of energy storage: This counts the number of times battery energy storage devices complete charging or discharging according to dispatch instructions within the third day. This indicator reflects whether energy storage devices participate in peak shaving and valley filling and supply-demand balance regulation.

[0054] 5. Number of Dispatch Instruction Corrections: This counts the number of times instruction corrections are triggered within the third day due to extreme supply-demand imbalances failing to meet the imbalance threshold. This indicator reflects the corrective capability of dispatch instructions before execution.

[0055] The data comparison table is as follows: As can be seen from the data in the table, under the same load data and the same equipment constraints, the average supply-demand gap of the method in this application decreased from 18.6kW to 9.4kW, and the maximum supply-demand gap decreased from 52.3kW to 27.8kW. This shows that after using a digital twin virtual model to predict the future operating state and combining it with the supply-demand gap to correct the dispatch instructions, the energy supply and demand matching degree of the park has been improved.

[0056] The daily energy purchase cost decreased from 8,630 yuan to 7,710 yuan because the method in this application can combine multi-energy coupling relationships and future supply-demand differences to prioritize the use of energy storage equipment and available energy equipment within the park, reducing unnecessary external energy purchases. The number of effective energy storage utilization times increased from 18 to 31, indicating that energy storage equipment is utilized during load peak-valley changes and supply-demand difference corrections. The number of dispatch command corrections was 7, indicating that the method in this application can identify supply-demand difference exceeding limits before dispatch commands are issued and correct the initial dispatch commands.

[0057] The above experiments demonstrate that, without altering the existing energy equipment hardware structure in the park, the method proposed in this application, through the participation of operating state sequences, load fluctuation sequences, multi-energy coupling characteristic matrices, and digital twin virtual models in scheduling calculations, can reduce the supply-demand gap, decrease the maximum supply-demand deviation, and lower daily energy purchase costs. The experimental results verify that the method proposed in this application has practical effects in reducing energy operating costs, improving energy utilization efficiency, reducing the risk of supply-demand imbalance, and enhancing the reliability of energy scheduling in the park.

[0058] Example 2: S1 includes obtaining equipment operation logs to generate a continuous operating status sequence and collecting real-time load data to obtain a real-time load fluctuation sequence; aligning the continuous operating status sequence and the real-time load fluctuation sequence to obtain a status fluctuation mapping relationship; constructing an energy operation map based on the status fluctuation mapping relationship; removing isolated nodes from the energy operation map if any exist, and obtaining an updated energy operation map; and obtaining the continuous operating status sequence and real-time load fluctuation sequence of the park's energy physical system within a historical time period through the updated energy operation map.

[0059] In this embodiment, when obtaining the equipment operation logs to generate a continuous operating status sequence, the equipment operation logs of each energy device are first read from the park's energy management platform.

[0060] Specifically, the equipment operation log includes equipment number, log time, start / stop status, input power, output power, operating efficiency, energy storage charge status, supply water temperature, return water temperature, gas consumption, alarm status, and control feedback status. For each sampling interval, the equipment operation logs falling within that sampling interval are searched. For input power, output power, operating efficiency, energy storage charge status, supply water temperature, return water temperature, and gas consumption, the arithmetic mean of the corresponding log values ​​within that sampling interval is taken as the status value of that sampling interval. For start / stop status, alarm status, and control feedback status, the last log record before the end of that sampling interval is taken as the status value of that sampling interval.

[0061] Subsequently, the status values ​​of each energy device within the same sampling interval are arranged into a group of device operation status data according to the device number order. Then, the device operation status data of each group are arranged in the order of the 288 sampling intervals to generate a continuous operation status sequence.

[0062] Furthermore, when collecting real-time load data to obtain the real-time load fluctuation sequence, metering data of electrical load, heat load, cooling load, gas load, and charging load are read from the metering modules of smart meters, heat meters, cooling meters, gas flow meters, and charging piles, respectively.

[0063] For each sampling interval, the various load metering values ​​falling within that sampling interval are statistically analyzed. When there are more than two metering values ​​of the same load type in the same sampling interval, the arithmetic mean of all metering values ​​in that sampling interval is taken as the load value of that sampling interval. When there is only one metering value of the same load type in the same sampling interval, that metering value is taken as the load value of that sampling interval.

[0064] Then, the load values ​​of the same load type in two adjacent sampling intervals are subtracted, and the load value of the previous sampling interval is subtracted from the load value of the next sampling interval to obtain the load fluctuation value of that load type in the next sampling interval; the first sampling interval does not have the previous sampling interval participating in the subtraction, and its load fluctuation value is set to 0; the load fluctuation values ​​of each sampling interval are arranged in the order of electric load, heat load, cooling load, gas load and charging load, and then arranged in the order of the 288 sampling intervals to obtain the real-time load fluctuation sequence.

[0065] Furthermore, when aligning the continuous operating status sequence with the real-time load fluctuation sequence, a 5-minute sampling interval is used as a unified time reference. For each sampling interval, the equipment operating status data corresponding to that sampling interval is bound to the load fluctuation data corresponding to that sampling interval to form a set of status fluctuation corresponding data.

[0066] Specifically, if the recording time of the equipment operation log is not exactly the same as the recording time of the load metering data, the correspondence is determined by the 5-minute sampling interval to which it belongs; if a certain energy device does not have an equipment operation log in a certain sampling interval, the equipment operation log that is closest to the end time of the sampling interval is read and the status value in the log is added to the sampling interval; if a certain load type does not have metering data in a certain sampling interval, the metering data that is closest to the end time of the sampling interval is read and the metering data is added to the sampling interval; after the addition is completed, each sampling interval has one set of equipment operation status data and one set of load fluctuation data. The 288 sets of status fluctuation corresponding data are arranged in chronological order to obtain the status fluctuation mapping relationship.

[0067] Furthermore, when constructing the energy operation map based on the state fluctuation mapping relationship, energy equipment, load type, and metering point are used as map nodes. Energy equipment nodes are determined by the park's energy management ledger, load type nodes include electrical load nodes, heat load nodes, cold load nodes, gas load nodes, and charging load nodes, and metering point nodes are determined by the equipment numbers of smart meters, heat meters, cold meters, gas flow meters, and charging pile metering modules.

[0068] For any energy device and any load type, the output power of the energy device in 288 sampling intervals is first obtained according to the continuous operating state sequence. Then, the change in output power of the energy device between adjacent sampling intervals is calculated by subtracting the output power of the previous sampling interval from the output power of the next sampling interval. The change in output power of the first sampling interval is set to 0.

[0069] Subsequently, the load fluctuation value of this load type within 288 sampling intervals is obtained based on the real-time load fluctuation sequence. Then, the output power change value and load fluctuation value within the same sampling interval are compared one by one: when both are simultaneously positive or simultaneously negative, it is recorded as a change in the same direction; when one is positive and the other is negative, it is recorded as a change in opposite directions; when either is 0, it is recorded as a no-change item. The number of simultaneous changes, the number of reverse changes, and the number of no-change items are counted within the 288 sampling intervals, and the result of subtracting the number of reverse changes from the number of simultaneous changes is used as the correlation count value between the energy device and the load type. When the correlation count value is not equal to 0, a connection edge is generated between the energy device node and the load type node, and the correlation count value is used as the edge attribute of the connection edge; when the correlation count value is equal to 0, the connection edge is not generated. After completing the connection judgment between all energy devices and all load types according to the above processing method, an energy operation map is formed.

[0070] In this embodiment, if there are isolated nodes in the energy operation graph, then the connection edges of each node in the energy operation graph are checked.

[0071] For any given node, the number of edges pointing to that node and the number of edges originating from that node are counted separately, and then the two counts are added together to obtain the total number of connections for that node. When the total number of connections is 0, the node is determined to be an isolated node. Isolated nodes do not have valid connections with energy equipment, load types, or metering points, and their corresponding data are not included in the subsequent feature extraction process. When removing isolated nodes, the isolated node is deleted from the energy operation graph, and the node attributes, data indexes, and cache records corresponding to the isolated node are deleted simultaneously. After all isolated nodes are deleted, the remaining nodes and their connecting edges constitute the updated energy operation graph.

[0072] As a preferred embodiment, when obtaining the continuous operating status sequence and real-time load fluctuation sequence of the park's energy physical system within a historical time period through the updated energy operation map, the energy equipment nodes retained in the updated energy operation map are read first, and the corresponding energy equipment's operating status data within 288 sampling intervals is extracted from the original continuous operating status sequence based on the equipment number of these energy equipment nodes.

[0073] Subsequently, the load type nodes and metering point nodes retained in the updated energy operation map are read, and load fluctuation data of the corresponding load in 288 sampling intervals are extracted from the original real-time load fluctuation sequence according to the corresponding load type and metering point number. The extracted operation status data are still arranged in the order of the sampling intervals to form the updated continuous operation status sequence. The extracted load fluctuation data are still arranged in the order of the sampling intervals to form the updated real-time load fluctuation sequence.

[0074] Example 3: S2 includes acquiring a continuous operating state sequence and a real-time load fluctuation sequence; performing synchronization alignment processing on the continuous operating state sequence and the real-time load fluctuation sequence to obtain a synchronized operating state sequence and a synchronized load fluctuation sequence; performing sequence truncation based on the synchronized operating state sequence and the synchronized load fluctuation sequence to obtain a multidimensional state subsequence and a multidimensional load subsequence; and extracting time-varying operating features from the multidimensional state subsequence and the multidimensional load subsequence through a long short-term memory network to obtain a feature set.

[0075] In this embodiment, when acquiring the continuous operating status sequence and the real-time load fluctuation sequence, the data output by S1 is read first. In the continuous operating status sequence, each set of energy equipment operating status data includes equipment start / stop status, input power, output power, operating efficiency, energy storage charge status, water supply temperature, return water temperature, gas consumption, and alarm status. In the real-time load fluctuation sequence, each set of load fluctuation data includes electrical load fluctuation value, heat load fluctuation value, cooling load fluctuation value, gas load fluctuation value, and charging load fluctuation value. After reading, the number of records in the two sequences is checked according to the order of the sampling intervals. When both sequences contain 288 sets of data, synchronization alignment processing is performed. When any sampling interval lacks equipment operating status data or load fluctuation data, completion processing is performed first.

[0076] Specifically, the rules for data completion are as follows: when equipment operating status data is missing, the set of equipment operating status data closest to the end time of the previous sampling interval is used to fill in the missing data; when load fluctuation data is missing, the set of load fluctuation data closest to the end time of the previous sampling interval is used to fill in the missing data; when the first sampling interval is missing, the first set of valid data following it is used to fill in the missing data.

[0077] Furthermore, when synchronizing and aligning the continuous operating status sequence and the real-time load fluctuation sequence, a 5-minute sampling interval is used as a unified time reference.

[0078] Specifically, the sampling interval to which each set of equipment operating status data belongs is first determined based on the data timestamp, and then the sampling interval to which each set of load fluctuation data belongs is determined based on the data timestamp. For the same sampling interval, the equipment operating status data and load fluctuation data within the sampling interval are bound together as a set of synchronized data. When there are more than two sets of continuous numerical operating status data within the same sampling interval, the arithmetic mean of these values ​​is taken as the operating status value of the sampling interval. When there are more than two sets of start / stop or alarm states within the same sampling interval, the last set of states before the end time of the sampling interval is taken as the status value of the sampling interval. When there are more than two sets of records of load fluctuation data within the same sampling interval, the arithmetic mean of these records is taken as the load fluctuation value of the sampling interval.

[0079] After the above processing is completed, each of the 288 sampling intervals has corresponding equipment operation status data and load fluctuation data. The equipment operation status data is arranged in chronological order to form a synchronous operation status sequence, and the load fluctuation data is arranged in the same chronological order to form a synchronous load fluctuation sequence.

[0080] Furthermore, when truncating the sequence based on the synchronous operating status sequence and the synchronous load fluctuation sequence, a fixed-length sliding truncation method is adopted.

[0081] Specifically, the truncation window is set to 12 consecutive sampling intervals, corresponding to 60 minutes of operating data; the sliding step is set to 1 sampling interval, corresponding to 5 minutes; the first truncation starts from the first sampling interval, taking the synchronous operating status data from the first to the 12th sampling interval as the first multidimensional state subsequence, and simultaneously taking the synchronous load fluctuation data from the first to the 12th sampling interval as the first multidimensional load subsequence; the second truncation starts from the second sampling interval, taking the data from the second to the 13th sampling interval to form the second multidimensional state subsequence and the second multidimensional load subsequence; this process is repeated, moving one sampling interval forward each time, until the end of the truncation window reaches the 288th sampling interval; since the total number of sampling intervals is 288 and the truncation window length is 12, the number of truncations is 277, resulting in 277 multidimensional state subsequences and 277 multidimensional load subsequences.

[0082] In this embodiment, each set of multidimensional state subsequences retains the operating status changes of each energy device within 12 sampling intervals, and each set of multidimensional load subsequences retains the fluctuation changes of various types of loads within the same 12 sampling intervals.

[0083] Specifically, the state subsequence and load subsequence are completely identical in terms of truncation start point, truncation end point, and number of sampling intervals. Therefore, the multidimensional state subsequence and multidimensional load subsequence with the same number constitute a set of long short-term memory network input samples. Before inputting into the network, the power, temperature, gas consumption, energy storage state of charge, and load fluctuation values ​​are subjected to dimension unification processing. The dimension unification processing uses predetermined upper and lower limits to convert each value to between 0 and 1. Start-up and alarm states are directly input according to the state codes in the equipment operation log, where running state is recorded as 1, stopped state is recorded as 0, alarm presence is recorded as 1, and no alarm presence is recorded as 0.

[0084] As a preferred embodiment, when extracting time-varying operational features from multidimensional state subsequences and multidimensional load subsequences using a long short-term memory network, data within the same truncated window are input item by item according to the time sequence of the sampling interval.

[0085] In this process, for each sampling interval, the equipment operating status data and load fluctuation data of that sampling interval are merged into one set of network input data. The Long Short-Term Memory (LSTM) network receives 12 sets of network input data sequentially and updates the memory unit with each input. The input gate determines the content to be written into the memory unit based on the equipment operating status and load fluctuation data of the current sampling interval. The forget gate determines the content to be retained based on the memory content of the previous sampling interval and the input content of the current sampling interval. The output gate generates the hidden state of the current sampling interval based on the updated memory content. After the recursion of 12 sampling intervals is completed, the hidden state of the last sampling interval is taken as the time-varying operating feature of the truncated window.

[0086] The above-mentioned long short-term memory network processing was sequentially applied to 277 sets of input samples to obtain 277 sets of time-varying operating features. Each set of time-varying operating features includes changes in equipment start-up and shutdown, equipment output, energy storage charging and discharging, energy conversion efficiency, load fluctuation trends, and equipment load response relationships within a corresponding 60-minute window. The 277 sets of time-varying operating features were summarized in chronological order according to the truncated window to obtain the feature set.

[0087] Example 4: S3 includes obtaining a feature set to construct an initial feature vector; receiving the initial feature vector through a random forest model to obtain a feature contribution sequence; calculating the coupling weights between different energy devices based on the feature contribution sequence; determining the multi-energy coupling relationship between different energy devices if the coupling weights meet the preset matching conditions; and performing matrix processing on the multi-energy coupling relationship to obtain the multi-energy coupling feature matrix.

[0088] In this embodiment, the feature set is output by the Long Short-Term Memory network. Each set of features corresponds to a truncated window. The feature content includes changes in energy equipment start-up and shutdown, changes in equipment output, changes in energy storage charging and discharging, changes in energy conversion efficiency, load fluctuation trends, and equipment load response relationships. The random forest model takes this feature set as input, calculates the contribution of each feature to the changes in energy equipment operation, and then converts the contribution results into the coupling weights between energy equipment.

[0089] Furthermore, when acquiring the feature set to construct the initial feature vector, all time-varying running features output by S2 are first read and processed group by group according to the sampling window order.

[0090] For the same sampling window, the operating characteristics of each energy device are first arranged according to the device number in the park's energy management ledger. For the same energy device, the following are arranged in sequence: start-stop changes, input power changes, output power changes, operating efficiency changes, energy storage state of charge changes, supply water temperature changes, return water temperature changes, and gas consumption changes. After the device characteristics are arranged, the following are arranged: electrical load fluctuations, heat load fluctuations, cooling load fluctuations, gas load fluctuations, and charging load fluctuations. After the arrangement is completed, the data corresponding to the sampling window constitutes an initial feature vector. All sampling windows are processed in the same order, so that each feature position corresponds to the same device and the same feature type in different sampling windows, avoiding misalignment of feature meanings during training of the random forest model.

[0091] In this embodiment, when the initial feature vector is received and the feature contribution sequence is obtained through the random forest model, a random forest regression model is established for each target energy device.

[0092] For the target energy equipment, the output power change, input power change, or operating efficiency change of the target energy equipment in the next sampling window is used as the prediction object, and the initial feature vector of the current sampling window is used as the input data. The random forest regression model sets 100 regression decision trees. When generating a node, each regression decision tree first calculates the dispersion of the predicted object value in the current node relative to the average value of the node. Then, it selects one candidate feature and the corresponding split value, divides the samples in the current node into the left branch and the right branch, and calculates the dispersion of the predicted object value in the left branch and the right branch respectively.

[0093] Next, according to the proportion of the number of samples in the left branch and the number of samples in the right branch in the current node's sample count, the dispersion of the left branch and the dispersion of the right branch are weighted and summed to obtain the dispersion after splitting. The dispersion before splitting minus the dispersion after splitting is used as the contribution value of the candidate feature to the node. If the obtained value is greater than 0, the node split is retained; if the obtained value is not greater than 0, the candidate split is not counted as a valid contribution.

[0094] In this embodiment, when the same feature is used for effective splitting at different nodes of a regression decision tree, the contribution values ​​of the feature at each node are added together to obtain the in-tree contribution value of the feature in the regression decision tree.

[0095] Specifically, after the 100 regression decision trees are calculated, the in-tree contribution values ​​of the same feature in the 100 regression decision trees are added together to obtain the total contribution value of the feature to the target energy equipment prediction object; then, the total contribution value of each feature is output according to the feature arrangement order of the initial feature vector to form the feature contribution degree sequence corresponding to the target energy equipment; the above processing is performed for each target energy equipment to obtain the feature contribution degree sequence corresponding to each target energy equipment.

[0096] As a preferred embodiment, when calculating the coupling weight between different energy devices based on the feature contribution sequence, the feature contribution sequence is first merged according to the feature source.

[0097] For any target energy device, its corresponding feature contribution sequence is read. The contribution values ​​from changes in start / stop, input power, output power, operating efficiency, energy storage state of charge, supply water temperature, return water temperature, and gas consumption originating from the same influencing energy device are summed to obtain the associated contribution value of that influencing energy device to the target energy device. Subsequently, the associated contribution values ​​of all energy devices other than the target energy device itself are summed to obtain the total external associated contribution value of the target energy device. The associated contribution value of a single influencing energy device is divided by the total external associated contribution value, and the result is used as the coupling weight from that influencing energy device to the target energy device. If the total external associated contribution value is 0, the coupling weight from each influencing energy device to the target energy device is recorded as 0.

[0098] If the coupling weights meet the preset matching conditions, the multi-energy coupling relationship between different energy devices is determined.

[0099] Specifically, the preset matching conditions consist of numerical conditions and connection conditions. The numerical condition is that the coupling weight is not less than 0.15. The connection condition is that the two energy devices have a direct connection edge in the energy operation map, or that the two energy devices have connection edges with the same energy type node or the same load node respectively. For any two energy devices, it is first determined whether their coupling weight is not less than 0.15. After the numerical condition is met, it is then checked whether there is a corresponding connection condition in the updated energy operation map. When both the numerical condition and the connection condition are met, the two energy devices are determined to have a multi-energy coupling relationship. If either condition is not met, the multi-energy coupling relationship between the two energy devices is not generated. Through this process, the feature contribution result must be limited by both the numerical contribution and the energy topology connection, and the resulting coupling relationship is consistent with the actual energy flow relationship of the park's energy physical system.

[0100] Furthermore, when performing matrix processing on multi-energy coupling relationships, the row and column order of the matrix is ​​first determined according to the equipment number order in the park's energy management ledger.

[0101] In this matrix, each row corresponds to one energy device as the source of influence, and each column corresponds to one energy device as the object of influence. For two energy devices with a confirmed multi-energy coupling relationship, the coupling weight from the source energy device to the object energy device is written to the corresponding position. For two energy devices with an undetermined multi-energy coupling relationship, the corresponding position is written as 0. The position corresponding to the same energy device itself is also written as 0. After all rows and columns are filled, a multi-energy coupling feature matrix is ​​obtained. Each non-zero value in this matrix corresponds to a confirmed multi-energy coupling relationship, and the value indicates the contribution ratio of the source energy device to the operational changes of the object energy device.

[0102] Example 5: S4 includes acquiring a multi-energy coupling feature matrix and the operating status of various energy devices. The multi-energy coupling feature matrix is ​​used to extract coupling weights, and the operating status is used to extract state mapping vectors. Energy flow relationships are determined based on the coupling weights and the state mapping vectors. It is then determined whether the coupling weights exceed a preset threshold. If the coupling weights exceed the preset threshold, flow nodes corresponding to the energy flow relationships are generated. A topology connection graph is constructed based on the flow nodes and energy flow relationships. This topology connection graph is used to generate virtual entities and mapping edges. An initial digital twin virtual model corresponding to the park's energy physical system is constructed using the virtual entities and mapping edges.

[0103] In this embodiment, the energy equipment in the park is based on the equipment number in the park's energy management ledger. The equipment includes photovoltaic power generation equipment, gas turbines, gas boilers, waste heat recovery equipment, chillers, heat pump units, battery energy storage equipment, cold storage equipment, charging piles, transformers, and heat exchangers. The row and column order of the multi-energy coupling feature matrix is ​​arranged according to the equipment number. The values ​​in the matrix that are located in the row of the source equipment and the column of the object equipment are used as the coupling weight from the source equipment to the object equipment.

[0104] Furthermore, when obtaining the multi-energy coupling feature matrix and the operating status of energy equipment, the multi-energy coupling feature matrix output by S3 is read first, and the coupling weights are extracted item by item according to the equipment number.

[0105] Subsequently, the operating status of each energy device at the current scheduling moment is read from the park's energy management platform. The operating status includes start / stop status, input energy type, output energy type, input power, output power, operating efficiency, energy storage charge status, rated capacity, occupied capacity, supply water temperature, return water temperature, gas consumption, and equipment alarm status. For each energy device, the status value is extracted in the order of device number, start / stop status, input energy type, output energy type, input power, output power, operating efficiency, and available capacity to form a status mapping vector.

[0106] Specifically, when the device is in operation, the start / stop status is written as 1; when the device is in a stopped state, the start / stop status is written as 0; when the device is in an alarm state, the available capacity is written as 0; when the device is not in an alarm state, the available capacity is the result of subtracting the occupied capacity from the rated capacity.

[0107] Furthermore, when determining the energy flow relationship based on the coupling weight and state mapping vector, one energy device is selected as the source of influence, and another energy device is selected as the object of influence.

[0108] Specifically, first read the coupling weight between the two, then read their state mapping vectors; when the start / stop state of the source device is 1, the start / stop state of the target device is 1, and the output energy type of the source device is consistent with the input energy type of the target device, it is determined that the two have the conditions for energy transfer.

[0109] The energy flow value is then determined as follows: when the output power of the source device is less than the input power of the target device, the output power of the source device is used as the energy flow value; when the output power of the source device is not less than the input power of the target device, the input power of the target device is used as the energy flow value; the energy flow value is then multiplied by the coupling weight to obtain the effective energy flow; when the effective energy flow is greater than 0, an energy flow relationship is determined to exist between the source device and the target device; when the effective energy flow is equal to 0, no corresponding energy flow relationship is generated.

[0110] Furthermore, when determining whether the coupling weight exceeds a preset threshold, the threshold is determined based on historical operating data.

[0111] Specifically, the operating data for the 30 calendar days prior to the current scheduling date is taken, with a sampling interval of 5 minutes, resulting in a total of 8640 sets of sampled data. First, based on the actual metering data of the equipment and the equipment connection ledger, the equipment pairs that have actually undergone energy transfer are identified, and the coupling weight of these equipment pairs within the 30 calendar days is calculated. Then, the equipment pairs that have not undergone actual energy transfer are identified, and the coupling weight of these equipment pairs within the same period is calculated.

[0112] In this embodiment, the minimum coupling weight is 0.18 for device pairs where actual energy transfer has occurred, and the maximum coupling weight is 0.12 for device pairs where no actual energy transfer has occurred. The arithmetic mean of 0.18 and 0.12 is taken to obtain a preset threshold of 0.15. Subsequently, each coupling weight in the multi-energy coupling feature matrix is ​​compared. If the coupling weight is greater than 0.15, it is determined that the coupling weight exceeds the preset threshold; if the coupling weight is not greater than 0.15, it is determined that the coupling weight does not exceed the preset threshold.

[0113] In this embodiment, if the coupling weight exceeds a preset threshold, a flow node corresponding to the energy flow relationship is generated.

[0114] When generating a flow node, the node attributes are written with the following information: the number of the source device, the number of the affected device, the energy type, the flow energy value, the effective flow volume, the coupling weight, and the current scheduling time. For electrical energy flow relationships, the node attributes record electrical power; for thermal energy flow relationships, the node attributes record thermal power, supply water temperature, and return water temperature; for cold energy flow relationships, the node attributes record cold energy, supply water temperature, and return water temperature; for gas energy flow relationships, the node attributes record gas consumption; and for energy storage flow relationships, the node attributes record charging power, discharging power, and energy storage state of charge. For device pairs with a coupling weight not exceeding 0.15, even if there is a similar energy input / output relationship, no flow node is generated.

[0115] Furthermore, when constructing the topology connection diagram based on the flow nodes and energy flow relationships, energy devices are treated as device nodes, and the already generated flow nodes are treated as intermediate nodes.

[0116] For each flow node, a connection edge is established from the source device node to the flow node, and another connection edge is established from the flow node to the affected device node. The former connection edge records the output energy type, output power, and output time; the latter connection edge records the input energy type, input power, and input time. If one energy device outputs energy to two energy devices, two flow nodes are generated respectively, and corresponding connection edges are established for each. If two energy devices input energy to one energy device, two connection paths are retained to enter the energy device through different flow nodes. After all device nodes, flow nodes, and connection edges are generated, a topology connection graph is formed.

[0117] As a preferred embodiment, a topological connection graph is used to generate virtual entities and map associated edges.

[0118] Specifically, for each device node in the topology connection diagram, a virtual device entity is generated, and the state mapping vector corresponding to the device node is written into the attribute field of the virtual device entity, so that the virtual device entity records the device number, device type, start / stop status, input energy type, output energy type, input power, output power, operating efficiency, available capacity, and alarm status; for each flow node in the topology connection diagram, an energy flow virtual entity is generated, and the energy type, flow energy value, effective flow amount, coupling weight, and flow time are written into the energy flow virtual entity.

[0119] Subsequently, mapping association edges are established according to the connection direction in the topology connection graph: device mapping association edges are established between physical devices and device virtual entities, flow mapping association edges are established between energy flow relationships and energy flow virtual entities, and energy transfer association edges are established between device virtual entities and energy flow virtual entities.

[0120] Furthermore, when constructing the initial digital twin virtual model using virtual entities and mapping association edges, the virtual entities of the equipment are written into the device object layer of the model, the virtual entities of energy flow are written into the flow object layer of the model, and the mapping association edges of the equipment, the flow mapping association edges, and the energy transmission association edges are written into the association object layer of the model.

[0121] Specifically, the device object layer is used to record the current operating status and operating constraints of each energy device; the flow object layer is used to record the energy type, effective flow volume and coupling weight between different energy devices; the association object layer is used to record the mapping path between physical devices, virtual entities and energy flow relationships; after the model is built, the device status, coupling weight and energy flow relationship at the current scheduling time are all written into the initial digital twin virtual model.

[0122] Example 6: S5 includes obtaining the predicted sequence and actual running sequence of the initial digital twin virtual model; processing the predicted sequence and actual running sequence using a dynamic time warping algorithm to obtain a sequence alignment matrix; generating an error distribution vector based on the sequence alignment matrix; determining whether the error distribution vector exceeds a preset threshold; if the error distribution vector exceeds the preset threshold, extracting the cumulative deviation; and using the cumulative deviation to determine the model deviation value.

[0123] In this embodiment, when obtaining the prediction sequence and actual running sequence of the initial digital twin virtual model, the 12 sets of prediction data output by the initial digital twin virtual model within the current time window are read first, and each set of prediction data corresponds to a 5-minute sampling interval.

[0124] Specifically, the forecast data includes forecasted electrical load, forecasted heat load, forecasted cooling load, forecasted gas load, forecasted equipment output power, and forecasted energy storage state of charge. Subsequently, actual operating data within the same current time window is read from smart meters, heat meters, cooling meters, gas flow meters, equipment controllers, and energy storage management units. The actual operating data is grouped according to the same sampling interval, forming 12 sets of actual operating data. If there are more than two metering records within the same sampling interval, the arithmetic mean of the continuous numerical data is taken. If there are more than two state records within the same sampling interval, the last state record before the end of that sampling interval is taken. After processing, both the forecast sequence and the actual operating sequence have 12 sets of data, and each set of data corresponds to one sampling interval.

[0125] Furthermore, when using the dynamic time warping algorithm to process the predicted sequence and the actual running sequence, the local difference between each sampling interval in the predicted sequence and each sampling interval in the actual running sequence is first calculated.

[0126] Specifically, when calculating the local difference value, the absolute values ​​of the differences between the predicted and actual electrical loads, the predicted and actual heat loads, the predicted and actual cooling loads, the predicted and actual gas loads, the predicted and actual equipment output power, and the predicted and actual energy storage state of charge are calculated separately. Then, the absolute values ​​of these six differences are added together to obtain one local difference value.

[0127] Since both the predicted sequence and the actual running sequence include 12 sampling intervals, a local difference result consisting of 12 rows and 12 columns can be obtained, where the rows correspond to the sampling intervals of the predicted sequence and the columns correspond to the sampling intervals of the actual running sequence.

[0128] Furthermore, when generating the sequence alignment matrix, the cumulative difference value in the first row and first column is determined as the local difference value between the first predicted sampling interval and the first actual sampling interval; for the remaining positions in the first row, the local difference value at the current position is added to the cumulative difference value at the left adjacent position to obtain the cumulative difference value at the current position.

[0129] Specifically, for the remaining positions in the first column, the local difference value of the current position is added to the cumulative difference value of the adjacent position above it to obtain the cumulative difference value of the current position. For positions that are not in the first row and not in the first column, the cumulative difference values ​​of the adjacent positions to the left, above, and to the upper left are read first, and the smallest value is taken and added to the local difference value of the current position to obtain the cumulative difference value of the current position. After completing the calculation for all positions in 12 rows and 12 columns, the sequence alignment matrix is ​​obtained. This sequence alignment matrix is ​​used to determine the minimum cumulative difference alignment path between the predicted sequence and the actual running sequence.

[0130] Furthermore, when generating the error distribution vector based on the sequence alignment matrix, the alignment path is searched backwards starting from the last row and last column of the sequence alignment matrix.

[0131] Specifically, during each search, the position with the smallest cumulative difference value among the adjacent positions to the left, above, and upper left of the current position is selected as the previous alignment position, until the first row and first column are reached. Based on this alignment path, the correspondence between the predicted sampling interval and the actual sampling interval can be obtained. For each set of correspondences on the alignment path, the electrical load error, thermal load error, cooling load error, gas load error, equipment output power error, and energy storage state of charge error are calculated respectively; each error is the absolute value of the predicted value minus the actual value. Arranging the errors on the alignment path in chronological order generates an error distribution vector.

[0132] As a preferred embodiment, when determining whether the error distribution vector exceeds a preset threshold, the preset threshold is determined before performing the comparison of the current time window.

[0133] Specifically, the threshold determination process is as follows: Historical data from the 30 calendar days prior to the current scheduling date, showing no equipment alarms, no communication interruptions, and completed scheduling commands, are selected with a sampling interval of 5 minutes, resulting in 8640 sets of historical samples. For each type of prediction object, the absolute value of the error between the historical predicted value and the historical actual value is calculated, and the average error value for that type of prediction object is obtained. Then, the square of the difference between each absolute error value and the average error value is calculated, and the square root of the average of all squared differences is taken to obtain the discrete error value for that type of prediction object. The average error value is added to twice the discrete error value to obtain the historical error limit for that type of prediction object. Then, the allowable error value in the corresponding metering equipment calibration file is read, and the historical error limit and the allowable error value are compared. The larger of the two values ​​is taken as the preset threshold for that type of prediction object. Thus, the threshold values ​​for electrical load, heat load, cold load, gas load, equipment output power, and energy storage state of charge are obtained respectively.

[0134] In this embodiment, during the threshold determination process, the error terms in the error distribution vector are read one by one, and the corresponding threshold is read according to the prediction object to which the error term belongs.

[0135] Specifically, the errors are compared with the electrical load threshold, the thermal load threshold, the cooling load threshold, the gas load threshold, the equipment output power threshold, and the energy storage state of charge threshold. If any error term exceeds its corresponding threshold, the error distribution vector is determined to exceed the preset threshold; if none of the error terms exceed their corresponding thresholds, the error distribution vector is determined to not exceed the preset threshold.

[0136] Furthermore, if the error distribution vector exceeds a preset threshold, the cumulative deviation is extracted. When extracting the cumulative deviation, only error terms that exceed the corresponding threshold are processed.

[0137] For each error term exceeding a threshold, the corresponding threshold is subtracted from the error term to obtain the over-limit error. Then, the over-limit errors are aggregated according to the predicted object: over-limit errors belonging to electrical load are accumulated as the electrical load deviation accumulation; over-limit errors belonging to heat load are accumulated as the heat load deviation accumulation; over-limit errors belonging to cooling load are accumulated as the cooling load deviation accumulation; over-limit errors belonging to gas load are accumulated as the gas load deviation accumulation; over-limit errors belonging to equipment output power are accumulated as the equipment power deviation accumulation; and over-limit errors belonging to energy storage state of charge are accumulated as the energy storage state of charge deviation accumulation. Error terms not exceeding the corresponding threshold are not included in the deviation accumulation.

[0138] Furthermore, when using the cumulative deviation to determine the model deviation value, weights are first assigned to each prediction object according to the park's energy dispatch strategy.

[0139] In this embodiment, the weights for electrical load (0.25), thermal load (0.15), cooling load (0.15), gas load (0.10), equipment output power (0.25), and energy storage state of charge (0.10) are summed to 1. Then, the cumulative deviation of electrical load is multiplied by 0.25, the cumulative deviation of thermal load is multiplied by 0.15, the cumulative deviation of cooling load is multiplied by 0.15, the cumulative deviation of gas load is multiplied by 0.10, the cumulative deviation of equipment power is multiplied by 0.25, and the cumulative deviation of energy storage state of charge is multiplied by 0.10. Finally, the six results are summed to obtain the model deviation value. This model deviation value is used to characterize the degree of deviation of the initial digital twin virtual model from the actual operating state of the park's energy physical system within the current time window.

[0140] Example 7: S6 includes determining whether the model deviation value is greater than a preset deviation threshold; if the model deviation value is greater than the preset deviation threshold, then a weight update vector is generated based on the energy conversion efficiency and load response gradient in the multi-energy coupling feature matrix; the coupling weights of the initial digital twin virtual model are corrected through the weight update vector to obtain the target digital twin virtual model; if the model deviation value is not greater than the preset deviation threshold, then the initial digital twin virtual model that has passed the verification using the multi-energy coupling feature matrix is ​​determined as the target digital twin virtual model.

[0141] In this embodiment, S6 is used to perform branching processing on the initial digital twin virtual model based on the model deviation value obtained within the current time window. The model deviation value originates from S5 and represents the degree of difference between the predicted sequence of the initial digital twin virtual model and the actual operating sequence of the park's energy physics system. Before the deviation value is used for judgment, a deviation threshold is determined. The deviation threshold is calculated from historical operating data, specifically: historical data within 30 natural days prior to the current scheduling date, where there were no equipment alarms, no communication interruptions, and scheduling instructions were executed, is selected. A statistical window is defined as 60 minutes, with a sampling interval of 5 minutes. For each statistical window, a historical model deviation value is obtained according to the calculation method in S5. A total of 720 statistical windows are formed over 30 natural days. The average historical deviation is obtained by summing the 720 historical model deviation values ​​and dividing by 720. Then, the difference between each historical model deviation value and the average historical deviation is calculated. The square of each difference is taken and the average is calculated. The square root of the average is then taken to obtain the discrete historical deviation value. The average historical deviation is added to twice the discrete historical deviation value to obtain the preset deviation threshold. When the model deviation value in the current time window is greater than the deviation threshold, coupling weight correction is performed. When the model deviation value in the current time window is not greater than the deviation threshold, model validation is performed.

[0142] Furthermore, if the model deviation value is greater than the preset deviation threshold, the energy device pairs that have formed a coupling relationship are read from the multi-energy coupling feature matrix, and the corresponding coupling weight, energy conversion efficiency and load response gradient of the energy device pair are obtained; the energy conversion efficiency is determined based on the input energy and output energy in the current time window.

[0143] For any energy device pair, the input and output power of the affected device are first read within 12 sampling intervals. Since the sampling interval is 5 minutes, the energy value of each sampling interval is calculated by multiplying the corresponding power by 5 minutes. The input energy in the 12 sampling intervals is added together to obtain the total input energy. The output energy in the 12 sampling intervals is added together to obtain the total output energy. If the total input energy is not 0, the energy conversion efficiency of the energy device pair is obtained by dividing the total output energy by the total input energy. If the total input energy is 0, the energy device pair does not participate in this weight update.

[0144] Furthermore, the load response gradient is used to represent the degree of correspondence between changes in energy equipment output and changes in load.

[0145] Specifically, during the calculation, firstly, the output power change of the source device within the current time window between adjacent sampling intervals is read. The output power change is obtained by subtracting the output power of the previous sampling interval from the output power of the later sampling interval. Next, the load change of the corresponding energy type of the affected device is read. The load change is obtained by subtracting the load value of the previous sampling interval from the load value of the later sampling interval. For each adjacent sampling interval, if the output power change is not zero, the response value of that adjacent sampling interval is obtained by dividing the load change by the output power change. If the output power change is zero, that adjacent sampling interval is not included in the response value calculation. All response values ​​participating in the calculation within the current time window are summed, and then divided by the number of response values ​​participating in the calculation to obtain the load response gradient of the energy device pair.

[0146] Furthermore, when generating the weight update vector, the update direction for each pair of energy devices is first determined.

[0147] If the predicted value of the energy device for the corresponding energy dimension is less than the actual value, it indicates that the model's estimate of the supply or response of that energy dimension is insufficient, and the update direction is determined to increase the coupling weight; if the predicted value of the energy device for the corresponding energy dimension is greater than the actual value, it indicates that the model's estimate of the supply or response of that energy dimension exceeds the actual operating result, and the update direction is determined to decrease the coupling weight.

[0148] Subsequently, the deviation value is subtracted from the deviation threshold to obtain the excess deviation amount; then the excess deviation amount is divided by the deviation threshold to obtain the deviation adjustment coefficient; for any energy equipment pair, its energy conversion efficiency, load response gradient and deviation adjustment coefficient are multiplied to obtain the weight update amount of the energy equipment pair; the weight update amounts of all energy equipment pairs are arranged in the row and column order of the multi-energy coupling feature matrix to form the weight update vector.

[0149] In this embodiment, when correcting the coupling weights of the initial digital twin virtual model through the weight update vector, the original coupling weights between each virtual entity in the initial digital twin virtual model are read and updated item by item according to the weight update vector.

[0150] Specifically, for energy device pairs whose update direction is increasing, the original coupling weight is added to the corresponding weight update amount to obtain the corrected coupling weight; for energy device pairs whose update direction is decreasing, the original coupling weight is subtracted from the corresponding weight update amount to obtain the corrected coupling weight; if the corrected coupling weight is less than 0, it is set to 0; if the corrected coupling weight is greater than 1, it is set to 1; the corrected coupling weight is written back to the mapping association edge between the device virtual entities and synchronously written to the corresponding energy flow virtual entity, so that the coupling strength in the initial digital twin virtual model corresponds to the actual deviation result in the current time window; after completing the correction of all coupling weights, the target digital twin virtual model is obtained.

[0151] If the model deviation value is not greater than the preset deviation threshold, the initial digital twin virtual model is verified using a multi-energy coupling feature matrix.

[0152] Specifically, during verification, the non-zero coupling weights in the multi-energy coupling feature matrix are first read, and it is checked whether there are virtual devices, virtual energy flow entities, and mapping edges corresponding to the non-zero coupling weights in the initial digital twin virtual model. If there are non-zero coupling weights in the matrix but no corresponding edges in the model, the verification fails. Then, it is checked whether each coupling weight in the model is between 0 and 1. If there are coupling weights less than 0 or greater than 1, the verification fails. Next, it is checked whether the energy flow direction is consistent with the device state mapping vector, that is, the output energy type of the source device should be consistent with the input energy type of the object device. If the two are inconsistent, an energy conversion device must exist as an intermediate virtual entity; otherwise, the verification fails.

[0153] As a preferred embodiment, when the coupling relationship, weight boundary, and energy flow direction are all verified, the initial digital twin virtual model is determined as the target digital twin virtual model. Therefore, when the model deviation value is greater than the deviation threshold, the initial digital twin virtual model is adjusted for coupling weight based on energy conversion efficiency and load response gradient. When the model deviation value is not greater than the deviation threshold, the initial digital twin virtual model is directly used as the target digital twin virtual model after completing the consistency verification through the multi-energy coupling feature matrix.

[0154] Example 8: S7 includes: simulating the future operating state of the park's energy physical system using a target digital twin virtual model; extracting predicted quantities based on the future operating state to obtain an expected supply-demand gap sequence; determining whether the maximum gap value of the expected supply-demand gap sequence is greater than an imbalance threshold; if the maximum gap value is greater than the imbalance threshold, constructing an instruction optimization space based on the expected supply-demand gap sequence and efficient scheduling objectives; and generating initial scheduling instructions by searching in the instruction optimization space. The efficient scheduling objectives include at least one of reducing energy operating costs, improving energy utilization efficiency, and reducing the degree of supply-demand imbalance.

[0155] In this embodiment, when simulating the park's energy physical system using the target digital twin virtual model, the status data of each virtual entity at the current scheduling time is first read, including equipment start / stop status, input power, output power, operating efficiency, energy storage charge status, available capacity, energy input type, energy output type, coupling weight, and energy flow direction.

[0156] Subsequently, state recursion is performed starting from the first future sampling interval. For photovoltaic power generation equipment, the power generation output of the sampling interval is determined based on the current output, rated output, and predicted irradiance data; for gas turbines, the electrical output and waste heat are determined based on the current output, ramp limit, gas input, and power generation efficiency; for gas boilers, the heating capacity is determined based on the gas input and boiler efficiency; for chillers, the cooling capacity is determined based on the input electrical power and cooling efficiency; for heat pump units, the heating capacity is determined based on the input electrical power and heating efficiency; for battery energy storage equipment, the charging and discharging status is determined based on the current energy storage state of charge, rated capacity, allowable charging power, and allowable discharging power. After the state calculation for the first future sampling interval is completed, this state is used as the initial state for the second future sampling interval, and so on, up to the 24th future sampling interval, thereby obtaining the future operating state.

[0157] In the above simulation, the energy flow between energy devices with coupling relationships also needs to be calculated; for any group of energy devices with coupling relationships, first read the output power of the source device in the current sampling interval, and then read the input demand of the affected device in the same sampling interval.

[0158] Specifically, when the output power is less than the input demand, the output power is used as the flowable power; when the output power is not less than the input demand, the input demand is used as the flowable power. Then, the flowable power is multiplied by the coupling weight corresponding to the group of energy devices to obtain the effective flowable power within the sampling interval. After processing all coupling relationships in this way, the target digital twin virtual model can obtain the flow results of electrical energy, thermal energy, cold energy, gas energy and energy storage within each future sampling interval.

[0159] Furthermore, when extracting predicted quantities based on future operating conditions to obtain the expected supply and demand gap sequence, the supply and demand quantities are extracted separately according to the sampling interval.

[0160] Specifically, the power supply is obtained by adding the output of photovoltaic power generation, the power output of gas turbines, the discharge power of batteries, and the power purchased from external sources, and then deducting the charging power of batteries; the power demand is obtained by adding the power load of the park's infrastructure, the power load of charging piles, the power consumption of chillers, and the power consumption of heat pump units; the heat supply is obtained by adding the heat supply of gas boilers, the waste heat recovery of gas turbines, and the heat supply of heat pumps; the heat demand is obtained by adding the building heat load and the process heat load; the cooling supply is obtained by adding the cooling capacity of chillers, the cooling capacity released by cold storage equipment, and the cooling capacity driven by waste heat; the cooling demand is obtained by adding the building cooling load and the process cooling load; the gas supply is determined by the amount of gas purchased from external sources, and the gas demand is obtained by adding the gas consumption of gas turbines and the gas consumption of gas boilers.

[0161] For each sampling interval, the supply and demand difference for that sampling interval is obtained by subtracting the demand from the supply of the corresponding energy type. Then, the supply and demand differences are arranged in the time sequence of the next 24 sampling intervals to obtain the expected supply and demand difference sequence.

[0162] Furthermore, when determining whether the maximum difference value of the expected supply-demand gap sequence is greater than the imbalance threshold, the absolute value of the supply-demand gap for each sampling interval in the expected supply-demand gap sequence is first taken, and then the largest value is selected from the 24 absolute values ​​as the maximum difference value. The imbalance threshold is determined before the current scheduling date.

[0163] Specifically, the process involves: selecting operational data from the 30 calendar days prior to the current scheduling date, where there were no alarms, uninterrupted communication, and completed scheduling commands; calculating the absolute value of the actual supply-demand difference for each sampling interval at a 5-minute sampling interval; summing all the absolute values ​​of the actual supply-demand difference and dividing by the sample size to obtain the historical average difference; then calculating the difference between each absolute value of the actual supply-demand difference and the historical average difference, squaring each difference and averaging it, and finally taking the square root of this average to obtain the historical discrete value of the difference; adding the historical average difference to twice the historical discrete value to obtain the historical imbalance limit. Subsequently, the energy that the backup equipment can compensate for within one sampling interval is read, and the historical imbalance limit is compared with the energy that the backup equipment can compensate for; the smaller of the two values ​​is taken as the imbalance threshold.

[0164] When the maximum difference value is greater than the imbalance threshold, the instruction optimization space construction process begins; when the maximum difference value is not greater than the imbalance threshold, the equipment operation setting value corresponding to the current simulation result is used as the candidate result of the initial scheduling instruction.

[0165] In this embodiment, if the maximum difference value is greater than the imbalance threshold, an instruction optimization space is constructed based on the expected supply and demand difference sequence and the scheduling target.

[0166] Specifically, during the construction process, the adjustable objects are first determined. These adjustable objects include the output of the gas turbine, the output of the gas boiler, the output of the chiller unit, the output of the heat pump unit, the charging power of the battery, the discharging power of the battery, the cold release capacity of the cold storage equipment, the power of external electricity purchase, the amount of external gas purchase, and the transferable load.

[0167] Then, value boundaries are determined for each adjustable object. The lower limit of equipment output is the minimum allowable output of the equipment, and the upper limit of equipment output is the rated output of the equipment; the upper limit of battery charging power is the smaller of the allowable charging power and the power converted from the remaining chargeable capacity; the upper limit of battery discharging power is the smaller of the allowable discharging power and the power converted from the current discharge capacity; the upper limit of external power purchase is the power purchase contract capacity; the upper limit of external gas purchase is the gas purchase contract capacity; the upper limit of transferable load is the user-authorized transfer capacity; then, each value boundary is discretized according to a pre-set adjustment step size, where the equipment output adjustment step size is 5% of the rated output, the energy storage power adjustment step size is 5% of the rated charging and discharging power, the external energy purchase adjustment step size is 5% of the contract capacity, and the transferable load adjustment step size is 5% of the authorized transfer capacity; the discrete value combination of all adjustable objects constitutes the instruction optimization space.

[0168] Furthermore, when searching and generating initial scheduling instructions in the instruction optimization space, each set of candidate instructions is input into the target digital twin virtual model, and the supply and demand status is re-deduced within the next 24 sampling intervals.

[0169] For each set of candidate instructions, the energy operating cost, energy utilization efficiency, and supply-demand imbalance are calculated separately. The energy operating cost is obtained by adding the external electricity purchase cost, external gas purchase cost, equipment start-up and shutdown cost, and equipment operation and maintenance cost. The external electricity purchase cost is obtained by multiplying the purchased electricity volume by the electricity price for the corresponding time period; the external gas purchase cost is obtained by multiplying the purchased gas volume by the gas price per unit; the equipment start-up and shutdown cost is obtained by multiplying the number of start-ups and shutdowns by the cost per start-up and shutdown; and the operation and maintenance cost is obtained by multiplying the equipment operating time by the unit operation and maintenance cost. The energy utilization efficiency is obtained by dividing the total effective output energy of all equipment by the total input energy of all equipment. When the total input energy is 0, the candidate instruction is not considered a valid candidate. The supply-demand imbalance is obtained by adding the absolute values ​​of the supply and demand differences in the next 24 sampling intervals.

[0170] Furthermore, when the scheduling objectives include reducing energy operating costs, improving energy utilization efficiency, and reducing the degree of supply and demand imbalance, the evaluation results of candidate instructions are calculated according to the objective weights.

[0171] In this embodiment, the weight of energy operating cost is set at 0.40, the weight of energy utilization efficiency is set at 0.30, and the weight of supply and demand imbalance is set at 0.30. For energy operating cost, the smaller the cost value, the better the corresponding evaluation result; for energy utilization efficiency, the larger the efficiency value, the better the corresponding evaluation result; for supply and demand imbalance, the smaller the imbalance value, the better the corresponding evaluation result.

[0172] In the specific evaluation, the operating cost, energy utilization efficiency, and supply-demand imbalance of each candidate instruction are first converted into evaluation values ​​of the same dimension, and then multiplied by the corresponding weights and added together to obtain the comprehensive evaluation value of the candidate instruction. After all candidate instructions have been calculated, the group of candidate instructions with the best scheduling target corresponding to the comprehensive evaluation value is selected as the initial scheduling instruction. The initial scheduling instruction includes equipment start-up and shutdown status, equipment output setpoint, battery charging and discharging power, cold storage equipment cooling capacity, external power purchase, external gas purchase, and transferable load adjustment.

[0173] Example 9: S8 includes parsing the expected energy supply sequence and expected energy demand sequence based on the initial scheduling instruction; performing time-slice alignment processing on the expected energy supply sequence and expected energy demand sequence to obtain the supply-demand difference fluctuation range; performing extreme value search on the supply-demand difference fluctuation range and removing outliers to obtain the extreme value of the supply-demand difference; determining whether the extreme value of the supply-demand difference is less than a preset imbalance threshold; if the extreme value of the supply-demand difference is less than the preset imbalance threshold, then the initial scheduling instruction is determined as the target scheduling instruction.

[0174] In this embodiment, when parsing the expected energy supply sequence and expected energy demand sequence according to the initial scheduling instruction, the control quantities of each device are first assigned to the corresponding 5-minute time slice according to the execution start time, execution end time and device number in the instruction.

[0175] Specifically, for electricity supply, the expected electricity supply within a given time slice is obtained by summing the output of photovoltaic power generation, gas turbine power output, battery discharge power, and externally purchased electricity, and then subtracting the energy corresponding to the battery charging power. For heat supply, the expected heat supply is obtained by summing the heat output of gas boilers, waste heat recovery from gas turbines, and heat pumps. For cooling supply, the expected cooling supply is obtained by summing the cooling capacity of chillers, the cooling capacity released by cold storage equipment, and the cooling capacity driven by waste heat. For gas supply, the expected gas supply is obtained by summing the externally purchased gas volume. The supply of each energy type is arranged in the order of the 24 time slices to form the expected energy supply sequence.

[0176] In this embodiment, the expected energy demand sequence is determined jointly by the load projection results of the target digital twin virtual model and the load adjustment amount in the initial scheduling instructions. For each time slice, the electricity demand is obtained by adding the basic power load of the park, the power load of charging piles, the power consumption of chillers and heat pump units, deducting the power load transferred out in that time slice, and adding the power load transferred in from other time slices. The heat demand is obtained by adding the building heat load and the process heat load. The cooling demand is obtained by adding the building cooling load and the process cooling load. The gas demand is obtained by adding the gas consumption of gas turbines and the gas consumption of gas boilers. All of the above demands are arranged in the same 24 time slices as the expected energy supply sequence to form the expected energy demand sequence.

[0177] Furthermore, when aligning the expected energy supply and demand sequences using time slices, the time slice number is used as the alignment benchmark. For the same time slice, only the supply and demand of the same energy type within that time slice are compared, and data from adjacent time slices are not mixed in the calculation. When calculating the supply-demand gap, the supply is subtracted from the demand to obtain the gap value for that energy type in that time slice. A positive gap value indicates that the supply exceeds the demand, while a negative gap value indicates that the supply is insufficient to cover the demand. After calculating the gaps for 24 time slices for each of the following energy types—electricity, heat, cooling, and gas—the supply-demand gap fluctuation ranges are obtained in chronological order.

[0178] Within this supply-demand gap fluctuation range, each time slice has a corresponding difference in electricity, heat, cold, and gas.

[0179] In this embodiment, before performing an extreme value search on the supply and demand difference fluctuation range, an outlier removal limit is first determined.

[0180] Specifically, this limit is calculated from valid historical data within 30 calendar days prior to the current scheduling date. Valid historical data is limited to data where there are no equipment alarms, communication is uninterrupted, and scheduling instructions are executed successfully. Using a 5-minute sampling interval, the actual supply-demand difference within each historical sampling interval is first calculated, then the change in supply-demand difference between adjacent historical sampling intervals is calculated, and the absolute value of the change is taken. The absolute values ​​of all changes are summed and divided by the number of samples to obtain the historical average change value. The difference between each absolute value of change and the historical average change value is squared, and the square root of the average of all squared values ​​is taken to obtain the historical discrete value. The historical average change value is added to three times the historical discrete value, which serves as the outlier removal limit.

[0181] Furthermore, during outlier removal, the supply-demand gap fluctuation range is checked in chronological order.

[0182] Specifically, for time slices 2 through 23, the absolute value of the difference between the supply and demand difference of the current time slice and the supply and demand difference of the previous time slice, as well as the absolute value of the difference between the current time slice and the supply and demand difference of the next time slice, are calculated respectively. When both absolute values ​​of the difference exceed the outlier removal limit, the supply and demand difference of the current time slice is removed as an outlier. For the first time slice, the absolute value of the difference is compared only with the second time slice. For the 24th time slice, the absolute value of the difference is compared only with the 23rd time slice. When the absolute value of the difference of the endpoint time slice exceeds the outlier removal limit, the corresponding endpoint difference value is removed. The remaining difference values ​​after removal constitute the effective supply and demand difference set.

[0183] As a preferred embodiment, when performing extreme value search, the absolute value of each difference value in the effective supply and demand difference set is taken, and the one with the largest value among all absolute values ​​is selected as the extreme value of the supply and demand difference.

[0184] If the park sets imbalance thresholds for electricity, heat, cold energy, and gas energy respectively, then the search is performed separately according to energy type to obtain extreme values ​​for electricity supply and demand, heat supply and demand, cold energy supply and demand, and gas energy supply and demand. If the park adopts a unified verification standard, then heat, cold energy, and gas energy are first converted into a unified energy unit according to the energy conversion factor, and then the converted difference value is included in the extreme value search along with the electricity difference value to obtain a unified supply and demand difference extreme value.

[0185] Furthermore, when determining whether the extreme value of the supply-demand gap is less than the preset imbalance threshold, the imbalance threshold is calculated before the current scheduling date.

[0186] Specifically, the determination method is as follows: Select the operating data of the equipment within 30 natural days before the current scheduling date, where there are no alarms, no communication interruptions, and the scheduling instructions have been executed. Calculate the absolute value of the actual supply and demand difference for each sampling interval according to a 5-minute sampling interval. Add all the absolute values ​​of the actual supply and demand difference and divide by the number of samples to obtain the historical average difference. Then, square the difference between each absolute value of the actual supply and demand difference and the historical average difference, average all the squared values, and take the square root to obtain the historical discrete value of the difference. Add the historical average difference to twice the historical discrete value of the difference to obtain the historical imbalance limit.

[0187] Then, the energy supply that the backup equipment in the park can compensate within one time slice is read, and the historical imbalance limit is compared with the energy supply that the backup equipment can compensate. The smaller value is taken as the preset imbalance threshold. If the verification is performed separately according to the energy type, the imbalance threshold is calculated using their respective historical data and the corresponding backup equipment compensation capacity for electrical energy, thermal energy, cold energy and gas energy.

[0188] During the comparison, the extreme value of the supply-demand difference is numerically judged against the corresponding imbalance threshold. When the extreme value of the supply-demand difference is less than the preset imbalance threshold, it means that the supply-demand deviation formed by the initial scheduling instruction within the next 24 time slices has not exceeded the allowable boundary, and the initial scheduling instruction meets the execution conditions. At this time, the initial scheduling instruction is determined as the target scheduling instruction, and the equipment start-up and shutdown status, equipment output setpoint, battery charging and discharging power, cold storage equipment cooling capacity, external power purchase, external gas purchase, and transferable load adjustment are written into the target scheduling instruction.

[0189] Example 10: S9 includes parsing the initial scheduling command according to the efficient scheduling objective if the extreme value of the supply-demand difference is not less than the preset imbalance threshold, obtaining the equipment response delay and energy conversion loss; constructing the command correction vector based on the equipment response delay and energy conversion loss; processing the initial scheduling command using the command correction vector to obtain the correction magnitude; updating the initial scheduling command according to the correction magnitude to obtain the target scheduling command, and outputting the target scheduling command to control various energy devices in the park's energy physical system to perform park energy scheduling.

[0190] In this embodiment, the preset imbalance threshold is determined before the current scheduling date.

[0191] Specifically, historical operational data from the 30 calendar days prior to the current scheduling date, where there were no equipment alarms, no communication interruptions, and all scheduling instructions were executed, is selected. The absolute value of the actual supply-demand difference for each sampling interval is calculated at a 5-minute sampling interval. The sum of all the absolute values ​​of the actual supply-demand difference is then divided by the number of samples to obtain the historical average difference. The difference between each absolute value of the actual supply-demand difference and the historical average difference is then calculated. The square of each difference is taken, and the square root of the average of all squared values ​​is taken to obtain the historical discrete value of the difference. The historical average difference is then added to twice the historical discrete value of the difference to obtain the historical imbalance limit.

[0192] Then, the energy supply that the backup equipment in the park can compensate within one time slice is read, the historical imbalance limit is compared with the energy supply that the backup equipment can compensate, and the smaller value is taken as the preset imbalance threshold.

[0193] Furthermore, when the extreme value of the supply-demand gap is not less than the preset imbalance threshold, it indicates that there is a time slice in the future operation result corresponding to the initial scheduling instruction where the supply is insufficient or the supply surplus exceeds the allowable boundary.

[0194] At this point, the initial scheduling instructions are first analyzed based on the goal of efficient scheduling. The analysis objects include equipment start-up and shutdown times, equipment output setpoints, energy storage charging power, energy storage discharging power, cold storage equipment cooling capacity, external power purchase, external gas purchase, and transferable load adjustment. For the goal of reducing energy operating costs, the external power purchase, external gas purchase, equipment start-up and shutdown times, and equipment operating time are extracted for each time slice. For the goal of improving energy utilization efficiency, the input energy and effective output energy of each energy conversion device are extracted. For the goal of reducing the degree of supply and demand imbalance, the absolute value of the supply and demand difference is extracted for each time slice.

[0195] In this embodiment, the device response delay is determined by the time the instruction is issued and the time when the device reaches the instruction state.

[0196] Specifically, for gas turbines, the arrival time is defined as the moment when their actual electrical output reaches the commanded electrical output; for gas boilers, the arrival time is defined as the moment when their heating output reaches the commanded heating output; for chillers, the arrival time is defined as the moment when their cooling capacity reaches the commanded cooling capacity; for heat pump units, the arrival time is defined as the moment when their heating output reaches the commanded heating output; and for battery energy storage devices, the arrival time is defined as the moment when their charging or discharging power reaches the commanded power. The response delay of the device is obtained by subtracting the command issuance time from the arrival time. If the device does not reach the commanded state within one time slice, the response delay within that time slice is calculated by subtracting the command issuance time from the end time of that time slice.

[0197] As a preferred embodiment, energy conversion losses are determined by the difference between the device's input energy and its effective output energy.

[0198] For gas turbines, the input energy is the energy converted from the gas input, and the effective output energy is the sum of electrical output energy and recoverable waste heat energy. For gas boilers, the input energy is the energy converted from the gas input, and the effective output energy is the heating energy. For chillers, the input energy is the electricity consumption, and the effective output energy is the cooling capacity. For heat pump units, the input energy is the electricity consumption, and the effective output energy is the heating capacity. For each time slice, the energy conversion loss of the equipment in that time slice is obtained by subtracting the effective output energy from the input energy. The total energy conversion loss of the equipment in the future scheduling cycle is obtained by adding up the energy conversion losses of the 24 time slices.

[0199] Furthermore, when constructing the instruction correction vector based on equipment response delay and energy conversion loss, one correction item is set for each controlled device according to the device number sequence. Each correction item records the device number, correction direction, response delay, total energy conversion loss, adjustable capacity, and constraint boundary. The correction direction is determined based on the sign of the supply-demand difference and the efficient scheduling objective: when the supply-demand difference of the corresponding time slice is negative, the correction direction is to increase supply or decrease demand; when the supply-demand difference of the corresponding time slice is positive, the correction direction is to decrease supply or increase demand. For the objective of reducing energy operating costs, the output of devices with unit operating costs greater than the current average unit operating cost is corrected first. For the objective of improving energy utilization efficiency, the output of devices with energy conversion loss ratios greater than the current average loss ratio of devices is corrected first. For the objective of reducing the degree of supply-demand imbalance, the output of devices that can complete the response within the current time slice or the load adjustment amount are corrected first.

[0200] Furthermore, when processing the initial scheduling command using the instruction correction vector, the energy to be compensated for each controlled device is calculated item by item.

[0201] For devices with response delays, the output difference that the device failed to reach within the delay time is first calculated, and then this output difference is multiplied by the delay time to obtain the delayed energy deficit. For devices with energy conversion losses, the input energy is multiplied by the proportion of energy conversion losses to input energy to obtain the loss compensation energy. The delayed energy deficit and the loss compensation energy are added together to obtain the energy to be compensated for the corresponding device. The energy to be compensated is used to determine the correction range.

[0202] In this embodiment, the correction range is calculated separately according to the device type.

[0203] Specifically, for power generation equipment and cold / heat source equipment, the energy to be compensated is divided by the corresponding time slice length to obtain the required output adjustment amount, which is then rounded according to the equipment output adjustment step size to form the equipment output correction range; the equipment output adjustment step size is 5% of the rated output of the equipment; for battery energy storage equipment, the energy to be compensated is divided by the time slice length, and the energy storage charging and discharging correction range is determined in combination with the allowable charging power, allowable discharging power, and the current energy storage state of charge; for external power purchase interfaces and external gas purchase interfaces, the energy to be compensated is converted into an external energy purchase correction amount and limited to the power purchase contract capacity and gas purchase contract capacity; for transferable load control units, the energy to be compensated is converted into a load transfer amount and limited to the user-side authorized transfer capacity.

[0204] Furthermore, when updating the initial scheduling instructions based on the correction magnitude, they are executed in the order of time slices.

[0205] If a time slice experiences a supply shortage, the output of equipment with a response delay of 0 and not yet at its rated output will be increased first. Once the equipment reaches its rated output, the battery discharge power will be increased. Once the battery discharge power reaches its allowable discharge power, the external power or gas purchase volume will be increased. If the supply shortage is still not eliminated, the adjustable load of that time slice will be reduced according to the user-side authorized capacity. If a time slice has a supply surplus, the output of equipment with a unit operating cost greater than the current average unit operating cost will be reduced first. Once the equipment reaches its minimum allowable output, the battery charging power or the cold storage capacity of the cold storage equipment will be increased. If the supply surplus is still not eliminated, transferable loads will be transferred to that time slice.

[0206] Furthermore, after each correction is completed, the corrected instructions are input into the target digital twin virtual model to recalculate the supply and demand difference for the corresponding time slice, and the extreme value of the supply and demand difference is searched again.

[0207] Specifically, when the recalculated extreme value of the supply-demand gap is less than the preset imbalance threshold, the correction stops; when all adjustable objects reach the constraint boundary and the extreme value of the supply-demand gap is still not less than the preset imbalance threshold, the completed correction results are retained, and the unresolved supply-demand gap is marked as information awaiting manual confirmation.

[0208] Furthermore, after the correction is completed, the corrected equipment start-up and shutdown status, equipment output setting value, energy storage charging power, energy storage discharging power, cold storage equipment cooling capacity, external power purchase, external gas purchase, and transferable load adjustment amount will be written into the target dispatch command.

[0209] Before outputting, the target scheduling command is subject to boundary verification: the equipment output should be between the minimum allowable output and the rated output, the energy storage state of charge should be between the lower allowable limit and the upper allowable limit, the external power purchase and external gas purchase should not exceed the contracted capacity, and the transferable load adjustment should not exceed the user-side authorized capacity; after the boundary verification is passed, the park energy management platform will issue the target scheduling command to the corresponding energy equipment according to the equipment number, and control the park energy physical system to execute the park energy scheduling.

[0210] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A computational method for efficient energy scheduling in industrial parks that integrates digital twins, characterized in that, include: S1. Obtain the continuous operating status sequence and real-time load fluctuation sequence of the park's energy physical system, which includes various energy devices; S2. Extract the time-varying operation characteristics of the park's energy physical system to obtain a feature set; S3. For the feature set, identify the multi-energy coupling relationship between different energy devices and obtain the multi-energy coupling feature matrix; S4. Based on the multi-energy coupling characteristic matrix, the operating status of various energy devices, and the energy flow relationship between various energy devices, construct an initial digital twin virtual model; S5. Obtain the prediction sequence output by the initial digital twin virtual model within the current time window, compare the prediction sequence with the actual operation sequence of the park's energy physics system within the current time window, and determine the model deviation value. S6. If the model deviation value is greater than the preset deviation threshold, the coupling weight of the initial digital twin virtual model is corrected to obtain the target digital twin virtual model; if the model deviation value is not greater than the preset deviation threshold, the model stability is verified, and the initial digital twin virtual model that passes the verification is determined as the target digital twin virtual model. S7. Simulate the future operating state of the park's energy physical system using the target digital twin virtual model to obtain the expected supply and demand difference sequence and generate initial scheduling instructions; S8. Determine the extreme value of the supply-demand gap corresponding to the initial scheduling instruction; if the extreme value of the supply-demand gap is less than the preset imbalance threshold, determine the initial scheduling instruction as the target scheduling instruction. S9. If the extreme value of the supply-demand gap is not less than the preset imbalance threshold, the initial scheduling instruction is modified according to the expected supply-demand gap sequence and the efficient scheduling objective to obtain the target scheduling instruction.

2. The method for efficient energy scheduling calculation in a park integrating digital twins as described in claim 1, characterized in that: S1 includes: Obtain equipment operation logs to generate a continuous operating status sequence, and collect real-time load data to obtain a real-time load fluctuation sequence; Align the continuous operating state sequence with the real-time load fluctuation sequence to obtain the state fluctuation mapping relationship; An energy operation map is constructed based on the state fluctuation mapping relationship. If there are isolated nodes in the energy operation map, the isolated nodes are removed to obtain an updated energy operation map. The updated energy operation map provides a continuous sequence of the park's energy physical systems' operating status and a real-time load fluctuation sequence over a historical period.

3. The method for efficient energy scheduling calculation in a park integrating digital twins as described in claim 1, characterized in that: S2 includes: Obtain continuous operating status sequences and real-time load fluctuation sequences; The continuous operating status sequence and the real-time load fluctuation sequence are synchronized and aligned to obtain the synchronized operating status sequence and the synchronized load fluctuation sequence. The sequence is truncated based on the synchronous operating state sequence and the synchronous load fluctuation sequence to obtain multidimensional state subsequence and multidimensional load subsequence; A feature set is obtained by extracting time-varying operational features from multidimensional state subsequences and multidimensional load subsequences using a long short-term memory network.

4. The method for efficient energy scheduling calculation in a park that integrates digital twins according to claim 1, characterized in that: S3 includes: Obtain the feature set and construct the initial feature vector; The initial feature vector is received through a random forest model to obtain a sequence of feature contributions. Calculate the coupling weights between different energy devices based on the characteristic contribution sequence; If the coupling weights meet the preset matching conditions, the multi-energy coupling relationship between different energy devices is determined. The multi-energy coupling relationship is matrixed to obtain the multi-energy coupling feature matrix.

5. The method for efficient energy scheduling calculation in a park that integrates digital twins according to claim 1, characterized in that: S4 includes: The multi-energy coupling feature matrix and the operating status of various energy devices are obtained. The multi-energy coupling feature matrix is ​​used to extract coupling weights, and the operating status is used to extract state mapping vectors. The energy flow relationship is determined based on the coupling weights and the state mapping vector. Determine whether the coupling weight exceeds a preset threshold; If the coupling weight exceeds a preset threshold, a flow node corresponding to the energy flow relationship is generated; A topological connection graph is constructed based on the flow nodes and energy flow relationships. The topological connection graph is used to generate virtual entities and map associated edges. An initial digital twin virtual model corresponding to the park's energy physical system is constructed using virtual entities and mapping associated edges.

6. The method for efficient energy scheduling calculation in a park integrating digital twins according to claim 1, characterized in that: S5 includes: Obtain the prediction sequence and actual execution sequence of the initial digital twin virtual model; The dynamic time warping algorithm is used to process the predicted sequence and the actual running sequence to obtain the sequence alignment matrix; Generate an error distribution vector based on the sequence alignment matrix; Determine whether the error distribution vector exceeds a preset threshold; If the error distribution vector exceeds a preset threshold, the cumulative deviation is extracted. The model deviation value is determined by using the cumulative deviation.

7. The method for efficient energy scheduling calculation in a park integrating digital twins as described in claim 1, characterized in that: S6 includes: Determine whether the model deviation value is greater than the preset deviation threshold; If the model deviation value is greater than the preset deviation threshold, a weight update vector is generated based on the energy conversion efficiency and load response gradient in the multi-energy coupling feature matrix. The coupling weights of the initial digital twin virtual model are corrected by updating the weight vector to obtain the target digital twin virtual model. If the model deviation value is not greater than the preset deviation threshold, the initial digital twin virtual model that passes the verification using the multi-energy coupling feature matrix will be determined as the target digital twin virtual model.

8. The method for efficient energy scheduling calculation in a park integrating digital twins according to claim 1, characterized in that: S7 includes: The future operating status of the park's energy physical system is obtained by using a target digital twin virtual model; The expected supply and demand gap sequence is obtained by extracting predicted quantities based on future operating conditions. Determine whether the maximum difference in the expected supply-demand gap sequence is greater than the imbalance threshold; If the maximum difference value is greater than the imbalance threshold, then construct the instruction optimization space based on the expected supply and demand difference sequence and the goal of efficient scheduling; An initial scheduling instruction is generated by searching the instruction optimization space, wherein the efficient scheduling objective includes at least one of reducing energy operating costs, improving energy utilization efficiency, and reducing the degree of supply and demand imbalance.

9. The method for efficient energy scheduling calculation in a park that integrates digital twins according to claim 1, characterized in that: S8 includes: The expected energy supply sequence and expected energy demand sequence are analyzed based on the initial dispatch instructions; Time-slice alignment is performed on the expected energy supply series and the expected energy demand series to obtain the fluctuation range of the supply-demand difference. By performing extreme value search on the fluctuation range of the supply and demand gap and removing outliers, the extreme values ​​of the supply and demand gap can be obtained. Determine whether the extreme value of the supply-demand gap is less than the preset imbalance threshold; If the extreme value of the supply-demand gap is less than the preset imbalance threshold, the initial scheduling instruction will be determined as the target scheduling instruction.

10. The method for efficient energy scheduling calculation in a park by integrating digital twins according to claim 1, characterized in that: S9 includes: If the extreme value of the supply-demand gap is not less than the preset imbalance threshold, the initial scheduling instruction is parsed according to the efficient scheduling objective to obtain the equipment response delay and energy conversion loss. Construct a command correction vector based on equipment response delay and energy conversion loss; The initial scheduling instruction is processed using the instruction correction vector to obtain the correction magnitude; The target scheduling instruction is obtained by updating the initial scheduling instruction based on the correction magnitude, and the target scheduling instruction is output to control various energy devices in the park's energy physical system to perform park energy scheduling.

Images