Distributed energy station AI intelligent regulation and control method and system based on machine learning

Abstract

The invention discloses a distributed energy station AI intelligent regulation and control method and system based on machine learning. The method comprises a data acquisition step, an analysis step, a calculation step, a regulation and control step, a judgment step, an energy supply step and a feedback step. According to the invention, by establishing the preset time sequence energy scheduling model trained by the historical power load data, periodicity and abrupt characteristics of power consumption behaviors of users can be effectively captured, complementary characteristics among distributed energy stations can be effectively reflected, local supply and demand imbalance in a specific time period can be relieved, a scheduling strategy can be timely optimized, and the scheduling efficiency can be improved. And the supply stability of the energy load is improved, so that the accuracy of energy regulation and control is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy regulation, in particular to an AI intelligent regulation method and system for distributed energy stations based on machine learning. BACKGROUND

[0002] After the distributed energy station is built, the energy supply situation between the energy central station and each energy station needs to be dispatched, the use demand of users under each energy station is analyzed, and the energy reserves of each distributed energy station are timely regulated to realize efficient allocation and utilization of energy and reduce the probability of energy waste or insufficient energy supply.

[0003] In the existing energy dispatching technology, each energy-consuming device used by the user needs to be monitored, such as heating, refrigeration, and electrical equipment, to meet the user's energy demand and energy quality requirements, count user data and analyze energy dispatching, and propose optimization suggestions to obtain reasonable energy allocation and dispatching strategies, respond to economic and carbon emission policies, and improve efficient, safe, and reliable use of energy.

[0004] At present, the Chinese patent with publication number CN119535986A proposes an intelligent building resource dispatching system and method under multiple scene weather, which determines and obtains the total operation data of all devices in the target area through an energy analysis module, divides the total operation data into nonlinear data and linear data, calculates the proportion of nonlinear data and linear data, calculates the dispatching allocation result according to the proportion of nonlinear data and linear data through an energy dispatching module, and controls the energy supply of solar energy, wind energy, geothermal energy, or electric energy to the target area through an energy supply module according to the dispatching allocation result.

[0005] For the above technical solution, this method only divides the dispatching result by the proportion of linear data and nonlinear data, the dispatching strategy is fixed, lacks environmental adaptability, and it is difficult to optimize the dispatching strategy in time, reducing the accuracy of energy regulation. SUMMARY

[0006] The purpose of the present application is to provide an AI intelligent regulation method and system for distributed energy stations based on machine learning to solve the problems raised in the background art.

[0007] In the first aspect, the present application provides an AI intelligent regulation method for distributed energy stations based on machine learning, which realizes the purpose of the application by adopting the following technical solution: The AI intelligent regulation method for distributed energy stations based on machine learning comprises the following steps: Obtain data: the energy central station connects N distributed energy stations and M groups of users thereof, and obtains historical power load of users and historical power load of energy stations, basic energy storage regulation amount and net amount of electricity exchanged; Analysis: Establish a preset time series energy dispatch model based on the user's historical electricity load, and output the predicted electricity load of the mth group of users at the nth energy station in the tth period ; Calculation: The predicted power load of the nth energy station in period t is: ,in is the priority weight of the mth group of users at the nth energy station; Control: The predicted power load of the energy station during period t is: , where η is the loss coefficient, is the energy storage regulation of the nth energy station in period t, The calculation model is: ,in is the mean value of the net amount of electricity exchanged at the nth energy station; Energy supply: During the period t, according to Dispatch the energy supply of the energy station, according to Regulate the energy storage regulation of the nth energy station according to Allocate energy supply to the nth energy station.

[0008] By adopting the above technical solution and establishing a preset time-series energy scheduling model trained with historical electricity load data, the periodic and sudden characteristics of user electricity consumption behavior can be effectively captured. Compared with the traditional proportional division method or the prediction method based on a fixed growth rate, the machine learning model can also adaptively learn the impact of external factors such as seasons, weather, holidays and energy prices on electricity load, reducing the prediction deviation caused by empirical assumptions. By aggregating the load demand at the user level to the distributed energy station level and further counting it at the energy center level, the calculation complexity can be reduced to a certain extent. Differentiated energy distribution can be achieved through priority weights, and the supply stability of key energy loads can be improved. The calculation of the energy storage regulation amount introduces the mean of the historical net power exchange amount, which can effectively reflect the complementary characteristics between distributed energy stations, alleviate the local supply and demand imbalance in a specific period of time, improve the supply stability of energy loads, and thus improve the accuracy of energy regulation.

[0009] Optionally, in the control step, The calculation model is modified. The calculation model is updated as follows: ,in is the dynamic adjustment weight of the nth energy station, , The value of is determined based on the user's historical electricity load. is the static adjustment weight of the nth energy station, .

[0010] By adopting the technical scheme, the energy storage strategy of different energy stations can be adjusted according to the historical load characteristics of the energy stations by setting the dynamic adjustment weight and the static adjustment weight, the value of the dynamic adjustment weight is determined based on the historical load data of the user, so that the model can adapt to the long-term load change trend, the static adjustment weight serves as a fixed coefficient, and the influence of short-term fluctuations on the overall energy dispatching strategy can be reduced, for example, for an energy station with frequent power exchange, the static adjustment weight is increased to amplify the influence of the net power exchange, and when the temporary power exchange net amount of multiple energy stations increases sharply due to extreme weather, the static adjustment weight can play a restraining role, reduce the violent shock of the energy storage adjustment amount, improve the supply stability of the energy load, and thus improve the accuracy of energy regulation.

[0011] Optionally, in the data acquisition step, the average user electricity load of the nth energy station is also acquired ; In the regulation step, the value of the dynamic adjustment weight of the nth energy station is determined, and the value calculation model of the dynamic adjustment weight is specifically as follows: . .

[0012] By adopting the technical scheme, the dynamic adjustment weight is quantified by the difference between the real-time load and the average value of the user in a specified period, so that the energy storage adjustment amount can respond to load fluctuations in a timely manner, for example, for an energy station with large load fluctuations, the dynamic adjustment weight can be increased to increase the flexibility of the basic adjustment amount, thereby reserving more energy storage buffer space, reducing the probability of dispatching failure due to demand surge, improving the supply stability of the energy load, and thus improving the accuracy of energy regulation.

[0013] Optionally, the regulation step is followed by a judgment step and a feedback step. Judgment: the energy storage adjustment change amount of the nth energy station in the t period is , the average energy storage adjustment change amount in the t period is , the alert error in the t period is , the set alert value is E, if , it is judged that the adjustment is normal, and the energy supply step is executed, if , it is judged that the adjustment is abnormal, and the feedback step is executed. Feedback: generate an energy supply adjustment abnormality early warning in the t period and upload it to the management end.

[0014] By adopting the above technical solution, abnormality monitoring is carried out by calculating the warning error of the energy storage regulation change, and the warning error is used to measure the discrete degree of the energy storage regulation change of each energy station. It is possible to effectively identify abnormal situations where the group decision-making behavior of the energy station deviates. According to the comparison result of the warning error and the warning value, an abnormal early warning mechanism is established to reduce the probability of the diffusion execution of the erroneous scheduling strategy, improve the accuracy of the coordination and joint decision-making between sites, improve the overall energy efficiency of distributed energy sites, and improve the supply stability of energy loads, thereby improving the accuracy of energy regulation.

[0015] Optionally, in the judgment step, let For the first period, For the second period, , get the warning error of the first period and the warning error of the second period ,according to and The value of the warning value E is determined. The specific calculation model of the value of E is: .

[0016] By adopting the above technical solution, the calculation of the warning value is dynamically adjusted through the warning errors of the first time period and the second time period, so that the warning value can be automatically adjusted following the periodic or seasonal load changes. The average of the warning errors of similar time periods at a certain interval is used as the warning value. This can reduce the situation where the warning value suddenly changes due to abnormal data in a single time period, reduce the dependence on experience presets, and reduce the probability of manual repeated intervention in the adjustment strategy. For example, the warning value is increased during the peak electricity consumption season to tolerate greater fluctuations in the energy storage regulation amount. The system can adaptively update the warning value, reduce operation and maintenance costs, and improve the supply stability of energy load, thereby improving the accuracy of energy regulation.

[0017] In a second aspect, the present invention provides an AI intelligent control system for distributed energy stations based on machine learning, which uses the following technical solutions to achieve the invention objectives: The AI ​​intelligent control system for distributed energy stations based on machine learning includes the following modules: Acquisition module: The output end is connected to the input end of the analysis module. The energy station is connected to N distributed energy stations and their M groups of users to obtain the historical power load of users and the historical power load of energy stations, as well as the basic energy storage regulation. and net amount of electricity exchanged; Analysis module: The input end is connected to the output end of the acquisition module, and the output end is connected to the input end of the calculation module. It is used to establish a preset time series energy scheduling model based on the user's historical electricity load and output the predicted electricity load of the mth group of users at the nth energy station in the t period ; The calculation module is connected with the output end of the analysis module and the input end of the regulation module, and is used for calculating the predicted electricity load of the nth energy station in the t period according to the historical electricity load data of the nth energy station and the priority weight of the nth energy station and the priority weight of the mth group of users of the nth energy station The calculation module is connected with the output end of the analysis module and the input end of the regulation module, and is used for calculating the predicted electricity load of the nth energy station in the t period according to the historical electricity load data of the nth energy station and the priority weight of the nth energy station ; The regulation module is connected with the output end of the calculation module and the input end of the energy supply module, and is used for calculating the energy storage adjustment amount of the nth energy station in the t period according to the predicted electricity load of the nth energy station in the t period, the predicted electricity load of the total energy station in the t period, the power exchange net amount of the nth energy station in the t period, and the loss coefficient η and the power exchange net amount of the nth energy station The calculation module is connected with the output end of the analysis module and the input end of the regulation module, and is used for calculating the predicted electricity load of the nth energy station in the t period according to the historical electricity load data of the nth energy station and the priority weight of the nth energy station , and the loss coefficient η 、 The calculation module is connected with the output end of the analysis module and the input end of the regulation module, and is used for calculating the predicted electricity load of the nth energy station in the t period according to the historical electricity load data of the nth energy station and the priority weight of the nth energy station ; The energy supply module is connected with the output end of the regulation module, and is used for scheduling the energy supply of the total energy station in the t period according to the predicted electricity load of the total energy station in the t period, the energy storage adjustment amount of the nth energy station in the t period, and the energy supply of the nth energy station The regulation module is connected with the output end of the calculation module and the input end of the energy supply module, and is used for calculating the energy storage adjustment amount of the nth energy station in the t period according to the predicted electricity load of the nth energy station in the t period, the predicted electricity load of the total energy station in the t period, the power exchange net amount of the nth energy station in the t period, and the loss coefficient η The energy supply module is connected with the output end of the regulation module, and is used for scheduling the energy supply of the total energy station in the t period according to the predicted electricity load of the total energy station in the t period, the energy storage adjustment amount of the nth energy station in the t period, and the energy supply of the nth energy station The energy supply module is connected with the output end of the regulation module, and is used for scheduling the energy supply of the total energy station in the t period according to the predicted electricity load of the total energy station in the t period, the energy storage adjustment amount of the nth energy station in the t period, and the energy supply of the nth energy station

[0018] By adopting the above technical scheme, the preset time sequence energy scheduling model trained by the historical electricity load data can effectively capture the periodicity and burstiness characteristics of user electricity consumption behavior. Compared with the traditional proportional division method or the prediction method based on fixed growth rate, the machine learning model can simultaneously adaptively learn the influence of external factors such as season, weather, holiday, and energy price on electricity load, reduce the prediction deviation caused by experience hypothesis, and reduce the calculation complexity to a certain extent by summarizing the load demand of the user level to the distributed energy station level and further to the total energy station level. The differentiated energy allocation is realized through the priority weight, the supply stability of the key energy load is improved, the historical power exchange net amount is introduced into the calculation of the energy storage adjustment amount, the complementary characteristics between the distributed energy stations are effectively reflected, the local supply-demand imbalance in a specific period is alleviated, the supply stability of the energy load is improved, and the accuracy of energy regulation is improved.

[0019] Optionally, in the regulation module, the calculation model of the nth energy station is corrected according to the dynamic adjustment weight and the static adjustment weight of the nth energy station , and the calculation model after correction is updated , wherein , .

[0020] By adopting the technical scheme, the energy storage strategy of different energy stations can be adjusted according to the historical load characteristics of the energy stations by setting the dynamic adjustment weight and the static adjustment weight, the value of the dynamic adjustment weight is determined based on the historical load data of the user, so that the model can adapt to the long-term load change trend, the static adjustment weight is used as a fixed coefficient, which can reduce the influence of short-term fluctuations on the overall energy dispatching strategy, for example, for the energy station with frequent power exchange, the static adjustment weight is increased to amplify the influence of the net power exchange, and when the temporary power exchange net amount of multiple energy stations increases sharply due to extreme weather, the static adjustment weight can play a restraining role, reduce the violent shock of the energy storage adjustment amount, improve the supply stability of the energy load, and thus improve the accuracy of energy regulation.

[0021] Optionally, in the acquisition module, the user electricity load average of the nth energy station is also acquired ; In the regulation module, the value of the dynamic adjustment weight of the nth energy station is determined, and the value calculation model of the dynamic adjustment weight is . .

[0022] By adopting the technical scheme, the dynamic adjustment weight is quantified by the difference between the real-time load and the average value of the user in a specified period, so that the energy storage adjustment amount can respond to load fluctuations in time, for example, for the energy station with large load fluctuations, the dynamic adjustment weight can be increased to increase the flexibility of the basic adjustment amount, thereby reserving more energy storage buffer space, reducing the probability of dispatching failure due to demand surge, improving the supply stability of the energy load, and thus improving the accuracy of energy regulation.

[0023] Optionally, the system further comprises a judgment module and a feedback module. The judgment module: the input end is connected with the output end of the regulation module, and the output end is connected with the input end of the energy supply module and the input end of the feedback module, for calculating the energy storage adjustment change amount of the nth energy station in the t period as , calculating the energy storage adjustment change amount average in the t period as , calculating the warning error in the t period as , setting the warning value as E, if , it is judged that the adjustment is normal, and the energy supply module is executed, if , it is judged that the adjustment is abnormal, and the feedback module is executed. The feedback module: the input end is connected with the output end of the judgment module, for generating the energy supply adjustment abnormal early warning in the t period and uploading to the management end.

[0024] ​By adopting the technical scheme, the abnormality is monitored by calculating the alert error of the energy storage adjustment change amount, the alert error is used to measure the dispersion degree of the energy storage adjustment change amount of each energy station, the abnormal situation of the deviation of the group decision behavior of the energy stations can be effectively identified, the abnormal early warning mechanism is established according to the comparison result of the alert error and the alert value, the probability of the diffusion execution of the wrong scheduling strategy is reduced, the accuracy of the collaborative and joint decision between the stations is improved, the overall energy efficiency of the distributed energy stations is improved, the supply stability of the energy load is improved, and thus the accuracy of the energy regulation is improved.

[0025] Optionally, in the judgment module, the alert value E is determined according to the alert error of the first period and the alert error of the second period. The first period is a first time interval. The second period is a second time interval. The alert error of the first period is obtained. The alert error of the second period is obtained. The value of the alert value E is determined according to the alert error of the first period and the alert error of the second period. The value of the alert value E is determined according to the alert error of the first period and the alert error of the second period. The value calculation model of the alert value E is specifically as follows. .

[0026] By adopting the technical scheme, the alert value is dynamically adjusted by the alert errors of the first period and the second period, so that the alert value can automatically adjust along with the periodic or seasonal load change, the mean value of the alert errors of the adjacent time intervals at a certain interval is used as the alert value, the sudden change of the alert value caused by abnormal data in a single time interval can be reduced, the dependence on experience preset is reduced, the probability of manual repeated intervention adjustment strategy is reduced, for example, the alert value is increased in the peak electricity consumption season to tolerate greater energy storage adjustment fluctuation, the system can adaptively update the alert value, the operation and maintenance cost is reduced, the supply stability of the energy load is improved, and thus the accuracy of the energy regulation is improved.

[0027] Compared with the prior art, the beneficial effects of the present application are as follows: 1. By establishing the preset time sequence energy scheduling model trained by the historical electricity load data, the periodicity and burstiness characteristics of the user electricity consumption behavior can be effectively captured, compared with the traditional proportioning method or the prediction method based on the fixed growth rate, the machine learning model can simultaneously adaptively learn the influence of external factors such as season, weather, holiday and energy price on the electricity load, the prediction deviation caused by experience assumption is reduced, the load demand at the user level is summarized to the distributed energy station level, and then further summarized to the energy total station level, so that the calculation complexity can be reduced to a certain extent, the differentiated energy allocation is realized through the priority weight, the supply stability of the key energy load is improved, the mean value of the historical power exchange net amount is introduced into the calculation of the energy storage adjustment amount, the complementary characteristics between the distributed energy stations can be effectively reflected, the local supply and demand imbalance in a specific period is alleviated, the supply stability of the energy load is improved, and thus the accuracy of the energy regulation is improved.

[0028] 2. By setting the dynamic adjustment weight and the static adjustment weight, different energy stations can adjust the energy storage strategy according to their historical load characteristics. The value of the dynamic adjustment weight is determined based on the historical load data of the user, so that the model can adapt to the long-term load change trend. The static adjustment weight is a fixed coefficient, which can reduce the influence of short-term fluctuations on the overall energy dispatching strategy. For example, for energy stations with frequent power exchange, the static adjustment weight is increased to amplify the influence of the net power exchange. In the case of temporary power exchange net surge of multiple energy stations caused by extreme weather, the static adjustment weight can play a restraining role, reducing the sharp fluctuation of energy storage adjustment, improving the supply stability of energy load, and thus improving the accuracy of energy regulation.

[0029] 3. By quantifying the dynamic adjustment weight based on the difference between the real-time load and the average load of the user in a specified period, the energy storage adjustment can respond to load fluctuations in a timely manner. For example, for energy stations with large load fluctuations, the dynamic adjustment weight can be increased to increase the flexibility of the basic adjustment, thereby reserving more energy storage buffer space, reducing the probability of dispatch failure due to demand surge, improving the supply stability of energy load, and thus improving the accuracy of energy regulation.

[0030] 4. By calculating the warning error of the energy storage adjustment change, abnormal monitoring is performed. The warning error is used to measure the dispersion degree of the energy storage adjustment change of each energy station, which can effectively identify abnormal situations where the group decision-making behavior of energy stations deviates. According to the comparison result of the warning error and the warning value, an abnormal warning mechanism is established to reduce the probability of the spread of incorrect dispatching strategies, improve the accuracy of collaborative and joint decision-making between stations, improve the overall energy efficiency of distributed energy stations, improve the supply stability of energy load, and thus improve the accuracy of energy regulation.

[0031] 5. By dynamically adjusting the calculation of the warning value based on the warning error of the first period and the second period, the warning value can automatically adjust to periodic or seasonal load changes. The mean value of the warning error of similar periods at certain intervals is used as the warning value, which can reduce the sudden change of the warning value caused by abnormal data in a single period, reduce the dependence on empirical presets, and reduce the probability of manual repeated intervention in adjusting the strategy. For example, in the peak season of electricity consumption, the warning value is increased to tolerate larger energy storage adjustment fluctuations. The system can automatically update the warning value, reduce operation and maintenance costs, improve the supply stability of energy load, and thus improve the accuracy of energy regulation. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 Flowchart of the AI intelligent regulation method for distributed energy stations based on machine learning of the present application; Figure 2 Module block diagram of the AI intelligent regulation system for distributed energy stations based on machine learning of the present application. DETAILED DESCRIPTION

[0033] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings. Figure 1 and Figure 2 The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings.

[0034] Embodiment one: the embodiment discloses a distributed energy station AI intelligent regulation and control method based on machine learning, referring to Figure 1 , comprising the following steps: S1, data acquisition step: the energy center connects N distributed energy stations and M groups of users of each energy station, acquires historical user electricity load and historical electricity load of the energy station, basic energy storage regulation amount and power exchange net amount, acquires the user electricity load average of the nth energy station , wherein .

[0035] S2, analysis step: a preset timing energy scheduling model is established according to the historical user electricity load, and the predicted electricity load of the mth group of users of the nth energy station in the t period is output , wherein .

[0036] S3, calculation step: the predicted electricity load of the nth energy station in the t period is: , wherein is the priority weight of the mth group of users of the nth energy station.

[0037] S4, regulation step: the predicted electricity load of the energy center in the t period is: , wherein η is the loss coefficient, is the energy storage regulation amount of the nth energy station in the t period, The calculation model of is: , wherein is the dynamic adjustment weight of the nth energy station, The value calculation model of is specifically: , is the static adjustment weight of the nth energy station, , is the average value of the power exchange net amount of the nth energy station.

[0038] S5, judgment step: the energy storage regulation change amount of the nth energy station in the t period is , and the average value of the energy storage regulation change amount in the t period is The alert error of the t period is The alert value is set as E, and E is The first period is The second period is The alert error of the first period is obtained The alert error of the second period is obtained The value of the alert value E is determined according to And The value calculation model of E is: If , it is judged that the adjustment is normal, and the energy supply step S6 is executed, if , it is judged that the adjustment is abnormal, and the feedback step S7 is executed.

[0039] S6, energy supply step: in the t period, the energy supply of the energy total station is dispatched according to The energy storage adjustment amount of the nth energy station is controlled according to The energy supply of the nth energy station is distributed according to .

[0040] S7, feedback step: the energy supply adjustment abnormality early warning of the t period is generated and uploaded to the management end.

[0041] The implementation principle of the distributed energy station AI intelligent regulation method based on machine learning in this embodiment is: The energy total station is connected with N distributed energy stations, each of which is connected with M groups of users, the historical electricity load of the users and the historical electricity load of the energy station, the basic energy storage adjustment amount And the net amount of power exchange, the average electricity load of the users of the nth energy station is , a preset timing energy dispatching model is established according to the historical electricity load of the users, and the predicted electricity load of the mth group of users of the nth energy station in the t period is output .

[0042] The preset timing energy dispatching model trained by the historical electricity load data is established by machine learning technology, which can effectively capture the periodicity and burstiness characteristics of user electricity behavior, wherein the machine learning technology includes but is not limited to LSTM timing prediction model and neural network technology, etc. Compared with the traditional proportioning method or the prediction method based on fixed growth rate, the machine learning model can simultaneously adaptively learn the influence of external factors such as season, weather, holiday and energy price on electricity load, reduce the prediction deviation caused by experience hypothesis, for example, in the summer or winter period in the northern region, the model can automatically analyze and process the relationship between air conditioner load increase and temperature change, accurately predict the peak demand, and timely optimize the energy dispatching strategy, thereby improving the accuracy of energy regulation.

[0043] The predicted electricity load of the nth energy station in the t period is calculated as: wherein is the priority weight of the mth user group of the nth energy station, is a positive number, for example, industrial users or government users can be given a higher weight to improve the supply stability of the energy load of key energy-consuming equipment, and the predicted electricity load of the energy total station in the t period is: wherein η is the loss coefficient of energy transmission between the energy total station and the distributed energy station, is the energy storage adjustment amount of the nth energy station in the t period, The calculation model of is: wherein is the dynamic adjustment weight of the nth energy station, The calculation model of the value of is: , is the static adjustment weight of the nth energy station, , is the average value of the net power exchange amount of the nth energy station.

[0044] By aggregating the load demand of the user level to the distributed energy station level and further to the energy total station level, the calculation complexity can be reduced to a certain extent, the differentiated energy allocation is realized through the priority weight, the supply stability of the key energy load is improved, the average value of the historical net power exchange amount is introduced into the calculation of the energy storage adjustment amount, which can effectively reflect the complementary characteristics between the distributed energy stations, alleviate the local supply-demand imbalance in a specific period, for example, the energy station with resource redundancy in this period can transport energy to the energy station with insufficient energy storage or real-time electricity price changes in this period through the surplus of the energy storage adjustment amount, thereby improving the supply stability of the energy load and improving the accuracy of energy regulation.

[0045] Through the setting of the dynamic adjustment weight and the static adjustment weight, different energy stations can adjust the energy storage strategy according to their own historical load characteristics, the value of the dynamic adjustment weight is determined based on the historical load data of the user, so that the model can adapt to the long-term load change trend, the dynamic adjustment weight is quantified by the difference between the real-time load and the average value of the user in a specified period, so that the energy storage adjustment amount can respond to the load fluctuation in time, for example, for the energy station with large load fluctuation, the dynamic adjustment weight can be increased to increase the flexibility of the basic adjustment amount, thereby reserving more energy storage buffer space and reducing the probability of dispatch failure due to demand surge.

[0046] The static adjustment weight is a fixed coefficient, which can reduce the influence of short-term fluctuations on the overall energy scheduling strategy. For example, for an energy station with frequent power exchange, the static adjustment weight is increased to amplify the influence of the net power exchange, and for the case of temporary power exchange net surge of multiple energy stations caused by extreme weather, the static adjustment weight can play a restraining role to reduce the violent shock of energy storage adjustment amount, improve the stability of energy load supply, and thus improve the accuracy of energy regulation.

[0047] The energy storage adjustment change amount of the nth energy station in the t period is calculated as The average energy storage adjustment change amount in the t period is The alert error in the t period is The alert value E is set as The first period is The second period is The alert error of the first period is The alert error of the second period is The value of E is determined according to and The value calculation model of E is: If , it is judged that the adjustment is normal, and if , it is judged that the adjustment is abnormal.

[0048] When the adjustment is judged to be abnormal, the energy supply adjustment abnormality warning in the t period is generated and uploaded to the management end. When the adjustment is judged to be normal, the energy supply of the energy supply station is scheduled according to , the energy storage adjustment amount of the nth energy station is regulated according to , and the energy supply of the nth energy station is distributed according to .

[0049] The alert error of the energy storage adjustment change amount is calculated to monitor the abnormality. The alert error is used to measure the dispersion degree of the energy storage adjustment change amount of each energy station, which can effectively identify the abnormal situation of the deviation of the group decision-making behavior of the energy stations. According to the comparison result of the alert error and the alert value, an abnormal warning mechanism is established to reduce the probability of the spread of the execution of the wrong scheduling strategy, improve the accuracy of the collaborative and joint decision-making between the stations, improve the overall energy efficiency of the distributed energy stations, improve the stability of the energy load supply, and thus improve the accuracy of the energy regulation.

[0050] The calculation of the alert value is dynamically adjusted by the alert error of the first period and the second period, so that the alert value can automatically adjust following the periodic or seasonal load changes. The mean value of the alert error of the similar period at certain intervals is used as the alert value, which can reduce the sudden change of the alert value caused by abnormal data in a single period, reduce the dependence on empirical presets, reduce the probability of manual repeated intervention adjustment strategy, for example, in the peak season of electricity consumption, the alert value is increased to tolerate larger energy storage adjustment fluctuation. The system can automatically update the alert value, reduce the operation and maintenance cost, improve the stability of energy load supply, and thus improve the accuracy of energy regulation.

[0051] Embodiment two: the embodiment discloses a distributed energy station AI intelligent regulation system based on machine learning, referring to Figure 2 , comprising the following modules: The acquisition module: the output end is connected with the input end of the analysis module, the energy center is connected with N distributed energy stations and M groups of users of each energy station, and is used for acquiring the historical electricity load of the users and the historical electricity load of the energy station, the basic energy storage adjustment amount and the net power exchange amount, the average electricity load of the users of the nth energy station is .

[0052] The analysis module: the input end is connected with the output end of the acquisition module, and the output end is connected with the input end of the calculation module, which is used for establishing a preset time sequence energy scheduling model according to the historical electricity load of the users, and outputting the predicted electricity load of the mth group of users of the nth energy station in the t period .

[0053] The calculation module: the input end is connected with the output end of the analysis module, and the output end is connected with the input end of the regulation module, which is used for calculating the predicted electricity load of the nth energy station in the t period according to and the priority weight of the mth group of users of the nth energy station .

[0054] The regulation module: the input end is connected with the output end of the calculation module, and the output end is connected with the input end of the judgment module, which is used for calculating the energy storage adjustment amount of the nth energy station in the t period according to , the mean value of the net power exchange amount of the nth energy station, the dynamic adjustment weight and the static adjustment weight . , The value calculation model of , , and the predicted electricity load of the energy center in the t period , and the loss coefficient η are calculated according to .

[0055] Judgment module: The input end is connected to the output end of the control module, and the output end is connected to the input end of the energy supply module and the input end of the feedback module respectively, which is used to calculate the energy storage regulation change of the nth energy station in time period t: ,according to Calculate the mean value of energy storage regulation change in period t , calculate the warning error of period t as , set the warning value to E, let For the first period, For the second period, obtain the warning error of the first period and the warning error of the second period ,according to and The value of the warning value E is determined. The specific calculation model of the value of E is: ,like , then the adjustment is judged to be normal, and the energy supply module is executed. If , then it is judged that the adjustment is abnormal and the feedback module is executed.

[0056] Energy supply module: The input end is connected to the output end of the judgment module, and is used to Dispatch the energy supply of the energy station, according to Regulate the energy storage regulation of the nth energy station according to Distribute energy supply to the nth energy station.

[0057] Feedback module: The input end is connected to the output end of the judgment module, and is used to generate an abnormal warning of energy supply regulation in the t period and upload it to the management end.

[0058] The implementation principle of the AI ​​intelligent control system for distributed energy stations based on machine learning in this embodiment is as follows: The energy center is connected to N distributed energy stations, each of which is connected to M groups of users. The acquisition module obtains the historical power load of users and the historical power load of energy stations, as well as the basic energy storage regulation amount. The net amount of electricity exchanged with the electricity consumption of the nth energy station is obtained as The analysis module establishes a preset time series energy scheduling model based on the user's historical electricity load, and outputs the predicted electricity load of the m group of users at the nth energy station in the t period .

[0059] The preset time sequence energy scheduling model trained by the historical electricity load data through the machine learning technology can effectively capture the periodicity and burstiness of the user electricity consumption behavior, and can adaptively learn the influence of external factors such as season, weather, holiday and energy price on the electricity load, reduce the prediction deviation caused by experience assumption, accurately predict the peak demand, and timely optimize the energy scheduling strategy, thereby improving the accuracy of energy regulation.

[0060] The calculation module calculates the priority weight of the mth group of users of the nth energy station according to and the priority weight of the mth group of users of the nth energy station The calculation module calculates the predicted electricity load of the nth energy station in the t period as , The positive number, the regulation module calculates the energy storage adjustment amount of the nth energy station in the t period according to and the average net electricity exchange amount of the nth energy station , the dynamic adjustment weight and the static adjustment weight The calculation module calculates the predicted electricity load of the total energy station in the t period as , The value calculation model of , , and the loss coefficient η, and the predicted electricity load of the total energy station in the t period is calculated as , . .

[0061] By summarizing the load demand of the user level to the distributed energy station level, and further summarizing to the total energy station level, the calculation complexity can be reduced to a certain extent, the differentiated energy allocation is realized through the priority weight, the supply stability of the key energy load is improved, the average value of the historical net electricity exchange amount is introduced into the calculation of the energy storage adjustment amount, the complementary characteristics between the distributed energy stations can be effectively reflected, the local supply-demand imbalance in a specific period is relieved, the supply stability of the energy load is improved, and the accuracy of the energy regulation is improved.

[0062] By setting the dynamic adjustment weight and the static adjustment weight, different energy stations can adjust the energy storage strategy according to their own historical load characteristics. The value of the dynamic adjustment weight is determined based on the historical load data of the user, so that the model can adapt to the long-term load change trend. The dynamic adjustment weight is quantified by the difference between the real-time load and the average value of the user in the specified period, so that the energy storage adjustment amount can respond to the load fluctuation in time, reduce the probability of scheduling failure caused by demand surge, and the static adjustment weight as a fixed coefficient can reduce the influence of short-term fluctuations on the overall energy scheduling strategy, improve the supply stability of the energy load, and improve the accuracy of the energy regulation.

[0063] The judgment module calculates the energy storage adjustment change amount of the nth energy station in the t period as , the average of the energy storage adjustment change amount in the t period is calculated as , the alert error in the t period is , the alert value E is set, and is the first period, is the second period, , the alert error in the first period is obtained and the alert error in the second period , the value of the alert value E is determined according to and , and the value calculation model of E is: , if , it is judged that the adjustment is normal, and if , it is judged that the adjustment is abnormal.

[0064] When it is judged that the adjustment is abnormal, the feedback module generates an energy supply adjustment abnormality early warning in the t period and uploads it to the management end, and when it is judged that the adjustment is normal, the energy supply module in the t period, according to schedules the energy supply of the energy supply station, according to controls the energy storage adjustment amount of the nth energy station, and according to allocates the energy supply of the nth energy station.

[0065] By calculating the alert error of the energy storage adjustment change amount, the abnormality is monitored, the dispersion degree of the energy storage adjustment change amount of each energy station is measured by using the alert error, the abnormal situation that the group decision behavior deviates can be effectively identified, according to the comparison result of the alert error and the alert value, the abnormal early warning mechanism is established, the probability of the spread execution of the wrong scheduling strategy is reduced, the accuracy of the collaborative and joint decision between the sites is improved, the overall energy efficiency of the distributed energy station is improved, the supply stability of the energy load is improved, and the accuracy of the energy regulation is improved.

[0066] By dynamically adjusting the calculation of the alert value through the alert errors of the first period and the second period, the alert value can automatically adjust following the periodic or seasonal load change, the mean value of the alert errors of the similar periods at certain intervals is used as the alert value, the situation that the alert value suddenly changes due to abnormal data in a single period can be reduced, the dependence on experience preset is reduced, the probability of manual repeated intervention adjustment strategy is reduced, for example, the alert value increases in the peak electricity consumption season to tolerate larger energy storage adjustment amount fluctuation, the system can adaptively update the alert value, reduce the operation and maintenance cost, improve the supply stability of the energy load, and thus improve the accuracy of the energy regulation.

[0067] The above are preferred embodiments of the present application, which do not limit the protection scope of the present application, therefore: any equivalent changes made on the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. The AI ​​intelligent control method for distributed energy stations based on machine learning is characterized by: The following steps are involved: Data acquisition: The energy center connects N distributed energy stations and their M groups of users to obtain the historical power load of users and the historical power load of energy stations, as well as the basic energy storage regulation. and net amount of electricity exchanged; Analysis: Establish a preset time series energy dispatch model based on the user's historical electricity load, and output the predicted electricity load of the mth group of users at the nth energy station in the tth period ; Calculation: The predicted power load of the nth energy station in period t is: ,in is the priority weight of the mth group of users at the nth energy station; Control: The predicted power load of the energy station during period t is: , where η is the loss coefficient, is the energy storage regulation of the nth energy station in period t, The calculation model is: ,in is the mean value of the net amount of electricity exchanged at the nth energy station; Energy supply: During the period t, according to Dispatch the energy supply of the energy station, according to Regulate the energy storage regulation of the nth energy station according to Distribute energy supply to the nth energy station.

2. The AI ​​intelligent control method for distributed energy stations based on machine learning according to claim 1 is characterized by: In the control step, The calculation model is modified. The calculation model is updated as follows: ,in is the dynamic adjustment weight of the nth energy station, , The value of is determined based on the user's historical electricity load. is the static adjustment weight of the nth energy station, .

3. The AI ​​intelligent control method for distributed energy stations based on machine learning according to claim 2 is characterized by: In the data acquisition step, the average power load of users at the nth energy station is also obtained. ; In the control step, according to Dynamically adjust the weight of the nth energy station The value of is determined. The specific value calculation model of is: .

4. The AI ​​intelligent control method for distributed energy stations based on machine learning according to claim 1 is characterized in that: After the control step, there are also a judgment step and a feedback step; Judgment: The energy storage regulation change of the nth energy station in period t is The average change of energy storage regulation in period t is , the warning error in period t is , set the warning value to E, if , then the adjustment is judged to be normal and the energy supply step is executed. If , then it is judged that the adjustment is abnormal and the feedback step is executed; Feedback: Generate an abnormal warning of energy supply regulation in period t and upload it to the management terminal.

5. The AI ​​intelligent control method for distributed energy stations based on machine learning according to claim 4 is characterized in that: In the judgment step, let For the first period, For the second period, , get the warning error of the first period and the warning error of the second period ,according to and The value of the warning value E is determined. The specific calculation model of the value of E is: .

6. The AI ​​intelligent control system for distributed energy stations based on machine learning is characterized by: Includes the following modules: Acquisition module: The output end is connected to the input end of the analysis module. The energy station is connected to N distributed energy stations and their M groups of users to obtain the historical power load of users and the historical power load of energy stations, as well as the basic energy storage regulation. and net amount of electricity exchanged; Analysis module: The input end is connected to the output end of the acquisition module, and the output end is connected to the input end of the calculation module. It is used to establish a preset time series energy scheduling model based on the user's historical electricity load and output the predicted electricity load of the mth group of users at the nth energy station in the t period ; Calculation module: The input end is connected to the output end of the analysis module, and the output end is connected to the input end of the control module. and the priority weight of the mth group of users at the nth energy station Calculate the predicted power load of the nth energy station in time period t ; Control module: The input end is connected to the output end of the calculation module, and the output end is connected to the input end of the energy supply module. The average net amount of electricity exchanged with the nth energy station Calculate the energy storage regulation of the nth energy station in time period t , and according to 、 Calculate the predicted power load of the energy station during period t using the loss coefficient η ; Energy supply module: The input end is connected to the output end of the control module, and is used to Dispatch the energy supply of the energy station, according to Regulate the energy storage regulation of the nth energy station according to Distribute energy supply to the nth energy station.

7. The AI ​​intelligent control system for distributed energy stations based on machine learning according to claim 6 is characterized by: In the control module, the weight is adjusted dynamically according to the nth energy station and statically adjust weights right The calculation model is modified and the modified The calculation model is updated, where , .

8. The AI ​​intelligent control system for distributed energy stations based on machine learning according to claim 7 is characterized in that: In the acquisition module, the average power load of users at the nth energy station is also obtained. ; In the control module, according to Dynamically adjust the weight of the nth energy station The value of is determined. The specific value calculation model of is: .

9. The AI ​​intelligent control system for distributed energy stations based on machine learning according to claim 6 is characterized in that: It also includes a judgment module and a feedback module; Judgment module: The input end is connected to the output end of the control module, and the output end is connected to the input end of the energy supply module and the input end of the feedback module respectively, which is used to calculate the energy storage regulation change of the nth energy station in time period t: ,according to Calculate the mean value of energy storage regulation change in period t , calculate the warning error of period t as , set the warning value to E, if , then the adjustment is judged to be normal, and the energy supply module is executed. If , then it is judged that the adjustment is abnormal and the feedback module is executed; Feedback module: The input end is connected to the output end of the judgment module, and is used to generate an abnormal warning of energy supply regulation in the t period and upload it to the management end.

10. The AI ​​intelligent control system for distributed energy stations based on machine learning according to claim 9 is characterized in that: In the judgment module, let For the first period, For the second period, obtain the warning error of the first period and the warning error of the second period ,according to and The value of the warning value E is determined. The specific calculation model of the value of E is: .

Images