Campus solar charging control method and device and storage medium

Abstract

The application discloses a kind of solar energy charging control methods for campus, including the following steps, S1 gathers campus historical power consumption information, including different time period's power consumption, power consumption peak and low value, S2 uses data analysis tool and algorithm, in-depth analysis is carried out to historical power consumption data, S3 is constructed by the data analysis result of S2, power consumption prediction model, S4 is according to the power consumption model of S3 construction, customizes charging strategy and power consumption plan, S5 will charging strategy and power consumption plan with smart grid interact, according to the load condition of power grid and the change of electricity price, adjust the output power of campus solar power generation system and power demand.

Description

TECHNICAL FIELD

[0001] The present application relates to a solar energy charging control method, device and storage medium for a campus. BACKGROUND

[0002] In the solar energy charging control method for a campus, the electricity demand prediction is crucial. However, the existing technology has some obvious shortcomings in the prediction of campus electricity demand. The existing technology often only relies on historical electricity data for demand prediction, ignoring other important factors. For example, the activity arrangement, course changes, special events and other factors in the campus have a significant impact on electricity demand, but these factors are difficult to be fully considered in traditional prediction methods, resulting in a deviation between the prediction results and the actual demand, lack of real-time data update, and many existing prediction methods cannot timely incorporate the latest electricity data and environmental change information. The electricity consumption of the campus may change rapidly due to seasonal changes, weather changes, equipment failures and other factors, and if the data cannot be updated in real time and the prediction model cannot be adjusted, the prediction results will lose timeliness and accuracy. SUMMARY

[0003] The present application provides a solar energy charging control method, device and storage medium for a campus to solve the above-mentioned technical problems.

[0004] The technical solution of the present application is as follows:

[0005] A solar energy charging control method for a campus, comprising the following steps:

[0006] S1, collecting historical electricity information of the campus, including electricity consumption, peak and valley values in different time periods;

[0007] S2, using data analysis tools and algorithms to deeply analyze the historical electricity data;

[0008] S3, constructing a prediction model of electricity consumption based on the data analysis results of S2;

[0009] S4, customizing charging strategies and electricity plans based on the electricity consumption model constructed in S3;

[0010] S5, interacting the charging strategies and electricity plans with the smart grid, and adjusting the output power of the campus solar power generation system and the electricity demand according to the load condition and electricity price changes of the grid.

[0011] Preferably, S1 further comprises the following sub-steps:

[0012] S101, classifying and organizing the historical electricity data, and subdividing according to different regions in the campus, different seasons and different weather conditions, so as to better analyze the influence of the above factors on electricity demand;

[0013] S102, collect information about the course schedule, exam schedules, and large event plans within the campus; these activities will directly affect the electricity demand in specific areas, for example, the lighting and air conditioning in teaching buildings during exams may increase;

[0014] S103, install environmental monitoring equipment to collect real-time weather data within the campus, including temperature, humidity, and light intensity; weather conditions have a great impact on the use of electricity-consuming equipment such as air conditioning and lighting, for example, high temperature weather will cause a significant increase in air conditioning electricity consumption;

[0015] S104, monitor the flow of personnel within the campus, which can be achieved by installing people flow counting devices at the entrances of the campus, teaching building entrances, and other building entrances; changes in the number of personnel will affect the use of lighting and electrical equipment in public areas.

[0016] Preferably, the S2 further comprises the following sub-steps:

[0017] S201, use data visualization techniques to display historical electricity consumption data in the form of charts for more intuitive observation of electricity consumption patterns;

[0018] S202, correlate and analyze campus activity information with historical electricity consumption data to find the relationship between activities and electricity demand; for example, statistics on the increase in electricity consumption in teaching buildings during exams can determine the impact of large events on electricity demand in specific areas;

[0019] S203, analyze the relationship between real-time weather data and electricity demand; for example, establish a regression model of temperature and air conditioning electricity consumption to determine the trend of air conditioning electricity consumption under different temperatures; in this way, the change of air conditioning electricity demand can be predicted according to the real-time weather conditions.

[0020] Preferably, the step S3 further comprises the following sub-steps:

[0021] S301, select appropriate prediction algorithms according to the characteristics of campus electricity demand; common algorithms include time series analysis, regression analysis, neural networks, etc.; time series analysis is suitable for electricity consumption data prediction with obvious time trend; regression analysis can consider the influence of multiple factors on electricity demand; neural networks can capture complex nonlinear relationships;

[0022] S302, compare and evaluate different prediction algorithms, select algorithms with high prediction accuracy and good stability as the main electricity demand prediction method; the performance of the algorithm can be evaluated by cross-validation, root mean square error, etc.

[0023] S303, using the selected prediction algorithm, combined with the collected data, to build a campus power demand prediction model; the input of the model can include historical power consumption data, campus activity information, real-time environmental data, etc., and the output is the power demand prediction value in the future period;

[0024] S304, parameter adjustment and optimization of the prediction model to improve the prediction accuracy; it can be realized by adjusting the parameters of the algorithm, increasing the data features, improving the model structure, etc.; for example, different time window lengths, different regression variable combinations, etc. can be tried to find the optimal model parameter setting.

[0025] Preferably, the S4 further comprises the following steps:

[0026] S401, according to the power demand prediction results, combined with the solar resource monitoring data, to develop a reasonable charging control strategy; for example, in the low valley period of power demand, store the excess solar power; in the peak period of power demand, preferentially use the stored power or adjust the output power of the solar power generation system;

[0027] S402, considering the differences in power demand in different areas, develop individualized charging control strategies; for example, for the teaching building with large power consumption, the number of solar panels can be increased or the battery storage capacity can be increased; for areas with small power consumption, the configuration of the solar power generation system can be appropriately reduced;

[0028] S403, with the passage of time and the changes in actual power consumption, real-time adjustment of power demand prediction results and charging control strategy; if there is a large deviation between the actual power demand and the predicted value, the operation state of the solar power generation system can be adjusted in time to ensure the stable supply of campus power;

[0029] S404, work with the campus equipment management system, according to the power demand prediction results, arrange the maintenance and repair plan of the equipment in advance, avoid the influence of equipment failure on power demand; at the same time, according to the changes of power demand, dynamically adjust the operation mode of the equipment, improve the energy utilization efficiency.

[0030] A campus solar charging device, comprising at least one processor, at least one memory and computer program instructions stored in the memory, when the computer program instructions are executed by the processor, the above-mentioned one kind of campus solar charging control method is realized.

[0031] A storage medium, when the computer program instructions are executed by the processor, the above-mentioned one kind of campus solar charging control method is realized.

[0032] The present application solves the shortcomings that the prior art often only relies on historical power consumption data for demand prediction, and ignores other important factors, and also solves the influence of weather seasons and climate on predicted power consumption demand, realizes the purpose of accurate prediction and reduces waste, through accurate power consumption demand prediction, solar power generation storage and use can be reasonably arranged according to actual demand, avoiding excessive storage or insufficient supply of energy, greatly improving the utilization efficiency of solar energy, and also realizing the effect of dynamic adjustment and optimization configuration, with the change of actual power consumption and solar energy resources, the charging control strategy is adjusted in real time, and the solar power generation system is always operated in the optimal state. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 is a schematic diagram of a campus solar energy charging control method of the present application. DETAILED DESCRIPTION

[0034] To further illustrate the technical means and effects adopted by the present application to achieve the predetermined application purpose, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Instead, they are merely examples of methods and systems consistent with some aspects of the present application as detailed in the appended claims.

[0035] The terms used in the present application are merely for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "an" and "the" used in the present application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein means and includes any or all possible combinations of one or more associated listed items.

[0036] Embodiment 1

[0037] The present application is a campus solar energy charging control method, comprising the following steps:

[0038] S1, collecting campus historical power consumption information, including power consumption in different time periods, power consumption peak and valley;

[0039] S101, classifying and arranging historical power consumption data, and subdividing according to different regions in the campus, different seasons and different weather conditions, so as to better analyze the influence of the above factors on power consumption demand;

[0040] S102, collect information of the course arrangement, examination time, and large-scale activity plan in the campus; these activities will directly affect the electricity demand of specific areas, for example, the lighting and air conditioning of the teaching building during the examination period may increase;

[0041] S103, install environmental monitoring equipment to collect real-time weather data in the campus, including temperature, humidity, and light intensity; weather conditions have a great impact on the use of electricity-consuming equipment such as air conditioning and lighting, for example, high-temperature weather will cause a significant increase in air conditioning electricity consumption;

[0042] S104, monitor the personnel flow in the campus, which can be achieved by installing people flow statistical equipment at the entrances of the campus, teaching building, and other buildings; the change in the number of personnel will affect the use of lighting and electrical equipment in public areas.

[0043] S105, correlate the information collected in S101, S102, S103, and S104, that is, correlate the historical daily electricity consumption information with the campus course, campus weather, and personnel information, and calculate the electricity consumption of a single person in each different building on a single day.

[0044] S2, use data analysis tools and algorithms to conduct in-depth analysis of historical electricity consumption data;

[0045] S201, use data visualization technology to display the calculation results of step S105 in the form of charts, so as to more intuitively observe the electricity consumption rules;

[0046] S202, analyze the relationship between campus activity information and historical electricity consumption data; for example, statistics the increase in electricity consumption of the teaching building during the examination period to determine the impact of large-scale activities on the electricity demand of specific areas;

[0047] S203, analyze the relationship between real-time weather data and electricity demand; for example, establish a regression model of temperature and air conditioning electricity consumption to determine the change trend of air conditioning electricity consumption under different temperatures; in this way, the change of air conditioning electricity demand can be predicted according to the real-time weather conditions.

[0048] S3, build a prediction model of electricity consumption based on the data analysis results of S2;

[0049] S301, select a prediction algorithm according to the characteristics of the campus electricity demand;

[0050] S302, compare and evaluate different prediction algorithms, and select an algorithm with high prediction accuracy and good stability as the main electricity demand prediction method; the performance of the algorithm can be evaluated by cross-validation, root mean square error, and other indicators;

[0051] S303, using the selected prediction algorithm, combined with the collected data, to build a campus electricity demand prediction model; the input of the model can include historical electricity data, campus activity information, real-time environmental data, etc., and the output is the electricity demand prediction value in the future period;

[0052] S304, parameter adjustment and optimization of the prediction model to improve the prediction accuracy; it can be realized by adjusting the parameters of the algorithm, increasing the data features, improving the model structure, etc.; for example, different time window lengths, different regression variable combinations, etc. can be tried to find the optimal model parameter setting

[0053] S4, according to the electricity demand model constructed in S3, customize the charging strategy and electricity plan;

[0054] S401, according to the electricity demand prediction results, combined with solar resource monitoring data, develop a reasonable charging control strategy; for example, store the excess solar energy during the electricity demand low period; during the electricity demand peak period, preferentially use the stored electric energy or adjust the output power of the solar power generation system;

[0055] S402, considering the electricity demand difference of different areas, develop individualized charging control strategy; for example, for the teaching building with large electricity demand, the number of solar panels can be increased or the battery storage capacity can be increased; for the area with small electricity demand, the configuration of the solar power generation system can be appropriately reduced;

[0056] S403, with the passage of time and the change of actual electricity demand, real-time adjustment of electricity demand prediction results and charging control strategy; if there is a large deviation between the actual electricity demand and the prediction value, the operation state of the solar power generation system can be adjusted in time to ensure the stable supply of campus electricity;

[0057] S404, work with the campus equipment management system, according to the electricity demand prediction results, arrange the maintenance and repair plan of the equipment in advance, avoid the influence of equipment failure on electricity demand; at the same time, according to the change of electricity demand, dynamically adjust the operation mode of the equipment, improve the energy utilization efficiency.

[0058] S5, interact the charging strategy and electricity plan with the smart grid, according to the load condition and price change of the grid, adjust the output power of the campus solar power generation system and the electricity demand.

[0059] Preferably, the prediction algorithm is ARIMA algorithm, through the analysis of historical electricity data, the general formula of (ARIMA (p,d,q)) is:

[0060]

[0061] Yt Yt is the observed value (electricity consumption data) at time t.

[0062] c is the constant term.

[0063] p is the order of the autoregressive term, indicating the influence of the observations at the past p time points on the current value.

[0064] φ i are the autoregressive coefficients, i = 1, 2,..., p.

[0065] Y t-i is the observation at time t-i.

[0066] d is the difference order, used to make the time series stationary.

[0067] q is the order of the moving average term, indicating the influence of the random error terms at the past q time points on the current value.

[0068] i are the moving average coefficients, i = 1, 2,..., q.

[0069] t-j is the random error term at time t-j.

[0070] t is the random error term at time t, usually assumed to be white noise.

[0071] In this invention, according to the known past electricity consumption data Yt, Yt-1, Yt-2,..., and the rest of the collected information parameters, the estimated value of the random error term is calculated, initially for t+j (i = 1, 2,..., h), initialized to 0.

[0072] According to the formula the predicted value is calculated step by step.

[0073] Assuming that p = 2, q = 1 have been determined, and the parameters c = 10, φ1 = 0.5, φ2 = 0.3, 1 = 0.2 have been estimated, given the past electricity consumption data Y t-1 = 50, Y t-2 = 45, and the previously calculated t-1 = 2, now we need to predict Y t+1 , first calculate , here t assuming it is 0, then calculate

[0074] If Yt=48 is known at this time, the value of can be calculated.

[0075] Preferably, in step S1, the data is pre-processed, including checking the integrity and accuracy of the data, handling missing values and outliers.

[0076] Preferably, in step S2, the ACF and PACF of the historical electricity consumption data are calculated, and the order range of the autoregressive term and the moving average term is preliminarily judged according to the graphical characteristics.

[0077] In the ARIMA model, the random error term t, is set as white noise, which generally does not contain specific known parameters, and the machine error term itself usually does not contain directly determinable parameters, but its statistical characteristics are considered in the overall estimation and analysis of the model.

[0078] Further, for large-scale data sets and complex model calculations, parallel computing techniques such as distributed computing frameworks or GPU acceleration can be considered. This can greatly shorten the model training and prediction time and improve efficiency.

[0079] Further, the model is compressed and optimized to reduce the number of parameters and the computational complexity of the model. For example, pruning techniques can be used to remove redundant parameters in the model, or quantization techniques can be used to reduce the storage and computational requirements of the model. This not only improves the running efficiency of the model, but also improves the generalization ability of the model to some extent.

[0080] A campus solar charging device, comprising a solar charging device for implementing the above-mentioned campus solar charging control method.

[0081] A storage medium comprising computer program instructions for implementing the above-mentioned campus solar charging control method when executed by a processor.

[0082] Embodiment 2

[0083] The campus solar charging control method of this embodiment has the following distinguishing features from Embodiment 1:

[0084] Data collection and preprocessing optimization

[0085] Different areas of electricity data collection, the function and use characteristics of different areas within the campus are different, leading to differences in electricity mode. For example, the teaching building has a large electricity demand during class time, mainly for lighting, multimedia equipment, etc. The dormitory has an increased electricity demand at night and during rest time, which may involve lighting, computers, electrical equipment, etc. By collecting electricity data from different areas, the electricity usage patterns of these specific areas can be more accurately captured, providing more detailed information for the model, thereby improving the prediction accuracy of the overall campus electricity demand.

[0086] Analyzing electricity data for different areas can also reveal potential energy-saving opportunities. For example, if it is found that the electricity consumption of a certain area is abnormally high, further investigation can be conducted to find the cause, which may be due to equipment aging, improper use, or waste, so that appropriate energy-saving measures can be taken.

[0087] Special equipment or activity electricity data collection, special equipment in the campus, such as laboratory equipment, large computers, etc., usually have high electricity demand and use time is not fixed. Collecting electricity data of these devices can help the model better predict electricity demand fluctuations in special situations, so as to prepare for solar energy charging in advance.

[0088] Large-scale activities, such as sports meetings, cultural performances, etc., will increase temporary electricity demand during a specific time. Understanding the time and electricity demand of these activities can help adjust the output power of the solar power generation system or start the standby power supply in advance to ensure power supply during the activities.

[0089] Advanced data preprocessing:

[0090] Combination of various stationarity tests, ADF test is a commonly used stationarity test method, but it may misjudge in some cases. KPSS test checks the stationarity of data from another angle, and combining it with ADF test can improve the accuracy of judgment. For example, if ADF test shows that the data is not stationary, but KPSS test shows that the data is stationary, further analysis of the characteristics of the data is needed, and different difference methods or models may need to be used to process it.

[0091] According to the specific situation of the data, the appropriate difference method can be selected to make the data more stationary. Seasonal difference can handle data with seasonal characteristics, and first-order difference can handle data with trend. Combining the use of these two difference methods can be more flexible to adapt to different types of data.

[0092] Machine learning-based missing value and outlier processing: Traditional interpolation and correction methods may not be accurate enough for handling missing values and outliers, especially when the data is complex or has a large number of missing values. Using machine learning methods such as random forests or K-nearest neighbors can predict missing values based on other features of the data, improving the completeness of the data.

[0093] For the processing of outliers, machine learning methods can judge the rationality of outliers by analyzing the distribution and pattern of data. If the outliers are caused by data entry errors or sensor failures, etc., they can be corrected; if the outliers are reasonable anomalies caused by special circumstances, they can be marked separately and their influence can be considered in the model.

[0094] Data normalization: Different variables may have different scales and units, which can affect model training and comparison. Normalization can convert data to the same scale, making the weights of different variables in the model more reasonable. For example, Min-Max normalization can map data to the [0,1] interval, and Z-score standardization can make data have zero mean and unit variance.

[0095] Model establishment and optimization

[0096] Dynamic model selection:

[0097] Switching models according to data characteristics: Different time series data may have different characteristics, and a single ARIMA model may not be suitable for all situations. When the data shows obvious seasonality and trend, SARIMA model or Holt-Winters exponential smoothing model may be more suitable. For example, the demand for electricity in the campus may have significant differences in different seasons, and the SARIMA model can better capture this seasonal change.

[0098] Combining machine learning algorithms for model selection can take full advantage of the strengths of different models. For example, using Stacking or Blending methods to combine the prediction results of multiple different time series models can improve the accuracy and stability of the prediction. By weighting the average or voting of the prediction results of different models, the error and uncertainty of a single model can be reduced.

[0099] Establish a model library to automatically select: Establishing a model library can conveniently manage and compare different time series models. Regularly evaluate the performance of different models on historical data, and automatically select the optimal model for prediction based on the evaluation results. For example, cross-validation or other evaluation indicators can be used to compare the performance of different models, and the model that performs best on historical data can be selected for current prediction.

[0100] As data continues to update and change, the performance of the model may also change. Therefore, it is necessary to regularly re-evaluate and select the model to ensure that the model always maintains good prediction ability.

[0101] Advanced parameter optimization algorithms, traditional grid search and random search methods are less efficient when the parameter space is large. Advanced parameter optimization algorithms such as particle swarm optimization (PSO), simulated annealing algorithm, etc. can search in a larger parameter space and find the optimal model parameter combination more quickly. These algorithms simulate physical processes or biological behaviors in nature, constantly adjusting parameter values to find the optimal solution.

[0102] Bayesian optimization method uses prior knowledge and historical data to optimize model parameters more efficiently. By constructing a probability distribution model of parameters, the prior distribution is constantly updated based on historical data, gradually approaching the optimal parameters. This method can find a better parameter combination in fewer iterations, improving the accuracy and efficiency of parameter estimation.

[0103] Consider the time-varying nature of parameters, as time passes and data updates, model parameters may change. For example, the power consumption behavior of the campus may change with factors such as teaching arrangements, student numbers, equipment updates, etc. Therefore, considering the time-varying nature of parameters, using online learning algorithms to update model parameters in real time can better adapt to the dynamic changes of data.

[0104] Online learning algorithms can continuously adjust model parameters based on new data, allowing the model to reflect the latest changes in data in a timely manner. This method can improve the real-time performance and adaptability of the model, better responding to unexpected situations and changes.

[0105] In-depth analysis of external factors, in addition to temperature, season, and other common factors, changes in population density around the campus, urban development planning, and other factors may also have potential impact on campus power demand. For example, an increase in population density around the campus may lead to an increase in commercial activities around the campus, thereby affecting the power demand of the campus; infrastructure construction in urban development planning may affect the stability of power supply to the campus.

[0106] Considering these external factors can make the model more comprehensive in understanding the reasons for changes in power demand, improving the accuracy of prediction. By analyzing the relationship between external factors and power demand, possible changes in power demand can be predicted in advance, providing more accurate basis for solar charging control.

[0107] Deep learning methods extract relationships. Deep learning methods, such as convolutional neural networks (CNN) or long short-term memory networks (LSTM), can automatically extract complex relationships between external factors and electricity demand. These methods have strong feature extraction capabilities and can handle high-dimensional data and complex nonlinear relationships.

[0108] Input external factors as input features, along with historical electricity data into the deep learning model, so that the model can learn the influence pattern of external factors on electricity demand. For example, using LSTM can capture long-term dependencies in time series data and better handle the impact of external factors on electricity demand over time.

[0109] Establish a dynamic model of external factors, predict future temperature changes based on weather forecast data, and incorporate them into the electricity demand prediction model to consider the impact of temperature on electricity demand in advance. For example, before the arrival of high-temperature weather in summer, adjust the output power of solar power generation systems or increase the capacity of energy storage devices to meet the electricity demand of air conditioning and other equipment.

[0110] Establishing a dynamic model of external factors can improve the forward-looking and accuracy of the prediction. By continuously updating the predicted value of external factors, timely adjusting the electricity demand prediction and solar energy charging control strategy, and better adapting to changes in the external environment.

[0111] Model evaluation and verification optimization

[0112] Introduce more evaluation indicators. Traditional mean absolute error (MAE), root mean square error (RMSE) and other indicators mainly measure the absolute error and variance between predicted values and actual values. Introducing mean absolute percentage error (MAPE), symmetric mean absolute percentage error (SMAPE), Theil inequality coefficient and other indicators can evaluate the performance of the model from different angles.

[0113] MAPE can measure the relative error between predicted values and actual values, and has better comparability for data of different magnitudes; SMAPE solves the calculation problem of MAPE when the actual value is zero to some extent; Theil inequality coefficient can measure the proportional difference between predicted values and actual values, reflecting the prediction stability of the model.

[0114] Visual evaluation. Drawing comparison charts, error distribution charts and other visual charts of predicted values and actual values can intuitively observe the prediction effect and error distribution of the model. Through comparison charts, you can see the prediction accuracy of the model at different time periods and whether there is a systematic bias.

[0115] The error distribution chart can show the distribution of prediction errors, helping to determine whether the model's errors meet the assumptions of normal distribution, etc. If the error distribution has skewness or kurtosis, further adjustments to the model or improvements to the data processing method may be needed.

[0116] Cross-validation and robustness testing:

[0117] Strict cross-validation methods, such as leave-one-out cross-validation (LOOCV) and time series cross-validation, can more rigorously evaluate the generalization ability of the model. Leave-one-out cross-validation leaves only one sample as the test set and the rest as the training set, and performs multiple validations, which can fully utilize limited data for model evaluation.

[0118] Time series cross-validation considers the sequential nature of time series data, dividing the data into multiple subsets according to time order for cross-validation. This method can better simulate the actual application situation and avoid data leakage problems.

[0119] Robustness testing and hypothesis testing, robustness testing of the model can test the performance of the model under different data subsets, different parameter settings and different external factors. For example, randomly extract different data subsets for multiple validations to observe whether the model's prediction results are stable.

[0120] Introducing hypothesis testing methods such as t-test or F-test can test whether the difference between the model's prediction results and the actual values is statistically significant. If the difference is statistically significant, the model needs to be further optimized or the data processing method needs to be adjusted to improve the accuracy of the model.

[0121] Application and decision support optimization

[0122] Establishing a real-time data monitoring system to collect real-time campus electricity consumption data and solar resource data can allow the model to understand the current electricity demand and solar power generation situation in a timely manner. Through sensor networks and data acquisition systems, real-time monitoring of electricity consumption in various areas of the campus and solar power generation systems can be achieved.

[0123] Based on real-time data, continuously update model predictions to achieve real-time electricity demand prediction and solar energy charging control adjustment. For example, when electricity demand suddenly increases or solar power generation is insufficient, the model can immediately adjust the output power of the solar power generation system or start the backup power supply to ensure the stability of power supply.

[0124] Automated decision support system, the development of automated decision support system can automatically generate solar charging control strategy suggestions according to the prediction results of the model and real-time data. For example, when predicting that the electricity demand peak is coming, the system can automatically adjust the output power of the solar power generation system, optimize the charging and discharging strategy of the energy storage device, or start the standby power supply.

[0125] The automated decision support system can improve the efficiency and response speed of energy management, reduce the need for manual intervention, and reduce management costs. At the same time, the system can provide multiple strategy options according to different situations, allowing decision-makers to adjust according to actual conditions.

[0126] Establish an early warning mechanism. When the model predicts that the electricity demand may exceed the supply capacity of the solar power generation system or other abnormal situations occur, timely warning can allow relevant personnel to take appropriate measures. For example, adjusting the electricity consumption plan in advance, starting the emergency power supply, notifying users to save electricity, etc.

[0127] The early warning mechanism can improve the safety and reliability of energy management, avoiding equipment damage, interruption of teaching and scientific research activities, and other problems caused by insufficient power supply.

[0128] Long-term prediction ability, the long-term prediction ability of the model can provide decision support for the energy planning and sustainable development of the campus. According to the future electricity demand trend and solar resource change predicted by the model, long-term energy development and investment plans are made.

[0129] For example, if it is predicted that future electricity demand will continue to grow, it can be considered to increase the capacity of the solar power generation system, build energy storage facilities, or cooperate more closely with the power grid. At the same time, according to the change of solar resources, the layout and installation angle of the solar power generation system are reasonably planned to improve the utilization efficiency of solar energy.

[0130] Consider the synergy of energy systems, the energy system of the campus usually includes solar power generation, power grid power supply, energy storage equipment and other components. Considering the synergy of these energy systems can develop more optimized energy supply strategies.

[0131] For example, through interaction with the power grid, excess power can be sold to the grid when solar power generation is excessive, and power can be purchased from the grid when solar power generation is insufficient, achieving optimal allocation of energy. At the same time, reasonable allocation of energy storage devices can improve the utilization efficiency and stability of energy and reduce dependence on the power grid.

[0132] Developing user behavior analysis, according to the demand for electricity prediction results and user behavior data, to develop individualized energy saving suggestions and incentive measures can encourage teachers and students to actively participate in energy conservation and sustainable development action. For example, by analyzing the user's electricity habit, the user is provided with targeted energy saving suggestions, such as reasonable arrangement of electricity time, turning off unnecessary electrical equipment, etc.

[0133] At the same time, the incentive mechanism can be established to reward the users who save electricity and improve the users' energy saving awareness and enthusiasm. Developing user behavior analysis can manage energy from the demand side and realize sustainable use of energy.

Claims

1. A method for controlling the charging of a solar energy on a campus, characterized by, The method comprises the following steps: Switching models according to data characteristics, different time series data have different characteristics, when the data shows obvious seasonality and trend, SARIMA model or Holt-Winters exponential smoothing model is used; Combining machine learning algorithms for model selection, using Stacking or Blending method to combine the prediction results of multiple different time series models, through weighted average or voting of the prediction results of different models, the accuracy and stability of prediction are improved; Establish a model library to automatically select, regularly evaluate the performance of different models on historical data, and automatically select the optimal model for prediction according to the evaluation results; Cross-validation or other evaluation indicators are used to compare the performance of different models; With the continuous updating and change of data, the model is reevaluated and selected regularly to ensure that the model always maintains good prediction ability; Using particle swarm optimization algorithm PSO, simulated annealing algorithm or Bayesian optimization method for parameter optimization, searching in a larger parameter space to find the optimal model parameter combination faster; Considering the time variability of parameters, using online learning algorithm to update model parameters in real time to adapt to the dynamic changes of data; Deeply analyze the influence of external factors, including the change of population density around the campus and urban development planning factors, predict the change of electricity demand in advance by analyzing the relationship between external factors and electricity demand; Using deep learning methods, including convolutional neural network CNN or long short-term memory network LSTM, automatically extracting the complex relationship between external factors and electricity demand, inputting external factors as input features into deep learning model together with historical electricity data; Establishing an external factor dynamic model, predicting future temperature changes according to weather forecast data, and incorporating it into the electricity demand prediction model, and adjusting the electricity demand prediction and solar charging control strategy in time by continuously updating the prediction value of external factors; Introducing mean absolute percentage error MAPE, symmetric mean absolute percentage error SMAPE and Theil inequality coefficient index to evaluate the performance of the model; By drawing the comparison chart of predicted value and actual value, error distribution chart and visual chart, the prediction effect and error distribution of the model can be observed intuitively; Using leave-one-out cross-validation LOOCV and time series cross-validation to evaluate the generalization ability of the model, and conducting robustness test and hypothesis test to test the performance of the model under different data subsets, different parameter settings and different external factor changes; Establishing real-time data monitoring system to collect real-time data of campus electricity consumption and solar resource data, and continuously updating model prediction according to real-time data to realize real-time electricity demand prediction and solar charging control adjustment; Developing an automatic decision support system to automatically generate solar charging control strategy suggestions according to the prediction results of the model and real-time data; Establishing early warning mechanism to timely issue warning when the model predicts that the electricity demand may exceed the supply capacity of solar power generation system or other abnormal situations occur. Long-term energy development and investment plans are made based on predicted future electricity demand trends and solar resource changes using the long-term predictive capabilities of the models. Optimized energy supply strategies are developed considering the synergy of the energy system, including solar power generation, grid power supply, and energy storage components. Individualized energy-saving recommendations and incentives are developed based on electricity demand prediction results and user behavior data.

2. The method of claim 1, wherein, The automatic selection of the model library includes: Selecting the model that performs best on historical data for the current prediction.

3. The method of claim 1, wherein, The particle swarm optimization algorithm (PSO), simulated annealing algorithm, or Bayesian optimization method includes: By simulating physical processes or biological behaviors in nature, the parameter values are constantly adjusted to find the optimal solution. Using prior knowledge and historical data, a probability distribution model of the parameters is constructed, and the prior distribution is constantly updated based on historical data to gradually approach the optimal parameters.

4. The method of claim 1, wherein, The use of deep learning methods includes: Handling high-dimensional data and complex nonlinear relationships; Using LSTM to capture long-term dependencies in time series data to better handle the impact of external factors changing over time on electricity demand.

5. The method of claim 1, wherein, The dynamic model of external factors includes: Adjusting the output power of the solar power generation system or increasing the capacity of the energy storage device in advance to meet the electricity demand of air conditioning equipment before the arrival of high-temperature weather in summer.

6. The method of claim 1, wherein, The introduction of more evaluation indicators includes: MAPE measures the relative error between predicted and actual values, and has better comparability for data of different magnitudes. SMAPE solves the calculation problem of MAPE when the actual value is zero. Theil inequality coefficient measures the proportional difference between predicted and actual values, reflecting the prediction stability of the model.

7. The method of claim 1, wherein, Cross-validation and robustness testing include: Leave-one-out cross-validation leaves one sample as the test set and the rest as the training set, and performs multiple validations to fully utilize limited data for model evaluation. Time series cross-validation divides data into multiple subsets in chronological order, performs cross-validation, better simulates actual application conditions, and avoids data leakage problems. Randomly extract different data subsets for multiple validations to observe whether the prediction results of the model are stable; introduce t-test or F-test to test whether the difference between the prediction results of the model and the actual values has statistical significance. 8.A charging device for solar energy on campus, comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to realize the method of any one of claims 1-7.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the method of any one of claims 1-7.

Images