Energy management method and system based on ai algorithm

Abstract

The present application relates to the technical field of energy management, in particular to an energy management method and system based on AI algorithm, comprising the following steps: collecting terminal load data to construct a periodic fluctuation curve, modeling, extracting change characteristics, generating an adjustment table, identifying adjustment capacity in combination with a disturbance threshold, generating a label set, obtaining current load training to generate a priority queue, executing adjustment to extract response data to generate a log, comparing the log to determine whether to rearrange and update the adjustment table and the label set to output results, in the present application, the slope and fluctuation characteristics are extracted based on periodic load data, the terminal adjustment capacity is identified and the capacity label is generated, the adjustment priority queue is optimized in combination with real-time load input, the response time and change data are recorded to form a log after the adjustment is executed, if the task deviates, the queue is rearranged and the label and the adjustment table are updated, and the multi-dimensional improvement of energy scheduling accuracy, system response sensitivity and terminal resource utilization efficiency is carried out.

Description

Technical Field

[0001] This invention relates to the field of energy management technology, and in particular to energy management methods and systems based on AI algorithms. Background Technology

[0002] The field of energy management technology encompasses promoting energy conservation and optimized use through the efficient scheduling, allocation, and utilization of energy resources. Its core content revolves around the effective monitoring and management of multiple stages of energy production, transmission, storage, and consumption. Energy management technology not only focuses on real-time monitoring of energy consumption but also includes scheduling and forecasting based on data analysis and algorithm optimization. With increasing energy demand and environmental protection requirements, improving energy efficiency and reducing carbon emissions through intelligent means has become a crucial development direction in this field.

[0003] Among them, the AI-based energy management method and system refers to a method for intelligent scheduling and management of energy using artificial intelligence technology. Addressing the scheduling optimization problem in energy management, a scheme using AI algorithms for energy consumption prediction and intelligent scheduling is proposed. Specifically, the patent optimizes energy use strategies, reduces energy waste, and improves the system's flexibility and responsiveness by establishing a data-driven model that combines historical and real-time data.

[0004] While existing technologies in energy management possess real-time monitoring and forecasting capabilities, they fall short in terms of fine-grained analysis of load changes and the immediacy of disturbance response. Their scheduling decisions largely rely on predictive models based on historical trends and overall statistics, which are not sensitive enough to capturing instantaneous disturbances. This leads to delays in responding to sudden load fluctuations, potentially causing short-term supply-demand imbalances and energy waste. Furthermore, existing technologies lack dynamic priority adjustment mechanisms in multi-terminal environments, making it difficult to optimize scheduling sequences in a timely manner based on actual adjustment effects during execution. This can result in over- or under-adjustment at certain nodes. In the long term, these deficiencies lead to decreased energy utilization, increased operating costs, and exacerbated risks during peak grid pressure periods, such as localized overloads or power shortages during extreme weather or peak electricity demand, impacting system stability and service continuity. Summary of the Invention

[0005] To address the technical problems existing in the prior art, embodiments of the present invention provide an energy management method based on AI algorithms, comprising the following steps:

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an energy management method based on AI algorithms, comprising the following steps:

[0007] S1: Collect load value change data of power terminals and construct load cycle fluctuation curves, extract load change slope and fluctuation amplitude and normalize them, analyze the load response trend of power terminals through Bayesian model, and generate load adjustment table;

[0008] S2: Compare the load regulation table with the disturbance discrimination threshold to determine if there is load disturbance data in the load regulation table, identify the power load status based on the load disturbance, calculate the load regulation capacity of the power terminal and attach a tag to generate a regulation capacity tag set;

[0009] S3: Based on the load adjustment table and the adjustment capacity label set as input, the optimization objective between load status and load adjustment capacity is set through the near-end strategy optimization algorithm to generate a load adjustment task sequence;

[0010] S4: Perform load regulation of power terminals through the load regulation task sequence, extract multiple parameters such as the slope of power load change of power terminals after regulation, regulation response time and regional total load change, and construct regulation log dataset;

[0011] S5: Structure the task duration and sorting position in the adjustment log dataset and compare them with the load adjustment task sequence. If the task offset threshold is exceeded, rearrange the task sequence to generate power energy management results.

[0012] As a further embodiment of the present invention, the load adjustment table includes load change slope, fluctuation amplitude normalization value, and load response trend probability; the adjustment capacity label set includes load disturbance identification mark, power adjustment capacity, and power terminal label type; the load adjustment task sequence specifically includes task number and target adjustment value; the adjustment log dataset includes adjusted load change slope, adjustment response time interval, and regional total load curve change; and the power energy management result includes task structure mapping sequence, task offset, and rearranged task sequence table.

[0013] As a further aspect of the present invention, the specific steps of S1 are as follows:

[0014] S101: Collect load value change data of power terminals in continuous time periods, arrange them point by point and construct load cycle fluctuation curve, calculate the curve change rate and extract the load change slope, compare the range of load values ​​in the same cycle and extract the fluctuation amplitude, and generate cycle load characteristic quantity.

[0015] S102: Based on the periodic load characteristic quantity, the load change slope and fluctuation amplitude are normalized, multiple values ​​are constructed into proportional coefficients according to the period and transformed into a dimensionless sequence, and the processed results are recombined in time order to obtain the normalized feature set.

[0016] S103: Call the normalized feature set, set the prior probability distribution in the Bayesian model and construct the likelihood function in combination with the normalized feature set, calculate the response probability of the load state under multiple adjustment intervals, and pair the time series with the corresponding response probabilities in order to establish a load adjustment table.

[0017] As a further aspect of the present invention, the specific steps of S2 are as follows:

[0018] S201: Based on the comparison between the multi-period load response probability and the disturbance discrimination threshold in the load adjustment table, obtain the judgment result of whether there is a load disturbance in each period, mark the time point where a disturbance is judged as the disturbance period, and mark the other time points as the non-disturbance period, and generate a disturbance identification mark sequence.

[0019] S202: Call the disturbance identification marker sequence, count the frequency and average response probability of multiple power terminals during the disturbance period within the cycle, and calculate the disturbance response probability density in combination with the corresponding cycle length. At the same time, classify and number the disturbance response probability density according to the adjustment capability to obtain the adjustment capability value set.

[0020] S203: Based on the set of regulation capacity values, attach corresponding capacity tag text to the regulation level number of each power terminal, combine the number and tag into structured data, aggregate according to the terminal number and time sequence, and generate a set of regulation capacity tags.

[0021] As a further aspect of the present invention, the disturbance discrimination threshold is set by adjusting the load change trend and response time fluctuation range of multiple power terminals in the log dataset during task execution, statistically analyzing the boundary range of the load slope change rate under disturbance conditions, and combining it with the average response time value under normal tasks.

[0022] As a further aspect of the present invention, the specific steps of S3 are as follows:

[0023] S301: Based on the load regulation table and the regulation capacity tag set, obtain the load value and corresponding tag of the power terminal in each time period as input data, and establish a load regulation task dataset for each time period by combining the load status and regulation capacity.

[0024] S302: Based on the load regulation task dataset, the near-end strategy optimization algorithm is used for training. An optimization target between load status and regulation capacity is set, and the strategy parameters are iteratively adjusted. Through training, the optimal match between load status and regulation capacity in multiple time periods is obtained, and an optimized matching strategy set is obtained.

[0025] S303: Based on the optimized matching strategy set, a load regulation task sequence is formed according to the load value and regulation capacity of each power terminal and sorted according to the time series.

[0026] As a further aspect of the present invention, the specific steps of S4 are as follows:

[0027] S401: Based on the load regulation task sequence, perform load regulation of the power terminal, monitor the power change value before and after regulation, compare the load change amount and time interval of adjacent time periods, calculate the load change rate, and generate the load change slope;

[0028] S402: Call the load change slope, extract time interval data based on the start time of load value change of multiple power terminals and the response time point of reaching the adjustment target, perform interval calculation on the response process of multiple terminals when performing adjustment tasks, and obtain the adjustment response time interval;

[0029] S403: Based on the regulation response time interval and the total load value of multiple regional power terminals within the corresponding time period, calculate the total load difference between two adjacent regulation task execution periods, collect the differences between multiple time periods and establish a unified structure dataset, and combine it with existing regulation information to form a regulation log dataset.

[0030] As a further aspect of the present invention, the specific steps of S5 are as follows:

[0031] S501: Based on the aforementioned adjustment log dataset, extract the adjustment start time, response period, and load change slope of the multi-power terminal load adjustment tasks, establish a mapping relationship structure according to the task number and time order, and generate a task structured mapping value set.

[0032] S502: Call the task structured mapping value set, compare the task number and sorting position of the corresponding task in the load adjustment task sequence at the same time point, calculate the task offset duration, filter the set of tasks whose offset degree exceeds the adjustment task offset threshold, and obtain the adjustment task offset amount.

[0033] S503: Based on the adjustment task offset, rearrange the position order of all offset tasks in the original load adjustment task sequence, regenerate the terminal load response task list under the corresponding time axis, and collect the load change values ​​during the adjustment period to establish the power energy management results.

[0034] As a further aspect of the present invention, the task offset threshold is set by adjusting the time distribution range and numbering continuity of multiple tasks in the current response process in the log dataset, statistically analyzing the allowable sorting error and response time delay of multiple task types in the standard execution sequence, and combining the task type execution characteristics.

[0035] AI-based energy management systems include:

[0036] The load modeling module collects load value change data from power terminals and constructs load cycle fluctuation curves. It extracts the load change slope and fluctuation amplitude and normalizes them. It analyzes the load response trend of power terminals through a Bayesian model, generates a load adjustment table, and transmits it to the adjustment identification module.

[0037] The regulation identification module compares the load regulation table with the disturbance discrimination threshold to determine if there is load disturbance data in the table. Based on the load disturbance, it identifies the power load status, calculates the load regulation capacity of the power terminal and adds a tag, generates a regulation capacity tag set and transmits it to the regulation task module.

[0038] The adjustment task module, based on the load adjustment table and the adjustment capacity label set as input, sets the optimization target between load status and load adjustment capacity through the near-end strategy optimization algorithm, generates a load adjustment task sequence and passes it to the adjustment execution module;

[0039] The adjustment execution module performs load adjustment of power terminals through the load adjustment task sequence, extracts multiple parameters such as the slope of the power load change of the power terminals after adjustment, the adjustment response time and the change of the total regional load, constructs an adjustment log dataset and transmits it to the result evaluation module;

[0040] The results evaluation module structures the task duration and sorting position in the adjustment log dataset and compares them with the load adjustment task sequence for consistency. If the task offset threshold is exceeded, the task sequence is rearranged to generate power energy management results.

[0041] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0042] This invention utilizes periodic analysis and trend modeling of load change data, combined with fluctuation amplitude and slope extraction, to form a precise basis for load regulation. This allows for rapid identification and quantification of the regulation capabilities of multiple terminals when disturbances occur, generating a set of available resources with capability tags. Based on this, real-time load information is used as dynamic input, and a priority queue is formed through strategy optimization training, making the power regulation process more targeted and flexible. After regulation execution, a regulation log is established by extracting response time and total load changes, and consistency verification is performed with the established priority queue to ensure the accuracy and stability of the regulation process. When deviations from predetermined task thresholds occur, tasks can be reordered and the data base updated immediately, enabling energy allocation and scheduling to form an adaptive closed loop. This series of continuous data acquisition, feature extraction, dynamic evaluation, priority ranking, and closed-loop feedback improves the accuracy and real-time performance of energy scheduling, reduces energy waste caused by delayed or unbalanced regulation, enhances the system's sensitivity and response speed to load changes, and simultaneously maximizes resource utilization and rapidly restores regional supply and demand balance in a multi-terminal environment. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a schematic diagram of the steps of the present invention;

[0045] Figure 2 This is a detailed schematic diagram of S1 of the present invention;

[0046] Figure 3 This is a detailed schematic diagram of S2 of the present invention;

[0047] Figure 4 This is a detailed schematic diagram of S3 of the present invention;

[0048] Figure 5 This is a detailed schematic diagram of S4 of the present invention;

[0049] Figure 6 This is a detailed schematic diagram of S5 of the present invention;

[0050] Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0051] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0052] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0053] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0054] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0055] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0056] Please see Figure 1 This invention provides an energy management method based on AI algorithms, comprising the following steps:

[0057] S1: Collect load value change data of power terminals and construct load cycle fluctuation curves, extract load change slope and fluctuation amplitude and normalize them, analyze the load response trend of power terminals through Bayesian model, and generate load adjustment table;

[0058] S2: Compare the load regulation table with the disturbance discrimination threshold to determine if there is load disturbance data in the load regulation table. Identify the power load status based on the load disturbance, calculate the load regulation capacity of the power terminal and attach a label to generate a regulation capacity label set.

[0059] S3: Based on the load adjustment table and the adjustment capacity label set as input, the optimization objective between load status and load adjustment capacity is set through the near-end strategy optimization algorithm to generate a load adjustment task sequence;

[0060] S4: Perform load regulation of power terminals through load regulation task sequence, extract multiple parameters such as the slope of power load change of power terminals after regulation, regulation response time and regional total load change, and construct regulation log dataset;

[0061] S5: Structure the adjustment log dataset to determine the task duration and sorting position, and compare the consistency with the load adjustment task sequence. If the task offset threshold is exceeded, rearrange the task sequence to generate power energy management results.

[0062] The load adjustment table includes load change slope, fluctuation amplitude normalization value, and load response trend probability. The adjustment capacity label set includes load disturbance identification mark, power adjustment capacity, and power terminal label type. The load adjustment task sequence specifically includes task number and target adjustment value. The adjustment log dataset includes the adjusted load change slope, adjustment response time interval, and regional total load curve change. The power energy management results include task structure mapping sequence, task offset, and rearranged task sequence table.

[0063] Please see Figure 2 The specific steps of S1 are as follows:

[0064] S101: Collect load value change data of power terminals in continuous time periods, arrange them point by point and construct load cycle fluctuation curve, calculate the curve change rate and extract the load change slope, compare the range of load values ​​in the same cycle and extract the fluctuation amplitude, and generate cycle load characteristic quantity.

[0065] Load value change data of power terminals are collected over a continuous period of time. Load values ​​are collected at 1-minute intervals within a set time period using power monitoring devices. This includes collecting four data points from the residential power terminal between 18:00 and 18:03: time point... load kW, load kW, load kW, load kW, Representing time points, the data is arranged in chronological order as an array structure. Construct a load cycle fluctuation curve, connecting multiple points with time (min) on the horizontal axis and load (kW) on the vertical axis to form a broken line, calculate the rate of change of the curve, and perform subtraction on adjacent points to calculate the load difference. and time difference Due to fixed intervals min, calculate the rate of change of the curve Points 0 to 1: kW / min, points 1 to 2: kW / min, point 2 to 3: kW / min, to obtain the rate of change sequence Extract the slope of load change; the slope is directly taken as the rate of change value. kW / min, compare the range of load values ​​within the same cycle, and find the maximum load value in the array. kW and minimum load value kW, Represents the load value, and performs a subtraction operation on the range. kW, extract the fluctuation amplitude, the amplitude equals the range value. kW, generating periodic load characteristics, and combining slope sequences and amplitude value .

[0066] Table 1: Power Terminal Load Acquisition Data Table

[0067]

[0068] As shown in Table 1, the load data points collected in the embodiment are displayed. Based on these, the slope sequence and fluctuation amplitude are calculated, and characteristic quantities are generated.

[0069] S102: Based on the periodic load characteristics, the load change slope and fluctuation amplitude are normalized, multiple values ​​are constructed into proportional coefficients according to the period and transformed into a dimensionless sequence, and then the processed results are recombined in time order to obtain the normalized feature set.

[0070] Based on periodic load characteristics, the slope sequence is called. kW / min and fluctuation range kW, normalize the slope of load change, and find the slope with the largest absolute value in the sequence. kW / min, normalize the slope by performing division: , , , to obtain the sequence To normalize the fluctuation range, refer to the maximum amplitude. kW (obtained by statistically analyzing peak data from the same terminal), calculated Construct a scaling factor from multiple numerical values, and perform a division operation: Scaling factor = Normalized slope / Normalized amplitude. , , , to obtain the sequence The result is transformed into a dimensionless sequence (normalization has already achieved dimensionlessness), and the results are combined in chronological order into a standardized feature set, with each time point representing a specific time point. correspond , correspond At the same time, according to the time point combination , forming a set This yields the normalized feature set.

[0071] S103: Call the normalized feature set, set the prior probability distribution in the Bayesian model and construct the likelihood function in combination with the normalized feature set, calculate the response probability of the load state under multiple adjustment intervals, and pair the time series with the corresponding response probabilities in order to establish a load adjustment table;

[0072] Calling the normalized feature set, referencing the feature set, such as time points. proportionality coefficient , proportionality coefficient , proportionality coefficient In the Bayesian model, a prior probability distribution is set, and the frequency of load states is statistically analyzed based on the data: the probability of high load occurrence. (Accounting for 30% of the data), medium load (50%), low load (20% of the total), construct a likelihood function based on the feature set, set the proportional coefficient to follow a normal distribution, and the mean value under high load conditions. ,variance (Data fitting), medium load , low load , Regarding the time point proportionality coefficient Calculate the likelihood value: high load probability density Calculation of the index part , denominator ,result Similarly, the load likelihood is calculated in the same way. Low load probability density , Represents an exponential function, the exponential part , denominator ,result Calculate the load state response probability (posterior probability), perform multiplication, and then normalize: High load ratio medium load low load ,sum Normalized high load probability medium load low load Calculate the same (proportion coefficient) )and (proportion coefficient) The posterior probability of the time series is used to pair the response probability with the time series: Corresponding probability distribution , correspond , correspond Establish a load adjustment table with a structure consisting of three columns: time point, high / medium / low load probability.

[0073] Please see Figure 3 The specific steps of S2 are as follows:

[0074] S201: Based on the comparison between the load response probability of multiple time periods in the load adjustment table and the disturbance discrimination threshold, obtain the judgment result of whether there is a load disturbance in each time period, mark the time point that is judged to have a disturbance as the disturbance period, and mark the other time points as the non-disturbance period, and generate a disturbance identification label sequence.

[0075] Based on load adjustment tables (e.g., time points) Corresponding to high load probability ,middle ,Low , high ,middle ,Low , high ,middle ,Low The table structure includes a time index and probability values. A disturbance detection threshold is set, obtained through data analysis. The disturbance event database is accessed to statistically determine the critical point where a high-load probability triggers a disturbance. Minimum and maximum value searches are performed. The minimum high-load probability at the time of the disturbance in historical data is... The maximum is Take the average value As a threshold, the example setting value The load response probability is compared for each time period. For each time point, a comparison operation is performed between the high load probability value and a threshold. If the high load probability is greater than the threshold, a disturbance is determined to exist; otherwise, it is not. For example, at time point... High load probability After performing the comparison, the output is "no disturbance" at the specified time point. High load probability Output "Disturbance exists" at time point High load probability The output is "No disturbance". The result sequence is obtained as [No disturbance, Disturbance exists, No disturbance]. Disturbance periods are marked, and the time points where the result is "Disturbance exists" are assigned the label "Disturbance". For example... Mark "perturbation" and the rest as "non-perturbation", for example and Mark "unperturbed", generate a perturbation identification label sequence, and output the array [(t=0, unperturbed), (t=1, perturbed), (t=2, unperturbed)] in chronological order.

[0076] S202: Call the disturbance identification marker sequence, count the frequency and average response probability of multiple power terminals during the disturbance period within the cycle, and calculate the disturbance response probability density in combination with the corresponding cycle length. At the same time, classify and number the disturbance response probability density according to the regulation capability to obtain the regulation capability value set.

[0077] The generated disturbance identification marker sequence is invoked, such as the terminal A sequence [(0, no disturbance), (1, disturbance), (2, no disturbance)], the terminal B sequence [(0, disturbance), (1, no disturbance)], and the terminal C sequence [(0, no disturbance), (1, disturbance), (2, disturbance)], The probability density of the disturbance response (dimensionless) reflects the intensity of the disturbance response per unit time; the larger the value, the stronger the regulation capability. Total number of power terminals (integer), obtained directly by counting the number of terminals within the monitoring area. : No. The terminal disturbance frequency (non-negative integer) is obtained by performing a counting operation on the disturbance identification tag sequence generated by S201. : No. Terminal cycle length (unit: minutes). : Average cycle length (unit: minutes), through 3 power terminals ( ), disturbance frequency The invoked perturbation identification marker sequence is used to perform a counting operation on the perturbation markers: Terminal A sequence: [(t=0, no perturbation), (t=1, perturbation), (t=2, no perturbation)] → counting Terminal B sequence: [(t=0, perturbation), (t=1, no perturbation)] → counting Terminal C sequence: [(t=0, no perturbation), (t=1, perturbation), (t=2, perturbation)] → counting t represents the time point and the period length. Collection time index range (unit: min): Terminal A: → Terminal B: → Terminal C: → , Represents the index of the start time of the cycle. Represents the index of the end time of the period, average value. :calculate: Substitute into the formula Substitute the parameter values: Disturbance response probability density The disturbance response probability density is graded according to its adjustment capability. The grading threshold is based on density data statistics, which calls the database to perform quantile calculation, and the 25th percentile value is used. 75th percentile Set the range: Low capability (number 1) Medium ability (number 2) For high capability (number 3), density Located in [0.5, 2.0), therefore, the classification number is 2, and all terminals are numbered: Terminal A density calculation ( , , , , , (No. 2) Representing the Terminal cycle length, Representing the Terminal disturbance frequency Represents the average length of the period. This represents the probability density of the disturbance response. This represents the average cycle length of all power terminals, with terminal B (density calculated) (Number 2), Terminal C (density) (3), to obtain the set of regulatory capacity values ​​[2, 2, 3].

[0078] Table 2: Statistics and Density Calculation of Power Terminal Disturbance

[0079]

[0080] As shown in Table 2, the frequency, cycle length, density value and classification number of the three terminals are listed, and a set of regulation capability values ​​is generated based on this.

[0081] S203: Based on the regulation capacity numerical set, attach the corresponding capacity tag text to the regulation level number of each power terminal, and combine the number and tag into structured data. Aggregate the data according to the terminal number and time series order to generate a regulation capacity tag set.

[0082] Based on the obtained set of regulation capacity values ​​(terminal A number 2, terminal B number 2, terminal C number 3), a capacity label text is attached to the regulation level number of each power terminal. The label setting follows the classification rules: number 1 corresponds to "low capacity", number 2 corresponds to "medium capacity", and number 3 corresponds to "high capacity". A text mapping operation is performed, mapping terminal A number 2 to "medium capacity", terminal B number 2 to "medium capacity", and terminal C number 3 to "high capacity". The number and label are combined into structured data, creating key-value pairs. Terminal A outputs (2, medium capacity), terminal B outputs (2, medium capacity), and terminal C outputs (3, high capacity), according to the terminal number. Aggregate with the time series sequence, with terminal numbers taken from unique identifiers (such as ID-A, ID-B, ID-C), and the time series referencing the time point of S201. Perform array merging operations to generate a tuple sequence [ID-A, t=0, (2, medium capacity), ID-A, t=1, (2, medium capacity), ID-A, t=2, (2, medium capacity), ID-B, t=0, (2, medium capacity), ID-B, t=1, (2, medium capacity), ID-C, t=0, (3, high capacity), ID-C, t=1, (3, high capacity), ID-C, t=2, (3, high capacity)], and generate a set of adjustment capacity labels.

[0083] Please see Figure 4 The specific steps of S3 are as follows:

[0084] S301: Based on the load regulation table and regulation capacity label set, obtain the load value and corresponding label of the power terminal in each time period as input data, and combine the load status and regulation capacity to establish a load regulation task dataset;

[0085] Based on the load regulation table (including time points and high / medium / low load probabilities) and the regulation capacity tag set (including terminal ID, time point, and capacity tag), the load value of the power terminal in each time period is obtained. The load value retrieves the original collected data (e.g., the load of terminal A at t=0). The system retrieves the corresponding label (kW) and calls the corresponding adjustment capacity label set (e.g., the label for terminal A at t=0 is "medium capacity"). It combines the input data and performs a key-value matching operation, using the terminal ID and time point as the combined primary key. It then searches for the probability value in the load adjustment table and the label in the label set. For example, for terminal A at t=0: load value... kW, high probability Medium probability ,low probability Capabilities in tags form tuples Similarly, other data points are processed to create a load regulation task dataset, which is then stored as a structured array.

[0086] Table 3: Example Table of Load Adjustment Task Data

[0087]

[0088] Table 3 shows the composition structure of the input data, based on which a complete load regulation task dataset is established.

[0089] S302: Based on the load regulation task dataset, the near-end policy optimization algorithm is used for training. The optimization objective between load status and regulation capacity is set, and the policy parameters are iteratively adjusted. Through training, the optimal match between load status and regulation capacity in multiple time periods is obtained, and the optimal matching set is obtained.

[0090] Call the load regulation task dataset (as shown in Table 3) and set the optimization objective: minimize the conflict value between the high load probability and the regulation capacity. The conflict value is defined as follows: when the high probability > 0.7 and the label is medium or low capacity, perform the difference calculation: conflict value = high probability - 0.7; otherwise, it is 0. For example, terminal A at t=0: high probability Conflict value = 0, t = 1: High probability Furthermore, the conflict value of the labeled ability is 0.816 - 0.7 = 0.116. The algorithm is trained using a near-end policy optimization method, and the specific actions performed include initializing policy parameters. (Randomly generated [0.1, 0.3, 0.5]), the action space is defined as load adjustment suggestions (increase / decrease / maintain), and the state space is (load value, high probability, label number). The label number is quantified as follows: high capacity = 3, medium capacity = 2, low capacity = 1. For example, terminal A is in state (12, 0.816, 2) at t=1. The strategy parameters are adjusted iteratively, and the action selection is performed in each iteration. The softmax function is called to calculate the action probability. The state vector s is normalized. : indicates selecting the first The probability of each action, with an action space of {increase load, decrease load, maintain}. This may correspond to "increased load". : No. The policy parameter vector for each action, : The normalized state vector : Calculate the first The unnormalized score of each action, for example s=(12 / 20, 0.816 / 1, 2 / 3)=(0.6, 0.816, 0.666), initial θ=[0.1, 0.3, 0.5], calculate the probability of the load increase action: Similarly, calculate the load reduction. (Set the same θ) Maintain After probability normalization, the multi-action probability is approximately 0.333. After executing an action, the reward is calculated as -conflict value (the negative sign indicates minimizing conflict). For example, after selecting the load reduction action, the conflict value is set to 0, and the reward is 0. The policy parameters are updated, gradient ascent is adopted, and the reward variance adjustment θ is calculated. After 5 iterations, θ converges to [0.15, 0.35, 0.45], and the optimized matching set is obtained and stored as (state, optimal action) pairs, including ((12, 0.816, 2), load reduction).

[0091] S303: Based on the optimized matching set, a load regulation task sequence is formed according to the load value and regulation capacity of each power terminal and sorted according to the time series.

[0092] Based on the optimized matching set (e.g., the optimal action for terminal A at t=1 is to reduce load), load regulation tasks are generated according to the load value of the power terminal (e.g., 12kW at t=1) and the regulation capacity (capacity in the tag). The task content is encoded as: increase load = +1, reduce load = -1, maintain = 0. For example, the task for terminal A at t=1 is -1. The tasks are sorted by time sequence, arranged in ascending order by terminal ID and time point. An array sorting operation is performed, and the task sequence for terminal A is: t=0 → maintain (assuming the optimization result), t=1 → reduce load, t=2 → maintain, forming the sequence [(t=0, 0), (t=1, -1), (t=2, 0)]. Other terminal sequences are generated in the same way and aggregated into a load regulation task sequence, outputting a structured array [(ID-A, t=0, 0), (ID-A, t=1, -1), (ID-A, t=2, 0), (ID-B, t=0, 0)].

[0093] Please see Figure 5 The specific steps of S4 are as follows:

[0094] S401: Based on the load regulation task sequence, the load of the power terminal is regulated, the power change value before and after regulation is monitored, the load change amount and time interval are compared between adjacent time periods, the load change rate is calculated, and the load change slope is generated.

[0095] Based on the load adjustment task sequence, for example, the terminal A sequence includes time points t=0 (load adjustment task maintenance, task value 0), t=1 (load reduction, task value -1), and t=2 (maintaining, task value 0). Load adjustment operations are performed. At time point t=1, the load value before adjustment (12kW, from the original data in S101) is retrieved. According to the task of reducing the load, the load is adjusted to 10kW. The power change before and after adjustment is monitored, recording the power before adjustment as 12kW and the power after adjustment as 10kW. The instantaneous power change ΔL - instant = 10 - 12 = -2kW is calculated. However, for load changes in adjacent time periods, adjacent time periods are defined as continuous time index intervals. Load change rate (a dimensionless index) comprehensively reflects the severity of load changes; a higher value indicates a higher risk of system fluctuations. Load variation between adjacent time periods (unit: kW), obtained by calling the adjusted load value. and Obtained by performing a subtraction operation. The difference between adjacent time periods (unit: min) is calculated using the time index difference. Average time interval (unit: min): Calculate the arithmetic mean of all time intervals using the following formula: ,in Total number of time periods Maximum allowable load variation (unit: kW), safety threshold parameter. Maximum permissible time interval (unit: min), system response constraint threshold. Normalized reference for load variation (unit: kW), used to eliminate the influence of dimensions. Time interval normalization benchmark (unit: min), used to eliminate the influence of dimensions. Calculation of load values ​​after monitoring and adjustment, for example, the period from t=1 to t=2: (t=1), (t=2) → , Representative time period The stable load value after adjustment Representative time period The stable load value after adjustment Fixed time interval 1 minute → , Statistical time intervals (t=0, t=1: 1min, t=1, t=2: 1min) → , Based on equipment specifications (IEEE 1547 standard), the maximum allowable variation in single-phase power supply in residential areas is 5kW. , System protection action time: 2 minutes (relay protection standard) → , : 100 events | ΔL | Average value 2kW → , Δτ average value 1min→ Substitute the parameter values: Similarly, other time periods are processed, such as t=0 to t=1ΔL=0kW, Δτ=1min, v=0 is calculated, and the slope sequence [0, 4.04] is output to generate the load change slope.

[0096] S402: Call the load change slope, extract time interval data based on the start time of load value change of multiple power terminals and the response time point of reaching the adjustment target, perform interval calculation on the response process of multiple terminals when performing adjustment tasks, and obtain the adjustment response time interval;

[0097] The generated load change slope is invoked. For example, terminal A has a slope v≈4.04 from t=1 to t=2, and terminal B has a slope v=1.5 (set value) from t=0 to t=1. Based on the start time of load value changes of multiple power terminals and the response time point when the adjustment target is reached, the start time is defined as the time point when the adjustment task is executed. For example, if terminal A starts to reduce load at t=1, the response time point is defined as the time when the load stabilizes and reaches the target value. By monitoring the power curve, timestamp data is invoked, and linear interpolation calculation is performed. For example, if the start time of terminal A is t-start=1.0min, the load changes from 10kW, and the load stabilizes at 15kW (adjustment target) at t=1.2min, the response time is t-end=1.2min. The time interval data Δt=t-e is extracted. nd minus t-start = 1.2 - 1.0 = 0.2 min. Similarly, terminal B's start time t-start = 0.0 min, response time t-end = 0.3 min, Δt = 0.3 min. Terminal C's start time t-start = 0.5 min, response time t-end = 0.7 min, Δt = 0.2 min. The response process of multiple terminals performing adjustment tasks is calculated in intervals. All Δt values ​​are called, and minimum and maximum values ​​are searched. minΔt = min(0.2, 0.3, 0.2) = 0.2 min, maxΔt = max(0.2, 0.3, 0.2) = 0.3 min. The interval value [0.2, 0.3] min is output, and the adjustment response time interval value is obtained.

[0098] S403: Based on the regulation response time interval and the total load value of multiple regional power terminals within the corresponding time period, calculate the total load difference between two adjacent regulation task execution periods, collect the differences between multiple time periods and establish a unified structure dataset, and combine it with existing regulation information to form a regulation log dataset;

[0099] Based on the adjustment response time interval, such as [0.2, 0.3] min, and combined with the total load value of multiple regional power terminals within the corresponding time period, the time period is defined as the adjustment task execution period, such as t=1 to t=2. The adjusted load values ​​of all terminals at t=1 and t=2 are retrieved. Terminal A has a load of 10kW at t=1 and 15kW at t=2; Terminal B has a load of 8kW at t=1 and 8kW at t=2; Terminal C has a load of 5kW at t=1 and 7kW at t=2. The total load value is calculated: total load at t=1 = 10 + 8 + 5 = 23kW; total load at t=2 = 15 + 8 + 7 = 30kW. The difference in total load between two adjacent adjustment task execution periods is calculated as the time period difference ΔL. The total load difference is 30-23=7kW. Similarly, other time periods are processed, such as the total load difference from t=0 to t=1, ΔL-total=(10+8+5)-(10+8+5)=0kW (set value). The differences from multiple time periods are collected and stored in the difference array [0, 7]. A unified structured dataset is established, with fields including time period index, total load difference, and response time interval. Combined with existing adjustment information such as the slope sequence and interval value of S401, a data merging operation is performed. For example, the records from time period t=1 to t=2 (time period index 1-2, total load difference 7kW, response interval [0.2, 0.3]min, slope 4.04) are summarized to form an adjustment log dataset and output a structured array.

[0100] Please see Figure 6 The specific steps of S5 are as follows:

[0101] S501: Based on the regulation log dataset, extract the regulation start time, response period and load change slope of the load regulation tasks of multiple power terminals, establish a mapping relationship structure according to the task number and time order, and generate a task structured mapping value set.

[0102] Based on the adjustment log dataset (containing time period index, total load difference, response time interval, and load change slope), the adjustment start time of the load adjustment task of multiple power terminals is extracted. The start time value of the "time period index" field in the log is called, such as the time period index "1-2" corresponding to the start time of 1.0 min (time index is based on 0). The response period is called by the minimum and maximum values ​​of the "response time interval" field, such as the interval [0.2, 0.3] min, which means the response period range is 0.2 min to 0.3 min. The load change slope is called by the value of the "slope" field, such as 4.04. The task number is generated according to the rule of "terminal ID-time point", such as the task number "A-1" for terminal A at t=1. The task numbers are sorted in ascending order (ASCII code order), and the time order is sorted in ascending order of time points. A mapping relationship structure is established, and an array sorting operation is performed, first by task. Sort by number, then sort by time point. For example, task numbers "A-0" (time 0), "A-1" (time 1), "B-0" (time 0) are sorted into a sequence ["A-0", "B-0", "A-1"]. The mapping values ​​are stored as key-value pairs, where the key is the task number and the value is (start time, response period min, response period max, slope). Example: Terminal A task "A-1" mapping value (1.0, 0.2, 0.3, 4.04), terminal B task "B-0" mapping value (0.0, 0.3, 0.4, 1.5). Generate a set of structured mapping values ​​for the tasks. 4.04 comes from the slope calculated above. Output a structured array [("A-0", 0.0, 0.2, 0.3, 0), ("B-0", 0.0, 0.3, 0.4, 1.5), ("A-1", 1.0, 0.2, 0.3, 4.04)].

[0103] S502: Call the task structured mapping value set, compare the task number and sorting position of the corresponding task in the load adjustment task sequence at the same time point, calculate the task offset duration, filter the set of tasks whose offset degree exceeds the adjustment task offset threshold, and obtain the adjustment task offset amount.

[0104] The task structured mapping value set is invoked (e.g., [("A-0", 0.0, 0.2, 0.3, 0), ("B-0", 0.0, 0.3, 0.4, 1.5), ("A-1", 1.0, 0.2, 0.3, 4.04)]), compared with the original load adjustment task sequence (sequence, e.g., [("A-0", t=0, 0), ("A-1", t=1, -1)]), and the task is extracted at the same time point, e.g., at time point t=0, the mapping value set tasks "A-0" and "B-0", the original sequence task "A-0", and the task offset duration is calculated. The system calls the planned start time (original sequence time point, e.g., "A-0" planned start time 0.0 min), the current start time (start time in the mapping value, e.g., "A-0" current start time 0.0 min), and the offset duration = |current start time - planned start time| = |0.0 - 0.0| = 0 min. The task number sorting position is based on the index of the S501 mapping value set. For example, after sorting the mapping value set, the indexes are: "A-0" position 0, "B-0" position 1, "A-1" position 2. The original sequence sorting position is: "A-0" position 0, "A-1" position 2, "B-0" position 1, "A-1" position 2, "B-0" position 1, "B-0 ... Set to 1. If task "B-0" is at mapping position 1, and the original sequence has no task "B-0" (set missing), |1-undefined| is defined as the maximum value of 10 (handle missing). Set the adjustment task offset threshold. The offset duration threshold is based on data statistics. Analyze the absolute value of the offset duration of 100 adjustment events, calculate the average value of 0.1 min and the standard deviation of 0.05 min, and set the threshold = average value + 2 × standard deviation = 0.1 + 2 × 0.05 = 0.2 min. The number difference threshold is based on the proportion of the total number of tasks, with an average difference of 0.5, and set the threshold = 1.0 (integer). Filter the set of tasks whose offset exceeds the threshold. The judgment condition is: offset duration > 0.2 min or difference > 1.0. Example: Task "B-0" offset duration = |0.0-0.0| = 0 min (not exceeding), difference = |1-undefined| = 10 > 1.0 (exceeding), so it is filtered as an offset task. Task "A-1" offset duration = |1.0-1.0| = 0 min, difference = |2-1| = 1.0 (equal to the threshold, not exceeding), get the adjustment task offset amount, and store the offset task set [("B-0", offset duration 0 min, difference 10)].

[0105] Table 4: Example Table of Task Offset Calculation

[0106]

[0107] Table 4 shows the calculation results of the task offset parameter, which is used to filter tasks that exceed the threshold.

[0108] S503: Based on the adjustment task offset, rearrange the position order of all offset tasks in the original load adjustment task sequence, regenerate the terminal load response task list under the corresponding time axis, and collect the load change values ​​during the adjustment period to establish the power energy management results.

[0109] Based on the load regulation task offset (e.g., offset task set [(“B-0”, 0min, 10)]), rearrange the position order of all offset tasks in the original load regulation task sequence (sequence [(“A-0”, t=0, 0), (“A-1”, t=1, -1)]), extract the offset task “B-0” (which is not present in the original sequence), insert it into the correct position, and sort the positions according to the mapping value set. Task “B-0” should be after “A-0” and before “A-1”, with a new position index of 1. A and B are unique identifiers of the power terminal. The rearranged sequence is [(“A-0”, t=0, 0), (“B-0”, t=0, 0), (“A-1”, t=1, -1)] (task value settings are maintained). Regenerate the terminal load response task list under the corresponding time axis. The time axis is discretized at 0.1min intervals, and interpolation is performed. For tasks “A-0” and “B-0” at t=0.0min, the load response value is 0 (maintain). For task “A-1” at t=1.0min, the value is -1 (load reduction). Combined with the load change value during the adjustment period, the total load difference in the log is called. For example, the difference between t=0 and t=1 is 0kW (set). The data is collected and an array merging operation is performed. The stored fields include time point, task number, task value, and load change value. For example, at time point 0.0min, the records are (task “A-0”, 0, 0kW) and (task “B-0”, 0, 0kW). At time point 1.0min, the records are (task “A-1”, -1, 0kW). The power energy management results are established and the structured dataset is output as [(t=0.0, “A-0”, 0, 0), (t=0.0, “B-0”, 0, 0), (t=1.0, “A-1”, -1, 0)].

[0110] Please see Figure 7 AI-based energy management systems include:

[0111] The load modeling module collects load value change data from power terminals and constructs load cycle fluctuation curves. It extracts the load change slope and fluctuation amplitude and normalizes them. It analyzes the load response trend of power terminals through a Bayesian model, generates a load adjustment table, and transmits it to the adjustment identification module.

[0112] The regulation identification module compares the load regulation table with the disturbance discrimination threshold to determine if there is load disturbance data in the table. Based on the load disturbance, it identifies the power load status, calculates the load regulation capacity of the power terminal and adds a tag, generates a regulation capacity tag set and transmits it to the regulation task module.

[0113] The load adjustment task module takes the load adjustment table and the adjustment capacity label set as input, sets the optimization target between load status and load adjustment capacity through the near-end strategy optimization algorithm, generates a load adjustment task sequence and passes it to the adjustment execution module.

[0114] The regulation execution module performs load regulation of power terminals through load regulation task sequences, extracts multiple parameters such as the slope of power load change of power terminals after regulation, regulation response time and regional total load change, constructs regulation log dataset and transmits it to the result evaluation module;

[0115] The results evaluation module structures the task duration and sorting position in the adjustment log dataset and compares them with the load adjustment task sequence for consistency. If the task offset threshold is exceeded, the task sequence is rearranged to generate power energy management results.

[0116] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for energy management based on AI algorithm, characterized in that, The method comprises the following steps: S1: collecting load value change data of the power terminal and constructing a load cycle fluctuation curve, extracting load change slope and fluctuation amplitude and normalizing, analyzing the load response trend of the power terminal through a Bayesian model, and generating a load adjustment table; S2: comparing the load adjustment table with the disturbance discrimination threshold, judging whether the load adjustment table has data of load disturbance, identifying the power load state according to the load disturbance, calculating the load adjustment capacity of the power terminal and adding a label, and generating an adjustment capacity label set; S3: based on the load adjustment table and the adjustment capacity label set as input, setting the optimization target between the load state and the load adjustment capacity through a proximal policy optimization algorithm, and generating a load adjustment task sequence; S4: adjusting the power terminal load through the load adjustment task sequence, extracting the power load change slope, adjustment response time and regional total load change multiple parameters of the adjusted power terminal, and constructing an adjustment log data set; S5: structuring the task duration and sorting position in the adjustment log data set, and comparing with the load adjustment task sequence for consistency, if the adjustment task offset threshold is exceeded, rearranging the task sequence, and generating a power energy management result. 2.The AI algorithm-based energy management method of claim 1, wherein, The load adjustment table includes load change slope, fluctuation amplitude normalized value, load response trend probability, the adjustment capacity label set includes load disturbance identification mark, power adjustment capacity, power terminal label type, the load adjustment task sequence specifically is task number, target adjustment value, the adjustment log data set includes adjusted load change slope, adjustment response time interval, regional total load curve change, and the power energy management result includes task structure mapping sequence, task offset, and rearranged task order table. 3.The AI algorithm-based energy management method of claim 1, wherein, The specific steps of S1 are: S101: collecting load value change data of the power terminal in a continuous period, arranging point by point and constructing a load cycle fluctuation curve, calculating the curve change rate and extracting the load change slope, comparing the range of load values in the same period and extracting the fluctuation amplitude, and generating periodic load characteristic quantities; S102: based on the periodic load characteristic quantities, normalizing the load change slope and fluctuation amplitude, constructing a proportionality coefficient for multiple values according to the period and converting it into a dimensionless number sequence, and then recombining the processed results in time sequence to obtain a normalized feature set; S103: calling the normalized feature set, setting the prior probability distribution in the Bayesian model and constructing the likelihood function combined with the normalized feature set, calculating the response probability of the load state in multiple adjustment intervals, and pairing the time series with the corresponding response probability in order to establish a load adjustment table. 4.The AI algorithm-based energy management method of claim 1, wherein, The specific steps of S2 are: S201: comparing the multi-period load response probability in the load adjustment table with the disturbance discrimination threshold, obtaining the judgment result of whether there is a load disturbance in each period, marking the time points with disturbance as disturbance periods and the remaining time points as non-disturbance periods, and generating a disturbance identification mark sequence; S202: Call the disturbance identification marker sequence, count the frequency and response probability mean of the disturbance period of the multiple power terminals in the cycle, and calculate the disturbance response probability density combined with the corresponding cycle length, adjust the disturbance response probability density to grade the adjustment capacity and number, and obtain the adjustment capacity value set; S203: Based on the adjustment capacity value set, the adjustment level number of each power terminal is added with the corresponding capacity label text, and the number and label are combined into structured data, which is aggregated according to the terminal number and time sequence order to generate the adjustment capacity label set. 5.The AI algorithm-based energy management method according to claim 4, characterized in that, The disturbance discrimination threshold is set by adjusting the load change trend and response time fluctuation interval of multiple power terminals in the task execution process in the adjustment log data set, and the boundary range of the load slope change rate under the disturbance state is counted, and the average response time value under the normal task is combined to set. 6.The AI algorithm-based energy management method according to claim 1, wherein, The specific steps of S3 are: S301: Based on the load adjustment table and the adjustment capacity label set, the load value and corresponding label of the power terminal in each period are obtained as input data, and the load state and adjustment capacity are combined to establish a load adjustment task data set; S302: According to the load adjustment task data set, the proximal strategy optimization algorithm is used for training, the optimization target between the load state and the adjustment capacity is set, the strategy parameters are iteratively adjusted, the optimal matching of the multiple period load state and the adjustment capacity is obtained through training, and the optimization matching set is obtained; S303: Based on the optimization matching set, according to the load value and adjustment capacity of each power terminal, and in time sequence order, a load adjustment task sequence is formed. 7.The AI algorithm-based energy management method of claim 1, wherein, The specific steps of S4 are: S401: Based on the load adjustment task sequence, the load adjustment of the power terminal is carried out, the power change value before and after adjustment is monitored, the load change amount and time interval of adjacent periods are compared, the load change rate is calculated, and the load change slope is generated; S402: Call the load change slope, extract the time interval data according to the load value change start time of multiple power terminals and the response time point of reaching the adjustment target, calculate the response process interval of multiple terminal adjustment tasks, and obtain the adjustment response time interval; S403: According to the adjustment response time interval and the total load value of multiple regional power terminals in the corresponding period, the total load difference value between adjacent two adjustment task execution periods is calculated, the multiple period difference values are collected and a unified structure data set is established, and combined with the existing adjustment information to form an adjustment log data set. 8.The AI algorithm-based energy management method according to claim 1, wherein, The specific steps of S5 are: S501: Based on the adjustment log data set, the adjustment start time, response cycle and load change slope of the load adjustment task of multiple power terminals are extracted, the mapping relationship structure is established according to the task number and time sequence, and the task structured mapping set is generated; S502: Call the task structured mapping set, compare the task length and sorting position of the corresponding task in the load adjustment task sequence at the same time point, calculate the task offset length, filter the task set whose offset degree exceeds the adjustment task offset threshold, and obtain the adjustment task offset amount; S503: According to the adjustment task offset, rearrange the position sequence of all offset tasks in the original load adjustment task sequence, and regenerate the terminal load response task list under the corresponding time axis, combine the adjustment period load change value for collection, and establish the power energy management result. 9.The AI algorithm-based energy management method of claim 1, wherein, The adjustment task offset threshold is set by adjusting the time distribution range and number continuity of multiple tasks in the current response process in the adjustment log data set, and the allowed sorting error and response time delay of multiple task types in the standard execution sequence are counted, combined with the task type execution characteristics.

10. An energy management system based on AI algorithm, characterized by, The system is used to implement the energy management method based on the AI algorithm in any one of claims 1-9, and the system comprises: The load modeling module collects the load value change data of the power terminal and constructs the load cycle fluctuation curve, extracts the load change slope and fluctuation amplitude and normalizes them, analyzes the load response trend of the power terminal through the Bayesian model, generates the load adjustment table, and delivers it to the adjustment identification module; The adjustment identification module compares the load adjustment table with the disturbance judgment threshold, judges whether there is load disturbance data in the load adjustment table, identifies the power load state according to the load disturbance, calculates the load adjustment capacity of the power terminal and adds a label, generates the adjustment capacity label set, and delivers it to the adjustment task module; The adjustment task module takes the load adjustment table and the adjustment capacity label set as input, sets the optimization goal between the load state and the load adjustment capacity through the proximal strategy optimization algorithm, generates the load adjustment task sequence, and delivers it to the adjustment execution module; The adjustment execution module adjusts the power terminal load through the load adjustment task sequence, extracts the power load change slope, adjustment response time and regional total load change parameters of the power terminal after adjustment, constructs the adjustment log data set, and delivers it to the result evaluation module; The result evaluation module structures the task duration and sorting position in the adjustment log data set, and compares the consistency with the load adjustment task sequence. If the adjustment task offset threshold is exceeded, the task sequence is rearranged, and the power energy management result is generated.

Images