Power load management method based on flexible adjusting unit

Abstract

The invention belongs to the technical field of power load management, and discloses a flexible adjustment unit-based power load management method, which comprises the following steps of: performing region division according to the position of a flexible adjustment unit, acquiring real-time load data and a timestamp of each unit, and constructing operation characteristic data; using a pre-trained full-connection neural network load prediction model to generate a future load prediction curve for each unit based on the operation characteristic data; summarizing the prediction curves of all units in the region to obtain a region prediction load curve, and obtaining a target load value from a dispatching center to construct a region target load curve; and based on the regional prediction load curve and the target load curve, combining the characteristics of each adjustment unit, and adopting an optimization algorithm to formulate a cooperative adjustment strategy. Through regional management and intelligent prediction, the accuracy and efficiency of load management are improved, the constraint condition of each adjustment unit is considered, and a global optimal adjustment effect is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of power load management, and more specifically, to a power load management method based on a flexible regulation unit. Background Art

[0002] Electric load refers to the total amount of electric energy actually consumed by all electrical equipment in the power system at a certain moment, reflecting the user's immediate demand for electric energy. Traditional power load management mainly relies on the peak-shaving and frequency-regulating capabilities of large power plants and simple demand-side response measures. However, this management model has problems such as insufficient flexibility, slow response speed, and low regulation accuracy, making it difficult to adapt to the operation requirements of the power system under the new situation. In recent years, with the development of smart grid technology, the concept of flexible regulation units has been introduced into power load management. Flexible regulation units mainly refer to the adjustable load parts in the power system, such as large industrial users, commercial buildings, smart homes, etc. By intelligently monitoring and controlling these units, more sophisticated and flexible load management can be achieved.

[0003] However, there are still some shortcomings in the current management of flexible regulation units. First, there is a lack of effective regional division and coordination mechanism. Flexible regulation units in different geographical locations and grid topologies are often managed in a unified manner, ignoring local characteristics, reducing the accuracy and efficiency of regulation. Second, there is a lack of intelligent collaborative regulation strategies. Existing regulation methods mostly use fixed rules or simple optimization algorithms, which make it difficult to fully consider the characteristics and constraints of each regulation unit and cannot achieve the global optimal regulation effect. Summary of the invention

[0004] In order to overcome the problem that the prior art cannot achieve a globally optimal regulation effect, the present invention proposes a power load management method based on a flexible regulation unit to solve the above problem.

[0005] The present invention provides the following technical solutions:

[0006] The power load management method based on the flexible regulation unit includes:

[0007] According to the position of the flexible adjustment unit, the flexible adjustment unit is divided into regions;

[0008] Obtaining real-time load data and timestamps of each flexible regulation unit; constructing operation characteristic data based on the real-time load data and timestamps;

[0009] Using the pre-trained load forecasting model, based on the operating characteristic data, a load forecasting curve for the next n hours is generated for each flexible regulation unit, where n is an integer greater than or equal to 2;

[0010] Summarize the load forecast curves of all flexible regulation units in the region to obtain the regional forecast load curve;

[0011] The regional target load curve is constructed by obtaining the target load value in the region for the next n hours from the dispatch center;

[0012] Based on the regional forecast load curve and the negative regional target load curve, combined with the characteristics of each regulation unit, a coordinated regulation strategy is formulated.

[0013] Preferably, the real-time load data is obtained by collecting the average power in the last hour; the construction of the operating characteristic data based on the real-time load data and the timestamp includes: obtaining the timestamp, obtaining the hours, date, weekday and whether it is a weekday label according to the timestamp, and combining the hours, date, weekday, whether it is a weekday label and load data to obtain the operating characteristic data.

[0014] Preferably, the pre-trained load forecasting model includes:

[0015] Use fully connected neural networks as the basic framework to build machine learning models;

[0016] Obtain historical load data and corresponding timestamps, convert each piece of historical load data and timestamp into operating characteristic data, and combine it with the load data of the next n hours to obtain a piece of training data; obtain at least 100 pieces of training data to form a training data set;

[0017] Divide the training data set into a training set and a validation set according to a preset ratio;

[0018] Randomly select multiple groups of training data from the training set to form batch data, and the number of the batch data is at least two;

[0019] Select any batch of data and use the operating characteristic data as input parameters; input the input parameters into the machine learning model for forward propagation, and calculate the output value of the machine learning model, that is, the load forecast value for the next n hours;

[0020] The loss value of the current batch is calculated using the following formula:

[0021]

[0022] In the formula, L represents the loss value of the current batch, m represents the number of training data in the current batch, and Y i,j represents the actual value of the jth load of the i-th training data, represents the j-th load forecast value of the i-th training data predicted by the machine learning model;

[0023] Based on the loss value, the back-propagation algorithm is used to calculate the gradient of the loss function with respect to the model parameters. The calculated gradient is applied using the optimizer to update the weights and biases in the machine learning model.

[0024] The training process is repeated until the accuracy of the machine learning model exceeds the set threshold when tested with the validation set data. The machine learning model at this time is used as the trained load forecasting model.

[0025] Preferably, the calculation formula for the accuracy of the machine learning model is as follows:

[0026]

[0027] In the formula, Ac represents the accuracy of the load forecasting model, p represents the number of data in the validation set, I(·) represents the indicator function, and its value is 1 if the condition in the brackets is true, otherwise it is 0, and U q It represents the difference between the actual load value and the predicted load value of the qth data in the validation set, and ε is the preset error coefficient.

[0028] Preferably, the step of acquiring the regional predicted load curve includes: acquiring the load prediction curves of all flexible adjustment units in the region, adding the values ​​corresponding to all load prediction curves to obtain the regional predicted load curve.

[0029] Preferably, the constructing of the regional target load curve includes:

[0030] Establish data communication connection with the dispatch center;

[0031] Obtain regional target load values ​​for the next n hours one by one, and perform real-time data validity check on each obtained target load value, wherein the validity check includes: checking whether the target load value is within a predetermined range; checking whether the change range of the target load value relative to the previous target load value is within a predetermined range;

[0032] If the target load value at a certain time point fails the validity check, a data anomaly warning is immediately sent to the dispatch center, and a request is made to resend the target load value at that time point;

[0033] The target load values ​​that pass the validity check are stored and arranged in chronological order to form a regional target load curve.

[0034] Preferably, the formulation of a collaborative regulation strategy includes:

[0035] Define the optimization goal and collect the constraints of each adjustment unit;

[0036] Construct an optimization problem according to the optimization objective and the constraints of each adjustment unit;

[0037] Apply optimization algorithms to solve the constructed optimization problems;

[0038] Generate a collaborative adjustment strategy based on the solution of the optimization problem.

[0039] Preferably, the optimization goal is to minimize the difference between the regional forecast load curve and the regional target load curve;

[0040] The constraints of each regulating unit include: total regulating amount, maximum regulating capacity of each regulating unit and response time;

[0041] The construction of the optimization problem includes:

[0042] The objective function is set as: the sum of squares of the differences between the regional forecast load curve and the regional target load curve;

[0043] The constraints are set as follows: the total regulation amount in each time period does not exceed the total regulation demand, the regulation amount of each regulation unit does not exceed its maximum regulation capacity, and the regulation response time does not exceed the preset response time limit;

[0044] The optimization algorithm adopts a mixed integer programming method and uses a branch and bound algorithm to solve;

[0045] The coordinated regulation strategy includes the regulation start time, regulation duration and regulation power size of each regulation unit.

[0046] The present invention provides a power load management method based on a flexible regulation unit, which has the following beneficial effects:

[0047] By finely dividing the flexible regulation units into different regions, the geographical characteristics and grid structure differences of each region are fully considered, and more accurate load management is achieved. This regionalized management method not only improves the accuracy of regulation, but also significantly improves the overall regulation efficiency. By constructing a load forecasting model and formulating a regulation strategy for each region, it better adapts to the characteristics of the local power grid and avoids the limitations brought by unified management.

[0048] The load forecasting model constructed by using a fully connected neural network can effectively capture complex load change patterns and greatly improve the prediction accuracy. When formulating the regulation strategy, the mixed integer programming method and branch-and-bound algorithm are used, which not only considers the maximum regulation capacity, response time and other characteristics and constraints of each regulation unit, but also incorporates the regulation needs of the entire region into the optimization goal. This method breaks through the limitations of fixed rules, can dynamically adjust the strategy according to real-time conditions, and achieves the global optimal regulation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flow chart of the power load management method based on the flexible regulation unit of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] Example 1

[0052] See also Figure 1 In this embodiment, the power load management method based on the flexible adjustment unit includes:

[0053] S1. Divide the flexible adjustment unit into regions according to the position of the flexible adjustment unit;

[0054] S2. Obtain real-time load data and timestamp of each flexible regulation unit; construct operation characteristic data according to the real-time load data and timestamp;

[0055] The real-time load data is obtained by collecting the average power in the last hour; the construction of operating characteristic data based on the real-time load data and timestamp includes: obtaining the timestamp, obtaining the hours, date, weekday and whether it is a weekday label according to the timestamp, and combining the hours, date, weekday, whether it is a weekday label and load data to obtain the operating characteristic data.

[0056] In this embodiment, the process of dividing the flexible adjustment unit area and acquiring the operation characteristic data may be as follows:

[0057] According to the location information of the flexible regulation unit, it is divided into different areas. The flexible regulation unit here mainly refers to the adjustable load part of the power, such as large industrial users, commercial buildings, smart homes, etc. The division can be based on criteria such as geographical location, grid topology or administrative division. For example, the flexible regulation units in the same city or the same distribution network can be divided into one area.

[0058] For each flexible regulation unit, a smart meter or power monitoring device is installed to collect real-time load data. These devices are connected to the data center through a communication network and can transmit the collected data in real time. The data center can calculate the average power once an hour as the real-time load data of the flexible regulation unit. At the same time, the corresponding timestamp information is recorded. According to the timestamp, the relevant time feature information is extracted, including the number of hours, the number of days, the number of weeks, and the working day mark. Among them, the working day mark can be determined by the pre-set calendar information. The extracted time feature information is combined with the real-time load data to form the operation feature data. These data will be used for subsequent load forecasting analysis.

[0059] S3, using the pre-trained load forecasting model, based on the operating characteristic data, generating a load forecasting curve for the next n hours for each flexible regulation unit, where n is an integer greater than or equal to 2;

[0060] The pre-trained load forecasting model includes:

[0061] Use fully connected neural networks as the basic framework to build machine learning models;

[0062] Obtain historical load data and corresponding timestamps, convert each piece of historical load data and timestamp into operating characteristic data, and combine it with the load data of the next n hours to obtain a piece of training data; obtain at least 100 pieces of training data to form a training data set;

[0063] Divide the training data set into a training set and a validation set according to a preset ratio;

[0064] Randomly select multiple groups of training data from the training set to form batch data, and the number of the batch data is at least two;

[0065] Select any batch of data and use the operating characteristic data as input parameters; input the input parameters into the machine learning model for forward propagation, and calculate the output value of the machine learning model, that is, the load forecast value for the next n hours;

[0066] The loss value of the current batch is calculated using the following formula:

[0067]

[0068] In the formula, L represents the loss value of the current batch, m represents the number of training data in the current batch, and Y i,j represents the actual value of the jth load of the i-th training data, represents the j-th load forecast value of the i-th training data predicted by the machine learning model;

[0069] Based on the loss value, the back-propagation algorithm is used to calculate the gradient of the loss function with respect to the model parameters. The calculated gradient is applied using the optimizer to update the weights and biases in the machine learning model.

[0070] The training process is repeated until the accuracy of the machine learning model exceeds the set threshold when tested with the validation set data. The machine learning model at this time is used as the trained load forecasting model.

[0071] The accuracy of the machine learning model is calculated as follows:

[0072]

[0073] In the formula, Ac represents the accuracy of the load forecasting model, p represents the number of data in the validation set, I(·) represents the indicator function, and its value is 1 if the condition in the brackets is true, otherwise it is 0, and U q It represents the difference between the actual load value and the predicted load value of the qth data in the validation set, and ε is the preset error coefficient.

[0074] In this embodiment, the training and use process of the load forecasting model can be as follows: first, a machine learning model based on a fully connected neural network is constructed. The model includes an input layer, multiple hidden layers, and an output layer. The input layer receives the operating characteristic data, and the output layer outputs the load forecast value for the next n hours.

[0075] Prepare the training data set. Extract the historical load data and corresponding timestamps of the flexible regulation unit from the historical database. Convert each piece of historical data into the operating characteristic data format and combine it with the actual load data of the next n hours to form a complete training data. Repeat this process until at least 100 training data are obtained. Randomly divide the training data set into a training set and a validation set in a ratio of 8:2.

[0076] Multiple groups of data are randomly selected from the training set to form batches, and each batch contains at least two data.

[0077] For each batch, the running feature data is input into the model and the predicted value is calculated through forward propagation. The mean square error is used as the loss function to calculate the error between the predicted value and the actual value. The gradient is calculated using the back-propagation algorithm, and the optimizer (such as Adam) is used to update the model parameters.

[0078] Repeat the training process until the accuracy of the model on the validation set reaches a preset threshold (such as 95%). The calculation of the accuracy takes into account the error range between the predicted value and the actual value. After the training is completed, the model is used to predict the load of each flexible regulation unit for the next n hours. The latest operating characteristic data is input into the model to obtain the predicted load curve.

[0079] Through machine learning technology, we can effectively capture the complex patterns of load changes, improve prediction accuracy, and provide reliable data support for the formulation of subsequent regulation strategies.

[0080] S4, summarizing the load forecast curves of all flexible regulation units in the region to obtain a regional forecast load curve;

[0081] The step of acquiring the regional forecast load curve includes: acquiring the load forecast curves of all flexible adjustment units in the region, adding the values ​​corresponding to all load forecast curves to obtain the regional forecast load curve.

[0082] S5. Build a regional target load curve by obtaining the target load value in the region for the next n hours from the dispatching center;

[0083] The constructing of the regional target load curve comprises:

[0084] Establish data communication connection with the dispatch center;

[0085] Obtain regional target load values ​​for the next n hours one by one, and perform real-time data validity check on each obtained target load value, wherein the validity check includes: checking whether the target load value is within a predetermined range; checking whether the change range of the target load value relative to the previous target load value is within a predetermined range;

[0086] If the target load value at a certain time point fails the validity check, a data anomaly warning is immediately sent to the dispatch center, and a request is made to resend the target load value at that time point;

[0087] The target load values ​​that pass the validity check are stored and arranged in chronological order to form a regional target load curve.

[0088] In this embodiment, the process of constructing the regional target load curve may be as follows:

[0089] First, a secure data communication connection is established with the power dispatching center. This is usually achieved through a dedicated communication protocol and encryption method to ensure the security and reliability of data transmission.

[0090] Start to request and receive the regional target load values ​​for the next n hours one by one. For each received target load value, immediately perform a data validity check.

[0091] The validity check includes two aspects: first, check whether the target load value is within the pre-set reasonable range, such as whether it exceeds the historical maximum load or is lower than the minimum load. Second, check whether the change of the target load value at two adjacent time points is within the allowable range to avoid unreasonable load mutations.

[0092] If the target load value at a certain time point fails the validity check, a data anomaly warning will be immediately sent to the dispatch center. At the same time, a re-request will be automatically initiated, requiring the dispatch center to resend the target load value at that time point. All target load values ​​that pass the validity check will be temporarily stored in chronological order. When the data at all time points are collected and passed the check, the data will be arranged in chronological order to form a complete regional target load curve.

[0093] Through real-time data checking and exception handling, the accuracy and rationality of the target load curve are ensured, providing a reliable reference benchmark for the subsequent formulation of load regulation strategies.

[0094] S6. Based on the regional forecast load curve and the negative regional target load curve, combined with the characteristics of each regulation unit, a coordinated regulation strategy is formulated.

[0095] The formulation of the coordinated regulation strategy includes:

[0096] Define the optimization goal and collect the constraints of each adjustment unit;

[0097] Construct an optimization problem according to the optimization objective and the constraints of each adjustment unit;

[0098] Apply optimization algorithms to solve the constructed optimization problems;

[0099] Generate a collaborative adjustment strategy based on the solution of the optimization problem.

[0100] The optimization goal is to minimize the difference between the regional forecast load curve and the regional target load curve;

[0101] The constraints of each regulating unit include: total regulating amount, maximum regulating capacity of each regulating unit and response time;

[0102] The construction of the optimization problem includes:

[0103] The objective function is set as: the sum of squares of the differences between the regional forecast load curve and the regional target load curve;

[0104] The constraints are set as follows: the total regulation amount in each time period does not exceed the total regulation demand, the regulation amount of each regulation unit does not exceed its maximum regulation capacity, and the regulation response time does not exceed the preset response time limit;

[0105] The optimization algorithm adopts a mixed integer programming method and uses a branch and bound algorithm to solve;

[0106] The coordinated regulation strategy includes the regulation start time, regulation duration and regulation power size of each regulation unit.

[0107] In this embodiment, the collaborative adjustment strategy formulation process may be as follows:

[0108] First, the optimization objective is defined as minimizing the difference between the regional predicted load curve and the regional target load curve. This means that we want to make the actual load as close to the target load as possible by adjusting each flexible load unit.

[0109] At the same time, the constraints of each regulation unit are collected. This includes the upper limit of the total regulation amount of the entire area, the maximum regulation capacity of each regulation unit, and the response time of each unit. For example, some industrial users may have a large regulation capacity but a long response time, while some commercial buildings may have a small regulation capacity but a fast response.

[0110] Based on the above information, an optimization problem is constructed. The objective function is set to the sum of the squares of the differences between the regional forecast load curve and the regional target load curve. The constraints include: the total regulation amount in each time period does not exceed the total regulation demand, the regulation amount of each regulation unit does not exceed its maximum regulation capacity, and the regulation response time does not exceed the preset limit.

[0111] A mixed integer programming method is used to solve this optimization problem, specifically using the branch and bound algorithm. This method can effectively handle problems with a mixture of discrete variables (such as the start time of regulation) and continuous variables (such as the power size of regulation).

[0112] The algorithm first relaxes the integer constraints and solves the linear programming problem. Then, through the branching process, the relaxed solution is gradually guided to the integer solution. In this process, the upper and lower bounds are set to prune the branches to improve the solution efficiency.

[0113] Finally, a coordinated regulation strategy is generated based on the solution of the optimization problem. This strategy specifies in detail the regulation start time, regulation duration, and regulation power size of each regulation unit. For example, the strategy may instruct a flexible regulation unit to reduce the load by 500kW starting at 14:00 for 2 hours, while a commercial building starts to increase the load by 200kW at 15:30 for 1 hour.

[0114] Through mathematical optimization, the optimal coordinated regulation scheme is found while satisfying various constraints, which can not only achieve the load regulation target but also fully consider the characteristics and limitations of each regulation unit.

[0115] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only one, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into one another, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0116] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A power load management method based on a flexible regulation unit, characterized in that: include: According to the position of the flexible adjustment unit, the flexible adjustment unit is divided into regions; Obtain real-time load data and timestamps for each flexible regulation unit; Construct operation characteristic data based on real-time load data and timestamps; Using the pre-trained load forecasting model, based on the operating characteristic data, a load forecasting curve for the next n hours is generated for each flexible regulation unit, where n is an integer greater than or equal to 2; Summarize the load forecast curves of all flexible regulation units in the region to obtain the regional forecast load curve; The regional target load curve is constructed by obtaining the target load value in the region for the next n hours from the dispatch center; Based on the regional forecast load curve and the negative regional target load curve, combined with the characteristics of each regulation unit, a coordinated regulation strategy is formulated.

2. The power load management method based on the flexible regulation unit according to claim 1 is characterized in that: The real-time load data is obtained by collecting the average power in the last 1 hour; The method of constructing operation characteristic data based on real-time load data and timestamp includes: obtaining timestamp, obtaining hours, date, weekday and whether it is a weekday label based on the timestamp, and combining hours, date, weekday, whether it is a weekday label and load data to obtain operation characteristic data.

3. The power load management method based on the flexible adjustment unit according to claim 2 is characterized in that: The pre-trained load forecasting model includes: Use fully connected neural networks as the basic framework to build machine learning models; Obtain historical load data and corresponding timestamps, convert each piece of historical load data and timestamp into operating characteristic data, and combine it with the load data of the next n hours to obtain a piece of training data; obtain at least 100 pieces of training data to form a training data set; Divide the training data set into a training set and a validation set according to a preset ratio; Randomly select multiple groups of training data from the training set to form batch data, and the number of the batch data is at least two; Select any batch of data and use the operating characteristic data as input parameters; input the input parameters into the machine learning model for forward propagation, and calculate the output value of the machine learning model, that is, the load forecast value for the next n hours; The loss value of the current batch is calculated using the following formula: In the formula, L represents the loss value of the current batch, m represents the number of training data in the current batch, and Y i,j represents the actual value of the jth load of the i-th training data, represents the j-th load forecast value of the i-th training data predicted by the machine learning model; Based on the loss value, the back-propagation algorithm is used to calculate the gradient of the loss function with respect to the model parameters. The calculated gradient is applied using the optimizer to update the weights and biases in the machine learning model. The training process is repeated until the accuracy of the machine learning model exceeds the set threshold when tested with the validation set data. The machine learning model at this time is used as the trained load forecasting model.

4. The power load management method based on the flexible regulation unit according to claim 3 is characterized in that: The calculation formula for the accuracy of the machine learning model is as follows: In the formula, Ac represents the accuracy of the load forecasting model, p represents the number of data in the validation set, I(·) represents the indicator function, and its value is 1 if the condition in the brackets is true, otherwise it is 0, and U q It represents the difference between the actual load value and the predicted load value of the qth data in the validation set, and ε is the preset error coefficient.

5. The power load management method based on the flexible regulation unit according to claim 4 is characterized in that: The step of acquiring the regional forecast load curve includes: acquiring the load forecast curves of all flexible adjustment units in the region, adding the values ​​corresponding to all load forecast curves to obtain the regional forecast load curve.

6. The power load management method based on the flexible regulation unit according to claim 5 is characterized in that: The constructing of the regional target load curve comprises: Establish data communication connection with the dispatch center; Obtain regional target load values ​​for the next n hours one by one, and perform real-time data validity check on each obtained target load value, wherein the validity check includes: checking whether the target load value is within a predetermined range; checking whether the change range of the target load value relative to the previous target load value is within a predetermined range; If the target load value at a certain time point fails the validity check, a data anomaly warning is immediately sent to the dispatch center, and a request is made to resend the target load value at that time point; The target load values ​​that pass the validity check are stored and arranged in chronological order to form a regional target load curve.

7. The power load management method based on the flexible regulation unit according to claim 1 is characterized in that: The formulation of the coordinated regulation strategy includes: Define the optimization goal and collect the constraints of each adjustment unit; Construct an optimization problem according to the optimization objective and the constraints of each adjustment unit; Apply optimization algorithms to solve the constructed optimization problems; Generate a collaborative adjustment strategy based on the solution of the optimization problem.

8. The power load management method based on the flexible regulation unit according to claim 7 is characterized in that: The optimization goal is to minimize the difference between the regional forecast load curve and the regional target load curve; The constraints of each regulating unit include: total regulating amount, maximum regulating capacity of each regulating unit and response time; The construction of the optimization problem includes: The objective function is set as: the sum of squares of the differences between the regional forecast load curve and the regional target load curve; The constraints are set as follows: the total regulation amount in each time period does not exceed the total regulation demand, the regulation amount of each regulation unit does not exceed its maximum regulation capacity, and the regulation response time does not exceed the preset response time limit; The optimization algorithm adopts a mixed integer programming method and uses a branch and bound algorithm to solve; The coordinated regulation strategy includes the regulation start time, regulation duration and regulation power size of each regulation unit.

Images