Industrial load resource management system and method based on cost optimization and response time effectiveness

Abstract

This invention discloses an industrial load resource management system based on cost optimization and response time. It includes a load identification and key parameter extraction module that decomposes electrical operation data to extract key process parameters for each adjustable load; a load resource dynamic modeling module that divides all adjustable loads into interruptible loads and shiftable loads, and constructs corresponding interruption compensation cost functions and shift cost functions; a collaborative optimization and decision-making module that establishes a collaborative optimization model based on the interruption compensation cost function and shift cost function, and uses an improved swarm intelligence optimization algorithm to solve for the industrial load scheduling scheme; and a scheduling execution and performance evaluation module that executes the industrial load scheduling scheme, and calculates the response performance index of each adjustable load based on the actual response delay and actual compensation cost, as well as the corresponding promised response time and promised compensation cost. This invention improves the reliability, economy, and regulation response capability of the virtual power plant.

Description

Technical Field

[0001] This invention relates to the field of power load dispatching technology, specifically to an industrial load resource management system and method based on cost optimization and response time. Background Technology

[0002] In the operation of virtual power plants, aggregating adjustable load resources from industrial users to participate in grid regulation is an important means to improve system flexibility and economy. Currently, a static hierarchical compensation model is commonly used, classifying interruptible loads according to preset response time levels and binding them to fixed unit compensation costs. The scheduling process essentially involves resource selection and combination under a given cost function, but it neglects the profound impact of dynamic field factors such as real-time equipment thermal status and process buffer conditions on the response capability of adjustable loads. This leads to a discrepancy between the promised response time of adjustable loads and the actual executable response delay. To ensure the timeliness of grid regulation, high-cost level resources are often frequently called upon, or the risk of response timeouts must be accepted. Simultaneously, there is a lack of proactive value quantification and guidance mechanisms for transferable loads, making it difficult to reduce dependence on high-cost interruptible loads through scheduling strategies. Furthermore, due to the lack of a mechanism for closed-loop feedback and dynamic adjustment of resource parameters based on historical response performance, virtual power plants face the dilemma of persistently high regulation costs and difficulty in synergistically improving the overall reliability and response economy of the resource pool during long-term operation, easily leading to insufficient electricity market service capacity. Summary of the Invention

[0003] The purpose of this invention is to provide an industrial load resource management system and method based on cost optimization and response time. This invention improves the reliability, economy and regulation response capability of virtual power plants.

[0004] To achieve this objective, the present invention provides an industrial load resource management system based on cost optimization and response time, comprising: The load identification and key parameter extraction module is used to decompose the electrical operation data of industrial loads using a non-intrusive load identification model, obtain the independent operation curves of each adjustable load, and extract the key process parameters corresponding to each adjustable load from the independent operation curves. The load resource dynamic modeling module is used to divide all adjustable loads into interruptible loads and shiftable loads based on the key process parameters, and to construct an interruption compensation cost function for interruptible loads and a shift cost function for shiftable loads. The collaborative optimization and decision-making module is used to establish a collaborative optimization model based on the interruption compensation cost function and the translation cost function; based on the scenario set generated by load and electricity price forecasts, the improved swarm intelligence optimization algorithm is used to solve the collaborative optimization model to obtain an industrial load scheduling scheme. The scheduling execution and performance evaluation module is used to execute the industrial load scheduling scheme. Based on the actual response delay and actual compensation cost of each adjustable load, as well as the promised response time and promised compensation cost of the corresponding adjustable load in the industrial load scheduling scheme, the module calculates the response performance index of each adjustable load.

[0005] Preferably, the feedback optimization module is used to dynamically adjust the response time level and compensation cost parameters of the interruptible load according to the response performance index, and update the interruption compensation cost function and the collaborative optimization model according to the adjusted response time level and compensation cost parameters.

[0006] Preferably, the specific process for obtaining the key process parameters corresponding to each adjustable load includes: A non-intrusive load identification model is constructed by fusing convolutional neural networks and long short-term memory networks. The electrical operation data of industrial loads are decomposed through the non-intrusive load identification model to obtain the independent operation curves of each adjustable load. By fitting the transient process segment representing load start-up or shutdown in the independent operation curve with an exponential decay model, the time constant obtained from the fitting is used as the equipment start-up and shutdown inertia time. The buffer capacity is obtained by calculating the product of the power fluctuation standard deviation of the corresponding production process buffer stage in the independent operation curve and the duration of the production process buffer stage. Based on the equipment start-up and shutdown inertia time, the buffer capacity, and the preset safe shutdown threshold, the key process parameters corresponding to each adjustable load are obtained.

[0007] Preferably, the specific process for constructing an interruption compensation cost function for interruptible loads includes: Based on the key process parameters, interruptible loads are divided from the adjustable loads. The interruptible loads refer to equipment with a minimum safe interruption duration ≤ A and a process allowable number of interruptions ≥ B. For interruptible loads, an interruption compensation cost function is constructed with the promised response time as the independent variable. The expression for the interruption compensation cost function is as follows: in, This indicates the promised response time for interruptible loads. Indicates the basic compensation cost. A Indicates the range of cost adjustments. k Indicates the promised response time The attenuation coefficient, e It is a natural constant. This indicates the expected cost of compensation for the disruption.

[0008] Preferably, the specific process for constructing a translation cost function for a transferable load includes: Based on the key process parameters, the adjustable load is divided into movable loads. The movable loads refer to equipment with a single movable time ≥ D and a daily cumulative movable amount ≤ E% of the equipment's rated operating time. For the relocatable load, a relocation cost function is constructed with the relocation period as the independent variable. The expression of the relocation cost function is as follows: in, This represents the expected relocation compensation cost of the relocatable load. Indicates the cost of basic translation compensation. Indicates the incentive coefficient. h Indicates the start time of the translation. This represents the set of periods with low load. a This indicates the discount rate.

[0009] Preferably, the specific process for establishing a collaborative optimization model based on the interruption compensation cost function and the translation cost function is as follows: Based on the interruption compensation cost function and the translation cost function, a day-ahead scheduling layer model is constructed with the objective of minimizing the expected total cost. This model considers the expected interruption compensation cost and the expected translation compensation cost, combined with the set end-to-end response delay constraints. This yields the baseline plan for each adjustable load; The end-to-end response delay constraint is expressed as follows: in, For communication transmission delay, The estimated device execution delay accumulation value is based on the safety control timing instructions in the preset control parameter package. Thresholds specified for power grid dispatching; Within a preset time resolution, adjustable load forecasts and real-time electricity price signals are received. Based on the interruption compensation cost function, the shift cost function, adjustable load forecasts, and real-time electricity price signals, an intraday rolling adjustment layer model is established with the goal of minimizing actual adjustment costs, thereby obtaining real-time adjustment instructions for each adjustable load. Based on the baseline plans for each adjustable load obtained from the day-ahead scheduling layer model and the real-time adjustment instructions for each adjustable load obtained from the intraday rolling adjustment layer model, a day-ahead and intraday dual-layer collaborative optimization model containing end-to-end response delay constraints is jointly constructed.

[0010] Preferably, based on the scenario set generated by load and electricity price forecasts, the improved swarm intelligence optimization algorithm is used to solve the collaborative optimization model to obtain the industrial load scheduling scheme. The specific process is as follows: The virtual power plant uses a long short-term memory network model to generate load demand and electricity price forecast curves for future cycles using historical adjustable load data, meteorological elements, and date types. Based on the load demand and electricity price forecast curves, an initial scenario set is generated using the Latin hypercube sampling method. The initial scenario set is then processed by scenario reduction to obtain a representative scenario set. The day-ahead and intraday two-layer collaborative optimization model, which includes end-to-end response delay constraints, is combined with a representative set of scenarios and solved using an improved swarm intelligence optimization algorithm: Initialize a population, where individuals in the population represent scheduling decision variables using a hybrid encoding method. The first individual in the population... The bits are integers that represent the response time level of the interruptible load, where Number of interruptible loads; population individuals after The bit is a real number encoding, representing the transfer period of the transferable load, where The number of loads that can be moved; Based on the ratio of the current iteration number to the maximum iteration number, the inertia weight is dynamically adjusted using a quadratic decay formula, which is: in, The inertia weight at the current iteration time. As the initial inertia weight, To terminate the inertia weight, This represents the current iteration number. This represents the maximum number of iterations. After each iteration update, the normalized diversity of the population is calculated. If the diversity is lower than a preset threshold, simulated binary crossover and polynomial mutation operations are performed on a randomly selected subset of individuals. Calculate individual fitness; if an individual violates the end-to-end response delay constraint, add a penalty term. The formula for the penalty term is: in, The penalty coefficient is... For end-to-end response delay constraints, Thresholds specified for power grid dispatching V For non-compliant adjustable load sets, For adjustable load k power, Total regulating power; Repeat the above iterative process until the termination condition is met to obtain the industrial load scheduling scheme corresponding to the individual with the best fitness.

[0011] Preferably, the specific process for obtaining the response performance indicators of each adjustable load is as follows: The industrial load scheduling scheme is sent to the edge controller on the user side. The edge controller performs local security verification on the instructions in the industrial load scheduling scheme according to the preset control parameter package. After the verification is passed, it drives each adjustable load to perform the corresponding operation and collects the actual response delay and actual compensation cost of each adjustable load. For each scheduled adjustable load, based on the actual response delay and actual compensation cost of the corresponding adjustable load, as well as the promised response time and promised compensation cost of the corresponding adjustable load in the industrial load scheduling scheme, the response performance index of the corresponding adjustable load is calculated: in, As an indicator of the response performance of adjustable loads, The promised response time for adjustable loads, The actual response delay for adjustable loads, To compensate for the cost of adjustable load commitments, This represents the actual compensation cost for adjustable loads.

[0012] An industrial load resource management method based on cost optimization and response timeliness includes the following steps: The electrical operation data of industrial loads are decomposed using a non-intrusive load identification model to obtain independent operating curves for each adjustable load. Key process parameters corresponding to each adjustable load are then extracted from these independent operating curves. Based on the key process parameters, all adjustable loads are divided into interruptible loads and shiftable loads. An interruption compensation cost function is constructed for the interruptible loads, and a shift cost function is constructed for the shiftable loads. Based on the interruption compensation cost function and the shift cost function, a collaborative optimization model is established; based on the scenario set generated by load and electricity price prediction, the collaborative optimization model is solved using an improved swarm intelligence optimization algorithm to obtain an industrial load scheduling scheme. By implementing the industrial load scheduling scheme, the response performance index of each adjustable load is calculated based on the actual response delay and actual compensation cost of each adjustable load, as well as the promised response time and promised compensation cost of the corresponding adjustable load in the industrial load scheduling scheme.

[0013] A computer program product includes a computer program that, when executed by a processor, implements the steps of the above-described method.

[0014] The beneficial effects of this invention are: This invention accurately extracts key process parameters of industrial loads through non-intrusive load identification, enabling refined classification and dynamic cost modeling of load resources. It constructs interruption compensation cost functions and translation cost functions that align with the actual operating characteristics of equipment, resolving the mismatch between traditional static compensation models and the dynamic response capabilities of industrial sites. Furthermore, by establishing a day-ahead and intraday dual-layer collaborative optimization model with end-to-end response delay constraints, combined with an improved swarm intelligence optimization algorithm and a scenario reduction method that retains key risk scenarios, this invention generates an industrial load dispatching scheme that satisfies both the hard time constraints of the power grid and achieves optimal cost, ensuring the on-site executability and economic rationality of the industrial load dispatching scheme. Through a local pre-set control parameter package verification mechanism on the edge controller, this invention significantly reduces the delay in dispatching command execution, effectively bridging the gap between the promised response capability and the actual execution effect of the load, improving the efficiency of dispatching command execution and the safety of equipment operation. Finally, based on actual load response data, this invention calculates response performance indicators, forming a closed-loop feedback mechanism to dynamically adjust the response time level and compensation cost parameters, driving the virtual power plant load resource pool towards continuous evolution towards high reliability and low cost, thereby enhancing the long-term reliability, economy, and market service competitiveness of the virtual power plant. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the structure of the present invention; Figure 2 This is a flowchart of the present invention. Detailed Implementation

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: Example 1 An industrial load resource management system based on cost optimization and response time, such as Figure 1 As shown, it includes: The load identification and key parameter extraction module is used to decompose the electrical operation data of industrial loads (including three-phase voltage, current, active power, and switch status data of industrial loads) using a non-intrusive load identification model that characterizes the physical characteristics and process constraints of loads. This decomposes the data to obtain independent operating curves for each adjustable load (adjustable loads refer to industrial equipment that can be adjusted in the virtual power plant resource pool). From these independent operating curves, key process parameters (quantitative indicators that characterize the core physical characteristics, safety boundaries, and production process constraints of industrial equipment) corresponding to each adjustable load are extracted. This design uses a non-intrusive load identification model to decompose the electrical operation data of industrial loads and extract key process parameters. It can accurately perceive the physical characteristics and production process constraints of each adjustable load, providing an accurate data foundation for subsequent load classification and cost modeling, and avoiding modeling deviations caused by ambiguity in the understanding of load characteristics. The load resource dynamic modeling module is used to divide all adjustable loads into interruptible loads and shiftable loads based on the key process parameters. It constructs an interruption compensation cost function with the promised response time as the independent variable for interruptible loads and a shift cost function with the shift period as the independent variable for shiftable loads. This design divides interruptible loads and shiftable loads based on key process parameters and constructs corresponding cost functions to achieve differentiated and accurate quantification of load resource adjustment costs, which can reflect the economic costs of different load adjustment behaviors. The collaborative optimization and decision-making module is used to establish a collaborative optimization model with end-to-end response delay constraints based on the interruption compensation cost function and the shift cost function. Based on the representative scenario set generated by load and electricity price forecasts, the collaborative optimization model is solved using an improved swarm intelligence optimization algorithm to obtain an industrial load dispatching scheme. This design establishes a collaborative optimization model based on the interruption compensation cost function and the shift cost function, and solves the industrial load dispatching scheme by combining the load and electricity price forecast scenario set and the improved swarm intelligence optimization algorithm. It can achieve optimal control of dispatching costs while meeting the grid response timeliness requirements, thereby improving the timeliness and economy of industrial load dispatching. Based on the specific details of the obtained industrial load dispatching scheme, some optimized technical solutions include: An industrial load dispatching scheme includes instructions on when to dispatch which adjustable load, as well as the promised response time and promised compensation cost for each dispatched adjustable load.

[0017] The scheduling execution and performance evaluation module is used to execute the industrial load scheduling scheme. Based on the actual response delay and actual compensation cost of each adjustable load, as well as the promised response time and promised compensation cost of the corresponding adjustable load in the industrial load scheduling scheme, the module calculates the response performance index of each adjustable load. This design uses the calculated response performance index to quantitatively evaluate the actual execution effect of the scheduling instructions in each industrial load scheduling scheme, providing a scientific basis for the authenticity and reliability of the scheduling execution effect.

[0018] In the above technical solution, the feedback optimization module is used to dynamically adjust the response time level and compensation cost parameters of the interruptible load according to the response performance index, and update the interruption compensation cost function and the collaborative optimization model according to the adjusted response time level and compensation cost parameters. Regarding the specific methods for updating the interruption compensation cost function and the collaborative optimization model, some optimized technical solutions include: The system continuously receives the calculated response performance indicators for each interruptible load and calculates the arithmetic mean of the historical response performance indicators for each interruptible load based on its own historical statistical distribution (usually calculated using valid data from the last 90 calendar days). with standard deviation Dynamically set the specific increase and decrease judgment thresholds for each interruptible load; The formula for calculating the threshold for raising the judgment level is as follows: in, To increase the judgment threshold, This is the arithmetic mean of historical response performance indicators. The standard deviation of historical response performance indicators. This is the first preset coefficient; The formula for calculating the reduction of the judgment threshold is: in This is the second preset coefficient. To lower the judgment threshold; The sum of the increased and decreased judgment thresholds is greater than zero to ensure a false judgment interval between thresholds, avoiding misadjustment due to accidental fluctuations; when the response performance index of a certain interruptible load is detected to be higher than the specific increased judgment threshold for multiple consecutive evaluation periods (e.g., 3 periods). When this happens, the response time level of the corresponding interruptible load will be automatically upgraded by one level (e.g., from T2 to T1), and the unit compensation cost parameter will be correspondingly reduced (by 5%). If the response performance index of the interruptible load is lower than the reduction threshold for several consecutive cycles... If the response time level of the corresponding interruptible load is reduced by one level, the corresponding compensation cost parameter is increased (e.g., increased by 8%). The adjusted response time level and unit compensation cost parameter of each interruptible load are fed back in real time through the internal data bus to update the specific expression of the interruption compensation cost function corresponding to the interruptible load. In the next round of scheduling optimization calculation, the updated interruption compensation cost function is called to build a new collaborative optimization model. The above design dynamically adjusts the interruptible load response time level and compensation cost parameters based on response performance indicators, and updates the interruption compensation cost function and collaborative optimization model. It constructs a complete closed-loop feedback mechanism from scheduling execution to resource modeling and optimization decision-making, which can continuously optimize itself based on the actual load response performance, continuously improve the adaptability of industrial load scheduling schemes to actual load characteristics, and improve the overall economic efficiency of virtual power plant scheduling.

[0019] The specific process for obtaining the key process parameters corresponding to each adjustable load in the above technical solution includes: A non-intrusive load identification model is constructed by fusing convolutional neural networks and long short-term memory networks. The electrical operation data of industrial loads are decomposed through the non-intrusive load identification model to obtain the independent operation curves of each adjustable load. Regarding the specific methods for constructing non-intrusive load identification models, some optimized technical solutions include: A deep learning model integrating convolutional neural networks and long short-term memory networks is constructed. The convolutional neural network layer is responsible for automatically extracting local features and spatial patterns (such as transient power features when specific equipment starts and stops) from the input total load sequence data. The long short-term memory network layer is used to capture the long-term dependence and dynamic change law of adjustable load power consumption behavior in the time dimension, thereby constructing a non-intrusive load identification model. Regarding the specific methods for obtaining the independent operating curves of each adjustable load, some optimized technical solutions include: By deploying intelligent measurement terminals in user distribution cabinets, the three-phase voltage, current, active power, and harmonic distortion rate at the main industrial line are collected as electrical operation data to form continuous total load sequence data. The collected continuous 24-hour total load sequence data is input into a trained non-intrusive load identification model. The non-intrusive load identification model decomposes the aggregated signal through the learned feature representation and finally outputs the trajectory of the independent active power of each adjustable industrial device over time, i.e., the independent operating curve of each adjustable load. By fitting an exponential decay model to the transient process segment representing load start-up or shutdown in the independent operating curve, the time constant obtained from the fitting is used as the equipment start-up and shutdown inertia time. The buffer capacity is obtained by calculating the product of the power fluctuation standard deviation of the corresponding production process buffer stage in the independent operating curve and the duration of the production process buffer stage. Based on the equipment start-up and shutdown inertia time, the buffer capacity, and a preset safe shutdown threshold (the safe shutdown threshold is directly set according to the equipment manufacturer's technical manual), the key process parameters corresponding to each adjustable load are obtained. The above design constructs a non-intrusive load identification model by fusing convolutional neural networks and long short-term memory networks, and calculates the equipment start-up and shutdown inertia time and buffer capacity through exponential decay model fitting. This improves the scientific nature of the key process parameter extraction. The obtained key process parameters can improve the accuracy of load identification and make the load characteristic description more consistent with the actual operating law of industrial equipment.

[0020] In the above technical solution, the specific process of constructing the interruption compensation cost function for interruptible loads includes: Based on the key process parameters, interruptible loads are divided from the adjustable loads. Interruptible loads refer to equipment with a minimum safe interruption duration ≤ A (A is 30 minutes) and a process allowable number of interruptions ≥ B (B is 1 time / day). Interruptible loads are equipment that can be interrupted for a short time under safe conditions and can be quickly restored after the interruption (such as electric arc furnaces and rolling mills). For interruptible loads, an interruption compensation cost function is constructed with the promised response time as the independent variable. The expression for the interruption compensation cost function is as follows: in, This indicates the promised response time for interruptible loads. This represents the basic recovery cost (the basic recovery cost represents the inherent cost incurred to interrupt production). A Indicates the range of cost adjustments. k Indicates the promised response time The attenuation coefficient (used to quantify the sensitivity of compensation costs to changes in promised response time). e It is a natural constant. This indicates the expected cost of compensation for the disruption. The interruption compensation cost function determines the unit compensation cost based on the response time level of the promised response time of the interruptible load, where the shorter the response time level, the higher the unit compensation cost. Regarding the specific methods for classifying response time levels, some optimized technical solutions include: The promised response time for interruptible loads can be classified into three levels: T1 (promised response time ≤ 5 minutes), T2 (5 minutes < promised response time ≤ 15 minutes), and T3 (15 minutes < promised response time ≤ 30 minutes); the unit compensation cost is set as follows. , For the unit compensation cost of T1 level, For the unit compensation cost of T2 level, The specific unit compensation cost for T3 level will be dynamically adjusted based on the latest compensation standards of the power trading center. The above design clarifies the criteria for classifying interruptible loads and constructs an exponential interruption compensation cost function with the promised response time as the independent variable. This accurately quantifies the relationship between the promised response time and the expected interruption compensation cost, making the cost model more consistent with the response characteristics of industrial loads and the compensation logic of the electricity market.

[0021] The specific process of constructing a translation cost function for the transferable load in the above technical solution includes: Based on the key process parameters, the adjustable loads are divided into movable loads. The movable loads refer to equipment with a single movable time ≥ D (D is 15 minutes) and a daily cumulative movable amount ≤ E% (E is 30) of the rated operating time of the equipment. The movable loads refer to equipment whose operating time can be adjusted within the process window (such as injection molding machines and air compressors). For the relocatable load, a relocation cost function is constructed with the relocation period as the independent variable. The expression of the relocation cost function is as follows: in, This represents the expected relocation compensation cost of the relocatable load (represented when the relocatable load is in...).h When the time period begins to shift, compensation needs to be paid to the units that can be shifted negatively. This represents the basic translation compensation cost (the basic translation compensation cost is the set benchmark price, which usually refers to the minimum compensation standard required to perform a translation during non-incentive periods (i.e., non-off-peak periods)). Indicates the incentive coefficient. h Indicates the start time of the translation. This represents the set of periods with low load. a The above design clarifies the criteria for classifying transferable loads and constructs a transfer cost function with the transfer period as the independent variable and including a term related to the load off-peak period. This function can guide transferable loads to migrate to the load off-peak period, thereby achieving grid valley filling and efficiency improvement while reducing dependence on high-cost interruptible resources.

[0022] In the above technical solution, the specific process of establishing a collaborative optimization model based on the interruption compensation cost function and the translation cost function is as follows: Based on the interruption compensation cost function and the translation cost function, a day-ahead scheduling layer model is constructed with the objective of minimizing the expected total cost. This model considers the expected interruption compensation cost and the expected translation compensation cost, combined with the set end-to-end response delay constraints. This yields the baseline plan for each adjustable load; Regarding the specific methods for constructing a day-ahead scheduling layer model to obtain the baseline plan for each adjustable load, some optimized technical solutions include: The day-ahead scheduling layer model is constructed by taking the cost of purchasing electricity from the external grid, the expected interruption compensation cost to be paid for calling interruptible loads (calculated by the interruption compensation cost function), the expected migration cost generated by calling shiftable loads (calculated by the migration cost function), and the valley filling incentive term as inputs and embedding physical constraints, namely end-to-end response delay constraints. An improved swarm intelligence optimization algorithm is used to solve the day-ahead scheduling layer model to obtain the individual with the best fitness. All decision variables encoded by the individual (i.e. when and how each adjustable load is scheduled) constitute the baseline plan formulated by the day-ahead scheduling layer model.

[0023] The end-to-end response delay constraint is expressed as follows: in, Communication transmission delay (communication transmission delay is collected in real time by a network latency monitoring module deployed on the communication link, with a sampling period of 1 minute and a typical value of ≤300 milliseconds). This is the accumulated value of the equipment execution delay estimated based on the safety control timing instructions in the preset control parameter package (generated by the virtual power plant according to industrial equipment communication protocols and process safety specifications, including equipment thermal state threshold parameters, process buffer verification logic, and safety control timing instructions). The threshold specified for power grid dispatch (the threshold can be set to 120 seconds for frequency regulation scenarios); Within a preset time resolution (which can be set to 15 minutes), adjustable load forecasts and real-time electricity price signals are received (real-time electricity price signals are obtained in real time through the standard API interface provided by the power trading center, and the data update frequency can be set to 15 minutes). Based on the interruption compensation cost function, the shift cost function, the adjustable load forecasts, and the real-time electricity price signals, an intraday rolling adjustment layer model is established with the goal of minimizing the actual adjustment cost, thereby obtaining the real-time adjustment instructions for each adjustable load. Regarding the specific method for establishing an intraday rolling adjustment layer model to obtain real-time adjustment commands for each adjustable load, some optimized technical solutions include: At the start of each preset time resolution, adjustable load forecasts and real-time electricity price signals are received. Based on these signals, the objective function is to minimize the actual adjustment cost (actual adjustment cost specifically refers to the additional cost or savings relative to the baseline plan incurred due to real-time adjustments). The decision variables of the intraday rolling adjustment layer model mainly involve rearranging the shifting periods of each shiftable load. When constructing the objective function, the intraday rolling adjustment layer model calls the pre-constructed shifting cost function for the shiftable load. The shifting cost function embeds a cost adjustment term (discount coefficient) associated with the off-peak load period, so that shifting the load to the off-peak period can directly generate economic benefits (negative costs). This transforms the demand for filling off-peak periods into a clear economic incentive signal. Meanwhile, the intraday rolling adjustment layer model follows a priority adjustment strategy for periods with both low electricity prices and low net loads. That is, through logical judgment (real-time electricity price is lower than 80% of the daily average price and net load is lower than 80% of the predicted average), it dynamically identifies periods with both low electricity prices and low net loads as the target window for adjustment. Under the premise of meeting the process constraints of each load (such as minimum continuous operating time and allowable shift time window), the intraday rolling adjustment layer model calculates the cost change caused by adjusting each shiftable load to different periods with both low electricity prices and low net loads, and prioritizes the load migration that can bring the greatest cost reduction, thereby completing the establishment of the intraday rolling adjustment layer model and obtaining real-time adjustment instructions for each adjustable load. Based on the baseline plans of each adjustable load obtained from the day-ahead scheduling layer model and the real-time adjustment instructions of each adjustable load obtained from the intraday rolling adjustment layer model, a day-ahead and intraday dual-layer collaborative optimization model containing end-to-end response delay constraints is jointly constructed. Regarding specific methods for constructing a two-tiered collaborative optimization model that includes end-to-end response delay constraints, both day-ahead and intraday, some optimized technical solutions include: The day-ahead scheduling layer model solves for a preliminary industrial load scheduling scheme, namely the baseline plan, which serves as the starting point and constraint boundary for intraday rolling model adjustments. When the baseline plan is executed, the intraday rolling adjustment layer model aims to minimize the deviation cost (i.e., actual adjustment cost) from the baseline plan and uses the baseline plan output by the day-ahead scheduling layer model as the initial value of the decision variables of the intraday rolling adjustment layer model. During the operation of the intraday rolling adjustment layer model, the end-to-end response delay constraint is treated as a hard rule that must be followed to ensure that no real-time adjustment command will exceed the fastest response time that the adjustable load can physically achieve. The day-ahead scheduling layer model and the rolling adjustment layer model work together to construct a day-ahead and intraday two-layer collaborative optimization model that includes end-to-end response delay constraints. The final executable industrial load scheduling scheme is the result of the combined action of the day-ahead scheduling layer model and the rolling adjustment model.

[0024] The above design establishes a two-layer collaborative optimization model for day-ahead and intraday operations and embeds end-to-end response delay constraints. The day-ahead scheduling layer model formulates a long-term benchmark plan to ensure scheduling economy, while the intraday rolling adjustment layer model dynamically adjusts to adapt to real-time operational uncertainties. The hard constraint on end-to-end response delay is used to ensure that the industrial load scheduling scheme always meets the hard requirements of power grid timeliness, which can improve the robustness and practicality of the industrial load scheduling scheme.

[0025] In the above technical solution, the specific process of obtaining the industrial load scheduling scheme by solving the collaborative optimization model based on the scenario set generated by load and electricity price forecasts and using an improved swarm intelligence optimization algorithm is as follows: The virtual power plant uses a long short-term memory network model to generate load demand and electricity price forecast curves for future cycles using historical adjustable load data, meteorological elements, and date types. Based on these curves, an initial scenario set (containing a large number of possible scenarios) is generated using the Latin hypercube sampling method. The initial scenario set is then reduced to obtain a representative scenario set (this reduction can be achieved using clustering algorithms based on Kantorovich distance or K-means clustering). The representative scenario set prioritizes scenarios that cover the risk of excessive response latency (including scenarios corresponding to edge controller triggering anomalies, equipment security verification timeouts, and communication link interruptions). Regarding the specific method of generating an initial scene set using the Latin hypercube sampling method and then performing scene reduction processing on the initial scene set to obtain a representative scene set, some optimized technical solutions include: The initial scenario set is generated through Latin hypercube sampling and can contain 200 scenarios, covering adjustable load forecasting errors and price fluctuations in real-time electricity price signals. Kantorovich distance is used as the metric to reduce the number of scenarios in the initial set: the initial scenario set is used as the remaining scenario set, and a preset retention quantity K is set (K is 10 and can be dynamically adjusted according to computing resources and scheduling accuracy requirements). When the number of remaining scenarios exceeds K, the average distance between each scenario in the remaining scenario set and all other scenarios is calculated, and the scenario with the smallest average distance is selected as a candidate for deletion. In the initial scenario generation phase, high-risk scenarios are generated by injecting three types of abnormal events and adding risk labels, including edge controller triggering anomalies: the edge controller response latency increases by 50% after the simulated command is issued; device security verification timeout: the time taken to verify the simulated device thermal state parameters exceeds a preset threshold by 20%; and communication link interruption: the packet loss rate of the simulated scheduling command transmission is set to 15%. If a candidate deletion object has a risk label, the deletion operation is skipped, and a non-high-risk scenario with the smallest average distance is selected from the remaining scenario set for deletion. The above iterative process is repeated until the number of remaining scenarios equals K. The final remaining scenario set is the representative scenario set used for solving the collaborative optimization model. The day-ahead and intraday two-layer collaborative optimization model, which includes end-to-end response delay constraints, is combined with a representative set of scenarios and solved using an improved swarm intelligence optimization algorithm: Initialize a population, where individuals in the population represent scheduling decision variables using a hybrid encoding method. The first individual in the population... The bits are integers that represent the response time level of the interruptible load, where Number of interruptible loads; population individuals after The bit is a real number encoding, representing the transfer period of the transferable load, where The number of loads that can be moved; Based on the ratio of the current iteration number to the maximum iteration number, the inertia weight is dynamically adjusted using a quadratic decay formula, which is: in, The inertia weight at the current iteration time. This is the initial inertia weight (the initial inertia weight is the same as the maximum inertia weight, and can be 0.9). The termination inertia weight (the termination inertia weight is the minimum inertia weight, which can be 0.4) is used. This represents the current iteration number. This represents the maximum number of iterations (the maximum number of iterations can be 200, representing the set total number of iterations or termination condition). After each iteration update, the normalized diversity of the population is calculated. If the diversity is lower than the preset threshold (the preset threshold is 0.15), then simulated binary crossover (crossover probability of 0.9) and polynomial mutation operation (mutation probability of 0.9) are performed on a randomly selected subset of individuals (30% of all individuals). Calculate the fitness of an individual (each individual represents a candidate industrial load scheduling scheme). If the individual violates the end-to-end response delay constraint, add a penalty term. The formula for the penalty term is: in, The penalty coefficient is... For end-to-end response delay constraints, Thresholds specified for power grid dispatching V For non-compliant adjustable load sets, For adjustable load k power, Total regulating power; Repeat the above iterative process until the termination condition is met (the termination condition is reaching the maximum number of iterations (200 times)), and obtain the industrial load scheduling scheme corresponding to the individual with the best fitness. The above design uses a long short-term memory network for load and electricity price prediction, Latin hypercube sampling to generate an initial scene set and performs reduction processing with priority retention of risk scenes. Combined with the improved swarm intelligence optimization algorithm's hybrid encoding, secondary decay inertial weight, population diversity detection and power weighted penalty term solution optimization, it improves the solution efficiency of the individual with the best fitness and the global optimization ability, and can effectively avoid the risk of constraint violation, thereby improving the robustness and executability of the output industrial load scheduling scheme.

[0026] The specific process for obtaining the response performance indicators of each adjustable load in the above technical solution is as follows: The industrial load scheduling scheme is sent to the edge controller on the user side. The edge controller performs local security verification on the instructions in the industrial load scheduling scheme according to the preset control parameter package. After the verification is successful, it drives each adjustable load to perform the corresponding operation and collects the actual response delay of each adjustable load (actual response delay refers to the time from the issuance of the instruction in the industrial load scheduling scheme to the completion of the action) and the actual compensation cost (actual compensation cost refers to the cost calculated based on the actual response of the adjustable load). Regarding the specific method for edge controllers to perform local security verification of instructions in industrial load scheduling schemes based on preset control parameter packages, some optimized technical solutions include: After receiving the instructions from the industrial load scheduling scheme, the edge controller decrypts the preset control parameter package and performs triple verification: the first verification is of the real-time thermal status parameters of the equipment (such as whether the cooling water temperature is ≤ the threshold of 60℃±2℃ in the preset control parameter package), the second verification is of the process buffer conditions (such as whether the buffer material inventory is ≥ the safety threshold of 5 tons), and the third verification is of the completeness of the steps of the safety control timing instructions; after all verifications pass, the equipment control loop is triggered according to the preset timing.

[0027] For each scheduled adjustable load, based on the actual response delay and actual compensation cost of the corresponding adjustable load, as well as the promised response time and promised compensation cost of the corresponding adjustable load in the industrial load scheduling scheme, the response performance index of the corresponding adjustable load is calculated: in, As an indicator of the response performance of adjustable loads, The promised response time for adjustable loads (in seconds). The actual response delay for adjustable loads (in seconds). Cost of compensation for adjustable load (unit: yuan). The actual compensation cost of the adjustable load (unit: yuan) is calculated by comparing the promised response time and promised compensation cost of the adjustable load with the actual response delay and actual compensation cost. The normalization process eliminates the influence of dimensions and ensures that the index value range is within a reasonable range. It can quantify the deviation between response time and cost execution and objectively reflect the actual scheduling contribution and execution capability of the adjustable load.

[0028] Example 2 An industrial load resource management method based on cost optimization and response time, such as Figure 2 As shown, electrical operation data is decomposed to extract key process parameters for each adjustable load; all adjustable loads are divided into interruptible loads and shiftable loads, and corresponding interruption compensation cost functions and shift cost functions are constructed; a collaborative optimization model is established based on the interruption compensation cost function and shift cost function, and an improved swarm intelligence optimization algorithm is used to solve for the industrial load scheduling scheme; the industrial load scheduling scheme is executed, and the response performance index of each adjustable load is calculated based on the actual response delay and actual compensation cost of each adjustable load, as well as the corresponding promised response time and promised compensation cost.

[0029] The specific methods for industrial load resource management based on cost optimization and response timeliness include the following steps: The electrical operation data of industrial loads are decomposed using a non-intrusive load identification model to obtain independent operating curves for each adjustable load. Key process parameters corresponding to each adjustable load are then extracted from these independent operating curves. Based on the key process parameters, all adjustable loads are divided into interruptible loads and shiftable loads. An interruption compensation cost function is constructed for the interruptible loads, and a shift cost function is constructed for the shiftable loads. Based on the interruption compensation cost function and the shift cost function, a collaborative optimization model is established; based on the scenario set generated by load and electricity price prediction, the collaborative optimization model is solved using an improved swarm intelligence optimization algorithm to obtain an industrial load scheduling scheme. By implementing the industrial load scheduling scheme, the response performance index of each adjustable load is calculated based on the actual response delay and actual compensation cost of each adjustable load, as well as the promised response time and promised compensation cost of the corresponding adjustable load in the industrial load scheduling scheme.

[0030] Example 3 A computer program product includes a computer program that, when executed by a processor, implements the steps of the method described in Embodiment 2.

[0031] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0032] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0033] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0034] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0035] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit its scope of protection. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading the present invention, they can still make various changes, modifications or equivalent substitutions to the specific implementation of the invention, but these changes, modifications or equivalent substitutions are all within the scope of protection of the pending claims of the invention.

[0036] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

Claims

1. An industrial load resource management system based on cost optimization and response time, characterized in that, It includes: The load identification and key parameter extraction module is used to decompose the electrical operation data of industrial loads using a non-intrusive load identification model, obtain the independent operation curves of each adjustable load, and extract the key process parameters corresponding to each adjustable load from the independent operation curves. The load resource dynamic modeling module is used to divide all adjustable loads into interruptible loads and shiftable loads based on the key process parameters, and to construct an interruption compensation cost function for interruptible loads and a shift cost function for shiftable loads. The collaborative optimization and decision-making module is used to establish a collaborative optimization model based on the interruption compensation cost function and the translation cost function; Based on the scenario set generated by load and electricity price forecasts, an improved swarm intelligence optimization algorithm is used to solve the collaborative optimization model to obtain an industrial load scheduling scheme. The scheduling execution and performance evaluation module is used to execute the industrial load scheduling scheme. Based on the actual response delay and actual compensation cost of each adjustable load, as well as the promised response time and promised compensation cost of the corresponding adjustable load in the industrial load scheduling scheme, the module calculates the response performance index of each adjustable load.

2. The industrial load resource management system based on cost optimization and response time according to claim 1, characterized in that, It also includes: The feedback optimization module is used to dynamically adjust the response time level and compensation cost parameters of the interruptible load according to the response performance index, and update the interruption compensation cost function and the collaborative optimization model according to the adjusted response time level and compensation cost parameters.

3. The industrial load resource management system based on cost optimization and response time according to claim 1, characterized in that: The specific process for obtaining the key process parameters corresponding to each adjustable load includes: A non-intrusive load identification model is constructed by fusing convolutional neural networks and long short-term memory networks. The electrical operation data of industrial loads are decomposed through the non-intrusive load identification model to obtain the independent operation curves of each adjustable load. By fitting the transient process segment representing load start-up or shutdown in the independent operation curve with an exponential decay model, the time constant obtained from the fitting is used as the equipment start-up and shutdown inertia time. The buffer capacity is obtained by calculating the product of the power fluctuation standard deviation of the corresponding production process buffer stage in the independent operation curve and the duration of the production process buffer stage. Based on the equipment start-up and shutdown inertia time, the buffer capacity, and the preset safe shutdown threshold, the key process parameters corresponding to each adjustable load are obtained.

4. The industrial load resource management system based on cost optimization and response time according to claim 1, characterized in that: The specific process of constructing an interruption compensation cost function for interruptible loads includes: Based on the key process parameters, interruptible loads are divided from the adjustable loads. The interruptible loads refer to equipment with a minimum safe interruption duration ≤ A and a process allowable number of interruptions ≥ B. For interruptible loads, an interruption compensation cost function is constructed with the promised response time as the independent variable. The expression for the interruption compensation cost function is as follows: in, This indicates the promised response time for interruptible loads. Indicates the basic compensation cost. A Indicates the range of cost adjustments. k Indicates the promised response time The attenuation coefficient, e It is a natural constant. This indicates the expected cost of compensation for the disruption.

5. The industrial load resource management system based on cost optimization and response time according to claim 1, characterized in that: The specific process of constructing a translation cost function for a transferable load includes: Based on the key process parameters, the adjustable load is divided into movable loads. The movable loads refer to equipment with a single movable time ≥ D and a daily cumulative movable amount ≤ E% of the equipment's rated operating time. For the relocatable load, a relocation cost function is constructed with the relocation period as the independent variable. The expression of the relocation cost function is as follows: in, This represents the expected relocation compensation cost of the relocatable load. Indicates the cost of basic translation compensation. Indicates the incentive coefficient. h Indicates the start time of the translation. This represents the set of periods with low load. a This indicates the discount rate.

6. The industrial load resource management system based on cost optimization and response time according to claim 4 or 5, characterized in that: The specific process for establishing a collaborative optimization model based on the interruption compensation cost function and the translation cost function is as follows: Based on the interruption compensation cost function and the translation cost function, a day-ahead scheduling layer model is constructed with the objective of minimizing the expected total cost. This model considers the expected interruption compensation cost and the expected translation compensation cost, combined with the set end-to-end response delay constraints. This yields the baseline plan for each adjustable load; The end-to-end response delay constraint is expressed as follows: in, For communication transmission delay, The estimated device execution delay accumulation value is based on the safety control timing instructions in the preset control parameter package. Thresholds specified for power grid dispatching; Within a preset time resolution, adjustable load forecasts and real-time electricity price signals are received. Based on the interruption compensation cost function, the shift cost function, adjustable load forecasts, and real-time electricity price signals, an intraday rolling adjustment layer model is established with the goal of minimizing actual adjustment costs, thereby obtaining real-time adjustment instructions for each adjustable load. Based on the baseline plans for each adjustable load obtained from the day-ahead scheduling layer model and the real-time adjustment instructions for each adjustable load obtained from the intraday rolling adjustment layer model, a day-ahead and intraday dual-layer collaborative optimization model containing end-to-end response delay constraints is jointly constructed.

7. The industrial load resource management system based on cost optimization and response time according to claim 6, characterized in that: Based on the scenario set generated by load and electricity price forecasts, the collaborative optimization model is solved using an improved swarm intelligence optimization algorithm. The specific process for obtaining the industrial load scheduling scheme is as follows: The virtual power plant uses a long short-term memory network model to generate load demand and electricity price forecast curves for future cycles using historical adjustable load data, meteorological elements, and date types. Based on the load demand and electricity price forecast curves, an initial scenario set is generated using the Latin hypercube sampling method. The initial scenario set is then processed by scenario reduction to obtain a representative scenario set. The day-ahead and intraday two-layer collaborative optimization model, which includes end-to-end response delay constraints, is combined with a representative set of scenarios and solved using an improved swarm intelligence optimization algorithm: Initialize a population, where individuals in the population represent scheduling decision variables using a hybrid encoding method. The first individual in the population... The bits are integers that represent the response time level of the interruptible load, where Number of interruptible loads; population individuals after The bit is a real number encoding, representing the transfer period of the transferable load, where The number of loads that can be moved; Based on the ratio of the current iteration number to the maximum iteration number, the inertia weight is dynamically adjusted using a quadratic decay formula, which is: in, The inertia weight at the current iteration time. As the initial inertia weight, To terminate the inertia weight, This represents the current iteration number. This represents the maximum number of iterations. After each iteration update, the normalized diversity of the population is calculated. If the diversity is lower than a preset threshold, simulated binary crossover and polynomial mutation operations are performed on a randomly selected subset of individuals. Calculate individual fitness; if an individual violates the end-to-end response delay constraint, add a penalty term. The formula for the penalty term is: in, The penalty coefficient is... For end-to-end response delay constraints, Thresholds specified for power grid dispatching V For non-compliant adjustable load sets, For adjustable load k power, Total regulating power; Repeat the above iterative process until the termination condition is met to obtain the industrial load scheduling scheme corresponding to the individual with the best fitness.

8. The industrial load resource management system based on cost optimization and response time according to claim 1, characterized in that: The specific process for obtaining the response performance indicators of each adjustable load is as follows: The industrial load scheduling scheme is sent to the edge controller on the user side. The edge controller performs local security verification on the instructions in the industrial load scheduling scheme according to the preset control parameter package. After the verification is passed, it drives each adjustable load to perform the corresponding operation and collects the actual response delay and actual compensation cost of each adjustable load. For each scheduled adjustable load, based on the actual response delay and actual compensation cost of the corresponding adjustable load, as well as the promised response time and promised compensation cost of the corresponding adjustable load in the industrial load scheduling scheme, the response performance index of the corresponding adjustable load is calculated: in, As an indicator of the response performance of adjustable loads, The promised response time for adjustable loads, The actual response delay for adjustable loads, To compensate for the cost of adjustable load commitments, This represents the actual compensation cost for adjustable loads.

9. An industrial load resource management method based on cost optimization and response time, characterized in that, It includes the following steps: The electrical operation data of industrial loads are decomposed using a non-intrusive load identification model to obtain independent operating curves for each adjustable load. Key process parameters corresponding to each adjustable load are then extracted from these independent operating curves. Based on the key process parameters, all adjustable loads are divided into interruptible loads and shiftable loads. An interruption compensation cost function is constructed for the interruptible loads, and a shift cost function is constructed for the shiftable loads. A collaborative optimization model is established based on the interruption compensation cost function and the translation cost function; Based on the scenario set generated by load and electricity price forecasts, an improved swarm intelligence optimization algorithm is used to solve the collaborative optimization model to obtain an industrial load scheduling scheme. By implementing the industrial load scheduling scheme, the response performance index of each adjustable load is calculated based on the actual response delay and actual compensation cost of each adjustable load, as well as the promised response time and promised compensation cost of the corresponding adjustable load in the industrial load scheduling scheme.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method of claim 9.

Images