An intelligent management system for energy demand

Abstract

This application relates to an intelligent management system for energy demand, including a real-time demand prediction module, a long-term demand prediction module, a production process anomaly monitoring module, an overlimit attribution analysis module, an energy dashboard module, a parameter configuration module, etc. This application realizes the comprehensive intelligent control of energy demand through pre-event prevention, in-event suggestions, post-event analysis, and centralized configuration display; realizes intelligent prediction through intelligent algorithms, and intelligently monitors the operation status of blast furnace power generation; systematizes operation experience, with more accurate prediction and better measures; problems can be quantitatively tracked, and demand management is comprehensively and systematically realized, achieving refined management.

Description

Technical Field

[0001] The present application relates to an intelligent management system for energy demand, which is applicable to the technical field of optimal energy utilization. Background Art

[0002] For large power consumers such as iron and steel enterprises, good energy demand management can save the company's costs and reduce grid fluctuations caused by sharp changes in power consumption. Usually, in order to make full use of secondary energy sources such as blast furnace gas, converter gas, coke oven gas, steelmaking waste heat, sintering waste heat, and electric furnace waste heat, enterprises also have power generation equipment such as boiler power generation, waste heat power generation, or TRT. The power topology of an enterprise usually provides high-voltage electricity of 110KV or 220KV from the power grid to the main step-down substation of the enterprise. Each main step-down substation is downstream connected to multiple main transformers to step down the high voltage to 35KV or 10KV, respectively supplying different equipment. These equipment include power-consuming equipment and also the power generation equipment of the enterprise. In addition, these equipment can be switched at the bus tie between different transformers or even different main step-down substations, and all important equipment has two-way power supply so that normal operation can still be guaranteed when corresponding transformers, bus ties, etc. are under maintenance.

[0003] The actual externally purchased electricity energy demand of an enterprise is the total electricity consumption minus the electricity generation. The electricity bill is calculated according to the total meter count at the inlet of the main step-down substation. The demand of each main step-down substation is closely related to the power consumption load of the downstream connected equipment. The existence of power generation equipment also increases the complexity of energy demand control. When energy demand management is controlled by traditional methods, there are pain points such as large and difficult-to-predict fluctuations in energy demand, complex power systems and topologies of the whole plant, difficult to regulate, high dependence on experience, and difficult to detect in advance.

[0004] In response to this problem, in the prior art, only the energy demand of certain processes, workshops or equipment of an enterprise is monitored or controlled, and the functions such as real-time prediction, long-term prediction, attribution analysis, and equipment anomaly detection of energy demand cannot be effectively integrated systematically to achieve systematic and intelligent control of the energy demand of the whole enterprise. Summary of the Invention

[0005] The present application provides an intelligent management system for energy demand, which applies industrial intelligent algorithms and big data analysis technologies to solve the pain points faced in enterprise energy demand management and achieve good control of enterprise energy demand.

[0006] An intelligent management system for energy demand according to the present application includes the following modules:

[0007] A real-time demand prediction module, which establishes a real-time demand prediction model based on machine learning algorithms to achieve minute-level real-time demand prediction;

[0008] Long-term demand prediction module, which realizes the energy demand prediction for the next week level according to the production plan, maintenance plan, bus-coupler switching state and the data of past electrical equipment;

[0009] Production process anomaly monitoring module, which judges the anomalies of the data quality concerned in the production process, then detects the anomalies of the equipment operation state, and provides control suggestions when anomalies are detected;

[0010] Overlimit attribution analysis module, which automatically analyzes the reasons for the overlimit of energy demand and conducts manual feedback on whether the automatically analyzed overlimit reasons are correct or not;

[0011] Energy dashboard module, which can display the information on energy demand management;

[0012] Parameter configuration module, which can configure the parameters related to energy demand.

[0013] Among them, the real-time demand prediction module may include the following sub-modules:

[0014] Equipment classification module, which divides the equipment types according to the characteristics of equipment power consumption data;

[0015] Feature extraction module, which extracts features according to the equipment power consumption data, the key data affecting the equipment power consumption, and the key data representing the equipment state changes respectively;

[0016] Operating condition identification module, which identifies different operating conditions by identifying the bus-coupler state, establishes different models for different operating conditions, and trains all the simulated operating conditions respectively to obtain the prediction models for different operating conditions;

[0017] Model training module, which trains the prediction models for different operating conditions with the data after feature extraction processing by using machine learning algorithms, optimizes and adjusts the parameters, and obtains the prediction models for the future preset time;

[0018] Real-time prediction module, which uses the data of a recent period of time in real time, constructs features through the feature extraction module, and selects the model trained by the model training module to realize the prediction for the preset time.

[0019] Among them, the real-time demand prediction module may also include a model online update module. When the prediction result of the real-time prediction module exceeds the deviation threshold, the actual operating condition data is used to retrain the model through the feature extraction module again; the long-term demand prediction module divides the energy demand of the prediction period into two parts: one part is the predicted demand of the equipment related to the steel grade power consumption, and this part of the demand is calculated through the historical data and production plan of each steel grade; the other part is the predicted demand of the equipment other than the equipment related to the steel grade power consumption, which is obtained by calculating the demand change of the equipment with inconsistent operating states between the reference period and the prediction period and the demand of the equipment without start-stop changes.

[0020] Among them, the predicted required amount of the equipment related to the steel grade power consumption is calculated based on the on-off time of these equipment and the steel grades produced in the production and maintenance plan. The predicted required amount of the equipment i related to the steel grade power consumption at the prediction time t is as follows:

[0021]

[0022] Where: and are the statistical energy required amount values of the equipment i during the production of steel grade A and during shutdown for maintenance respectively, and set1 is the set of equipment related to the steel grade power consumption.

[0023] For the predicted required amount of other equipment, first, according to the total required amount data P in the actual power consumption data of the entire reference period 总需量_Base and the power consumption data of each equipment related to the steel grade calculate the reference data P of the predicted required amount of other equipment at the t-th moment Other_Base :

[0024]

[0025] Compare the switch conditions of other equipment in the reference period and the prediction period, and calculate the required amount of equipment that is running in the reference period and not running in the prediction period and the required amount of equipment that is not running in the reference period and running in the prediction period at the t-th moment respectively, so as to obtain the predicted required amount sum of other equipment at the t-th moment:

[0026]

[0027] Among them, are the statistical energy required amount values when the equipment j is running and not running respectively, set2 (t) is the set of equipment that is running in the reference period and not running in the prediction period at the t-th moment, and set3 (t) is the set of equipment that is not running in the reference period and running in the prediction period at the t-th moment;

[0028] Obtain the total required amount prediction value at the t-th moment of the prediction period

[0029] Among them, the predicted required amount per hour in the prediction period is the n-th quantile of the total required amount prediction value where 50% ≤ n ≤ 100%; finally, combine the bus-tie switching records and calculate the predicted required amount corresponding to each main substation and each moment according to the above steps respectively, and then obtain the predicted required amount per hour of each main substation in the prediction period.

[0030] Among them, the production process anomaly monitoring module may include the following sub-modules:

[0031] The data quality judgment module uses statistical or rule-based methods to preliminarily judge the quality of production process data collected by sensors, and then uses statistical methods or methods based on anomaly monitoring algorithms to determine whether the measured value deviates too much from the actual value;

[0032] The equipment anomaly detection module judges whether the operation state of the equipment is abnormal through rules or machine learning algorithms when the data quality is normal;

[0033] The regulation rule recommendation module judges the type of equipment anomaly and triggers regulation suggestions for the operator to provide reference when the equipment anomaly detection module judges that the equipment has an anomaly and the data quality judgment module judges that the data quality is normal; the regulation rule recommendation module may include an anomaly judgment standard configuration unit and a regulation rule recommendation library unit.

[0034] Among them, the operation method of the overlimit attribution analysis module includes the following steps:

[0035] (1) After judging and obtaining the overlimit moment corresponding to the main substation, divide all equipment into multiple equipment groups, and then calculate the demand change amount of each equipment group in a certain period of the past respectively;

[0036] (2) Set the weight coefficient of the demand change amount of each equipment group to obtain the sorting sequence of the weighted demand change amount of each equipment group, and select the equipment group with the largest weighted demand change amount as the main cause of overlimit;

[0037] (3) Set the weight coefficient of the demand change amount of each equipment under the equipment group with the largest weighted demand change amount, and find out the equipment with the largest weighted demand change amount under the equipment group as the main cause of demand overlimit.

[0038] Among them, the intelligent management system may further include a parameter configuration module, which can perform functions such as setting the demand control target, configuring the demand economic index, configuring the power or demand threshold of power generation and power consumption equipment, inputting and importing the production plan, and configuring the bus-tie switch; it may also include an abnormal event module, which can alarm and record all abnormal events related to energy demand.

[0039] Through the system cooperation of each functional module, this application assists in the standardized, digital, and intelligent management and operation of energy demand, upgrading the existing demand management mode mainly relying on manual experience to an intelligent management method of "intelligent system assistance + manual decision-making", realizing end-to-end intelligent auxiliary decision support, and providing possible reasons for abnormal demand and systematic operation suggestions, thus effectively realizing the intelligent control of the enterprise's energy demand and reducing the enterprise's electricity cost. Specifically, the system realizes the comprehensive intelligent control of energy demand through pre-event prevention, in-event suggestions, post-event analysis, and centralized configuration display; realizes intelligent prediction through intelligent algorithms, and intelligently monitors the operation status of blast furnace power generation; systematizes operation experience, makes more accurate predictions and better measures; can quantitatively track problems, comprehensively and systematically realize demand management, and realizes refined management. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the intelligent management system for the energy demand of this application.

[0041] Figure 2 It is a schematic diagram of the algorithm model of the long-term demand prediction module of this application.

[0042] Figure 3 It is a schematic diagram of the operation method of the over-limit attribution analysis module of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] To make the purpose, technical solutions, and advantages of this application clearer and more understandable, the embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined arbitrarily with each other. This application will be described by taking the energy demand of a steel enterprise as an example, and those skilled in the art know that the intelligent management system for the energy demand of this application can be used in other similar technical fields.

[0044] This application provides an intelligent management system for energy demand. Refer to Figure 1 , which includes the following modules:

[0045] Real-time demand prediction module: Classify equipment types according to the characteristics of equipment power consumption data, and divide the equipment into impact large-load equipment and non-large-load equipment; among them, the impact large-load equipment can be, for example, rolling lines, power generation equipment, electric furnaces, refining furnaces, or other impact large-load equipment in a steel enterprise. Extract corresponding different characteristics for different types of impact large-load equipment respectively, and also extract corresponding characteristics for non-large-load equipment, and then establish a real-time demand prediction model based on machine learning algorithms, which can achieve minute-level prediction for the future, such as 1 to 5 minutes.

[0046] Long-term demand prediction module: Based on the production schedule, maintenance plan, bus-coupler switching status, and past data of electrical equipment, it predicts the energy demand on a weekly basis for the next few days or even weeks. The prediction scope depends on the time that can be determined in advance by the production schedule and maintenance plan.

[0047] Production process anomaly monitoring module: This module first determines anomalies in the data quality of production process-related data such as the amount of blast furnace gas generated and the power generation of generators, and then detects anomalies in the operating status of key equipment such as blast furnaces and generator sets during production operation. When an anomaly occurs, it triggers the regulation rule recommendation module, which provides optimal regulation recommendations according to the type of anomaly for operators' reference.

[0048] Overlimit cause analysis module: The KPI dashboard can summarize the daily demand trend charts and overlimit times of each main substation of the previous day. The overlimit cause analysis module can automatically analyze possible overlimit causes. Operators can provide feedback on whether the automatically analyzed causes are correct or not. The system collects the feedback and conducts in-depth learning to achieve more intelligent overlimit cause analysis in the future.

[0049] Energy dashboard module: It realizes the information display of energy demand management, including real-time monitoring of the electricity load of key users, real-time monitoring of the power generation of key power generation equipment, monthly demand trend charts, and display of power economic indicators. The power economic indicators can include monthly total electricity charges, kWh electricity charges, basic electricity charges, and cost of purchased coal, etc.

[0050] Parameter configuration module: It realizes all parameter configurations related to energy demand, including functions such as setting demand control targets, electricity quantity thresholds for electrical equipment and power generation equipment, entry and import of production schedules, and bus-coupler switching configuration.

[0051] Abnormal event module: It realizes the alarm and recording of all abnormal events related to energy demand, including alarms for real-time predictions, as well as alarms and recordings when data quality is abnormal or when blast furnace power generation equipment is abnormal.

[0052] Historical query module: It realizes the historical query of information such as historical demand, electricity load, power economic indicators, overlimit alarm records of demand, regulation records, and regulation measures, which is convenient for quantitative tracking and post-event review.

[0053] The above modules can be classified by function as: pre-event prevention function, in-event recommendation function, post-event analysis function, and centralized configuration and display function. The specific implementation modules are as follows:

[0054] Pre-event prevention function: Real-time demand prediction module, long-term demand prediction module;

[0055] In-event recommendation function: Production process anomaly monitoring module, abnormal event module;

[0056] Post - analysis function: Over - limit attribution analysis module, historical query module;

[0057] Centralized configuration display function: Energy dashboard module, parameter configuration module.

[0058] The following will explain each of the above - mentioned modules in detail.

[0059] Real-time demand prediction module

[0060] The real - time demand prediction module can solve the prediction problem of real - time production. If some real - time factors cause the predicted energy demand to increase, some equipment can be temporarily shut down to avoid short - term demand over - limit. The real - time demand prediction module includes an equipment classification module, a feature extraction module, a working condition identification module, a model training module, a real - time prediction module, and a model online update module.

[0061] Equipment classification module: Classify equipment types according to the characteristics of equipment power consumption data, and divide equipment into impact large - load equipment and non - large - load equipment; among them, impact large - load equipment can be, for example, rolling lines, power generation equipment, electric furnaces, refining furnaces in steel enterprises, or other impact large - load equipment. Those skilled in the art can understand that different industries will correspond to different impact large - load equipment, and the large - load equipment here is only relative to other equipment, without specifying a specific load range. Non - large - load equipment can be equipment with relatively stable power consumption or equipment with less power consumption. Impact large - load equipment has a great impact on the fluctuation of demand. Therefore, for impact large - load equipment, it is necessary to specifically find more relevant dimensional data and extract corresponding features through the subsequent feature extraction module to predict the power consumption demand of the corresponding impact large - load equipment. For other equipment, since it has little impact on the total - drop demand fluctuation, the power consumption data of each equipment may not be required, and the difference between the total - drop import total meter and the power consumption of impact large - load equipment can be used for calculation.

[0062] Feature extraction module: Construct features to be extracted based on equipment power consumption data, key data affecting equipment power consumption, and key data representing equipment state changes respectively, and predict the future power consumption data of this type of equipment through the feature extraction module. For example, steam flow is the key data affecting equipment power consumption for a steam generator, and the gas consumption data of the rolling - line heating furnace is the key data representing the equipment state change for the rolling - line equipment.

[0063] Operating condition recognition module: By recognizing the state of the bus coupler, different operating conditions are identified. Different models are established for different operating conditions, and all simulated operating conditions are trained separately to obtain prediction models for different operating conditions. The equipment switches under different bus couplers, and the start-stop states of the equipment constitute different load mounting combinations of electric power. Considering that the electricity consumption of some equipment will change with day and night, seasonal climate changes, and the energy efficiency of some equipment will also change with time, operating conditions, etc. When there are many and complex electrical equipment, there may be some situations such as project transformation, commissioning power consumption, new equipment power consumption, and equipment scrapping reducing power consumption. Through this module, it is possible to train a model with almost no obvious decrease in prediction accuracy for almost all operating conditions without the historical data of all operating conditions.

[0064] Model training module: Use machine learning algorithms such as random forest, LightGBM, and LSTM to train the model with the data after feature extraction and processing, optimize and adjust the parameters to obtain a prediction model for the next 1-5 minutes or longer.

[0065] Real-time prediction module: It uses the data of a certain recent period in real time, constructs features through the feature extraction module, and selects the model trained by the model training module to achieve the prediction for the next 1-5 minutes or longer.

[0066] Model online update module: When the prediction result of the real-time prediction module exceeds the deviation threshold, use the actual operating condition data to retrain the model through the feature extraction module again. For example, if the equipment changes greatly, resulting in a large difference in the power consumption mode from the past, such as adding a lot of equipment with large impact loads such as rolling mills and electric furnaces, it may cause a decrease in prediction accuracy. According to the actual situation, the historical data of the corresponding operating conditions can be used to retrain the model.

[0067] Long-term demand prediction module

[0068] Since the energy demand is closely related to the total production volume, the type of steel produced, production maintenance, etc. For example, if the power consumption of all the steel produced on a certain day is large, then the energy demand on that day must be relatively high. Even if the real-time prediction module predicts an overlimit, relying on frequently shutting down equipment is not an optimal solution to avoid exceeding the energy demand limit. Another example is that if a generator is shut down for maintenance during a certain period of a certain day, if some equipment is not planned to be shut down in advance, the electricity consumption of purchased electricity will also increase, and the energy demand will exceed the limit. Therefore, it is necessary to consider the energy demand target during production scheduling and optimize the scheduling. By using the above real-time demand prediction module combined with the long-term demand prediction module, future production plans and energy balance plans can be formulated, comprehensively considering various energy factors, matching production requirements to achieve the optimal production plan, and meeting the goal of running along the line.

[0069] The algorithm model design of long-term demand prediction is shown in Figure 2, its basic logic is to divide the energy demand in the prediction period into two major parts: one part is the predicted demand of equipment related to steel types, such as rolling lines, electric furnaces, refining furnaces, etc. The electricity consumption of the equipment is related to the steel type, and this part of the demand is mainly calculated through the historical data and production scheduling plans of each steel type; the other part is the demand of the remaining equipment, which is obtained by calculating the change in the demand of equipment with inconsistent operating states between the reference period and the prediction period and the demand of equipment without start-stop changes. This part of the demand is obtained by adding or subtracting the demand of equipment with inconsistent operating states between the reference period and the prediction period from the remaining electricity data of the reference period except for the first part of the equipment. Finally, combined with the bus-coupler switching records, the predicted demand value per hour for each main substation in the prediction period is obtained. The following is a detailed description of the content therein.

[0070] Figure 2 In this, the prediction period is the production scheduling period to be predicted, and the reference period is a time range of the same length in the past, which is used as the basis for predicting the demand in the prediction period. For example, if the production scheduling and maintenance plan is in days, yesterday can be regarded as the reference period and tomorrow as the prediction period; if the production scheduling plan is in weeks, the week starting tomorrow is the prediction period, and yesterday and the previous 6 days are the reference period. The equipment demand data includes the statistical energy demand values of all operating and non-operating equipment. This value can be a fixed manually counted value, such as the power demand of 15000 KW when the rolling line is running and 2000 KW when it is shut down; it can also be updated periodically in real time according to data, such as updating the energy demand data of a certain type of steel in the electric furnace according to the electricity data of the past 100 heats of a certain common carbon steel in the electric furnace. In addition, for equipment such as electric furnaces and refining furnaces that continuously switch heats, the production cycles of normal electric furnaces and refining furnaces also need to be considered to improve the prediction accuracy.

[0071] The actual electricity data of the reference period: At least the total electricity data of each main substation in the entire reference period and the electricity data of equipment related to steel types are required. The latter is mainly used to subtract the electricity data of these equipment from the total electricity data of the main substation to calculate the prediction data of other equipment. The bus-coupler switching record at the start of the reference period: The reason for requiring this data is that some equipment will switch between different main substations, so long-term prediction needs to be calculated in combination with different time periods and different bus-coupler states. Moreover, the bus-coupler states of the reference period and the prediction period may not be exactly the same, so it is necessary to compare the different bus-coupler states of the reference period and the prediction period for refined calculation.

[0072] Figure 2 In this, the content description of the long-term demand prediction module is as follows:

[0073] The predicted demand of the equipment related to the steel grade power consumption is calculated based on the start-up and shutdown times of these equipment in the production and maintenance plan and the steel grades produced. For example, if rolling line 1 rolls steel grade A from 9:00 to 10:00 and stops for roll change from 10:00 to 10:30, the predicted demand of rolling line 1 from 9:00 to 10:00 is 15000KW, and the predicted demand from 10:00 to 10:30 is 2000KW. The predicted demand of the equipment i related to the steel grade power consumption at the predicted time t is as follows:

[0074]

[0075] Where: and are the statistical energy demand values of equipment i during the production of steel grade A and during shutdown for maintenance respectively, and set1 is the set of equipment related to the steel grade power consumption.

[0076] For Figure 2 the predicted demand of other equipment, first calculate the reference data P 总需量_Base of the predicted demand of other equipment at the t-th moment according to the total demand data P in the actual power consumption data of the entire reference period and the power consumption data of each equipment related to the steel grade: Other_Base :

[0077]

[0078] Compare the switching conditions of other equipment in the reference period and the predicted period, and calculate the demand of equipment that runs in the reference period and does not run in the predicted period and the demand of equipment that does not run in the reference period and runs in the predicted period at the t-th moment respectively, so as to obtain the sum of the predicted demand of other equipment at the t-th moment:

[0079]

[0080] Where, are the statistical energy demand values when equipment j runs and does not run respectively, set2 (t) is the set of equipment that runs in the reference period and does not run in the predicted period at the t-th moment, and set3 (t) is the set of equipment that does not run in the reference period and runs in the predicted period at the t-th moment.

[0081] In addition, for some equipment, in addition to the difference between running and not running, there are also working conditions such as partial power operation, which also belong to the situation of working condition change. For example, for a certain steam generator, due to the decrease in the amount of gas in the predicted period, it needs to operate at a power reduced from the rated power of 30000KW to 16000KW, which also belongs to the change of the equipment operation state and needs to be considered in the calculation.

[0082] In this way, the total predicted demand value at the t-th moment of the predicted period can be obtained The predicted hourly demand for the last prediction period is the total predicted demand value which is the nth quantile of, where n can be selected as a certain number in [50%, 100%] according to the situation. The larger the value, the more conservative the prediction.

[0083]

[0084]

[0085] Finally, combine the bus-coupler switching records and calculate the predicted demands corresponding to each main substation and each moment according to the above steps, and then the predicted hourly demand values for each main substation prediction period can be obtained.

[0086] Triggering method of the long-term demand prediction module: After uploading the production scheduling and maintenance plan, the long-term prediction model is automatically triggered to calculate the hourly demand distribution for the future prediction period, such as 24 hours or one week. The prediction range depends on the time that can be determined in advance by the production plan and the maintenance plan. This long-term demand prediction module can achieve accurate prediction without all equipment data, and the reference period is dynamically updated, which can greatly improve the problem of inaccurate prediction caused by the slow changes of equipment power consumption due to seasons, climate, and working conditions. In addition, this method also fully considers the demand changes brought by periodic human intervention behaviors in some production processes. For example, some steam generators will slightly reduce power generation during the electricity price valley to reduce the overall cost. And in this method, the reference period and the prediction period correspond to each moment. The change of the period when the steam generator reduces power generation due to human intervention will be reflected in the demand part of the equipment without start-stop changes. The corresponding purchased energy demand for the prediction of this period will increase accordingly, thus indirectly considering the energy demand changes brought by this periodic temporary human intervention in the production process.

[0087] Production process anomaly monitoring module

[0088] This module first makes an abnormal judgment on the data quality of the production process concerned, and then detects the abnormal operation status of the key equipment in the production operation. When an abnormality is judged to occur, the regulation rule recommendation module is triggered to automatically provide the optimal regulation recommendation according to the type of abnormality for the on-site operators to refer to. It specifically includes the following sub-modules:

[0089] Data quality judgment module: Use statistical or rule-based methods to make a preliminary judgment on the data quality of the production process collected by sensors, and then use statistical methods or methods based on abnormal monitoring algorithms to judge whether the measured value deviates too much from the actual value, so as to avoid misjudging the abnormal state of the equipment due to data quality problems or mis-triggering operation instructions and affecting production.

[0090] Device anomaly detection module: When the data quality is normal, it can judge whether the device running status is abnormal through rules or machine learning algorithms.

[0091] Regulation rule suggestion module: When the device anomaly detection module determines that the device is abnormal and the data quality judgment module determines that the data quality is normal, it judges the types of device anomalies and triggers regulation suggestions to provide references for operators. The regulation rule suggestion module can include an abnormal judgment standard configuration unit to further identify the abnormal types output by the device anomaly detection module; it can also include a regulation rule suggestion library unit to give the optimal regulation suggestions for the corresponding abnormal situations after the abnormal types are judged.

[0092] For the dispatching decision-making suggestions, here is a method for establishing a power demand regulation rule library: First, sort out the electricity loads of all the company's devices under various working conditions; then, divide the priorities of device regulation, comprehensively consider whether the production process allows a certain device to downshift or stop, how much load is allowed to be reduced, how long it takes to reduce the load, and the impacts of downshifting or stopping on the production plan, maintenance plan, and energy plan. By quantifying these indicators and referring to the actual regulation experience of the business, corresponding dispatching combinations are formed and finally solidified into a regulation rule suggestion library. Of course, the subsequent regulation rule library can also be further optimized in combination with the subsequent actual operation conditions.

[0093] Among them, the data quality judgment module uses a statistics-based or rule-based method to preliminarily judge the data quality of the production process data collected by sensors, and then uses statistical or anomaly monitoring algorithms to realize the data quality judgment. The specific judgment process is as follows:

[0094] (1) Use a statistics-based data quality judgment or a rule-based data quality judgment method to preliminarily judge the data quality of the important parameters of the production process.

[0095] This module will first preliminarily judge the data quality of the important parameters of the production process. The data quality preliminary judgment method can be designed in combination with the data quality problems that are most likely to occur in the production data. The judgment methods here mainly include statistics-based data quality judgment and rule-based data quality judgment. For the statistics-based data quality preliminary judgment method, the most commonly used is variance judgment. The key is the selection of the number of data used for calculating the variance and the variance judgment threshold. For the rule-based data quality preliminary judgment method, it is mainly to design corresponding data quality judgment rules in combination with the actual situation, including data communication interruption, the numerical value remaining continuously constant, etc. For example, it can be set that if data cannot be normally collected within 1 minute, it is a database communication anomaly and the data quality has problems.

[0096] (2) Use a statistics-based method or an anomaly detection algorithm-based method to determine whether the measured value deviates too much from the actual value.

[0097] After the preliminary judgment of data quality, there may still be problems such as excessive deviation between the measured value and the actual value of the data and numerical drift. For the statistics-based method, for example, a T-test can be used. If the deviation or ratio between the detected data and the relevant data does not deviate significantly from the normal value, the data quality is considered normal; if the deviation is significant, the data quality is considered abnormal. The method based on the anomaly detection algorithm includes using historical data and machine learning algorithms to establish a machine learning algorithm model for key impact data points and predicted detection points, and then performing model prediction on real-time data. If the algorithm output result is abnormal, the data quality at the detection point is abnormal; if the algorithm output result is normal, there is no problem with the data quality at the detection point.

[0098] Using the above data quality judgment module can avoid misjudging the abnormal state of the equipment due to data quality problems or mis-triggering operation instructions, which affects production.

[0099] Overlimit attribution analysis module

[0100] This module mainly realizes the following functions: summarizing the demand trend chart and the number of overlimit times of each main substation of the previous day every day; the overlimit attribution analysis module can intelligently analyze the reasons for overlimit; when the analysis of the reasons for overlimit is incorrect, the actual reasons can be manually input to form an operation closed-loop; the system combines manual feedback to realize more intelligent subsequent analysis of the reasons for overlimit; this module also provides the number of overlimit times within the query time period, the feedback rate and the statistics of the number of feedbacks of the user on the root cause analysis results, and displays them according to different time dimensions. For example, through the KPI dashboard, it can be intuitively displayed and analyzed, and at the same time, human participation can be added, fully realizing the integration of machine and human, and then improving the accuracy of overlimit attribution.

[0101] As Figure 3 shown, the operation method of the overlimit attribution analysis module includes the following steps:

[0102] (1) After determining and obtaining the overlimit moment corresponding to the main substation, divide all equipment into a power generation equipment group, a rolling line group, a refining furnace group, an electric furnace group and other equipment groups; then calculate the change amount of the demand of each equipment group in a certain period of time in the past, such as 15 minutes, respectively.

[0103] (2) Since the influence weights of the demand change of each equipment group on the overlimit of the total demand are different, set the weight coefficients of the demand change amount of each equipment group to obtain the sorting sequence of the weighted demand change amounts of each equipment group;

[0104] (3) Select the equipment group with the largest weighted demand change as the main cause of overlimit, and finally further determine the specific equipment causing the overlimit according to the methods in the first two steps; that is, set the weight coefficient of the demand change of each equipment, and find the equipment with the largest weighted demand change under the equipment group, which is the main cause of the demand overlimit.

[0105] If the overlimit cause feedback manually is inconsistent with the overlimit cause determined by the algorithm, then update the influence weight of the equipment group or the equipment under the equipment group in combination with the manual feedback, so as to achieve more accurate overlimit cause analysis. Specifically as follows: If the equipment group determined for the overlimit cause is inconsistent, only update the influence weight of the equipment group, otherwise update the influence weight of the equipment under the equipment group. For example, for any two equipment groups, the weighted demand change of a certain equipment group is the largest, but the overlimit cause determined manually is another equipment group, then update the influence weight of the other equipment group. Similarly, the equipment weight under the equipment group can also be updated in a similar way.

[0106] Energy dashboard module

[0107] The energy dashboard is a comprehensive display interface that realizes the display of all key information and key associated information for energy demand management, mainly including the following information: the actual controlled demand, planned controlled demand and real-time load of each main substation in the current month; the real-time demand, predicted demand, change curves of real-time load in the past period of time for each main substation, as well as the demand plan control line, demand actual control line, etc.; pop-up alarm and sound reminder when the demand exceeds the limit or is about to exceed the limit; real-time monitoring of the electricity consumption load or demand of key users, and the user can be reminded of abnormalities of relevant equipment through color changes or other methods; real-time monitoring of the power generation power or demand of key power generation equipment, and the user can be reminded of abnormalities of relevant equipment through color changes or other methods; display of main power economic indicators, such as monthly total electricity bill, kWh electricity bill, basic electricity bill, cost of purchased coal, etc. This module can also automatically configure the content to be displayed or of interest, and the user can display the corresponding content according to their own needs.

[0108] Parameter configuration module

[0109] Realize the configuration of all parameters related to energy demand, mainly including functions such as setting the demand control target, configuring the demand economic indicators, configuring the electricity or demand threshold of power consumption and power generation equipment, inputting and importing the production plan function, and configuring the bus-tie switchover. The main functions are introduced in turn below:

[0110] Demand control target setting: This function is used to set the planned demand control target for each month. When the real-time predicted demand exceeds this target, an alarm will be triggered to remind the operator to take measures to reduce the demand. In addition to the above-mentioned planned demand control target, there is also an actual demand control target. The actual demand control target is the threshold for the real-time demand prediction alarm. When the real-time predicted demand exceeds this target, an alarm will be triggered. The actual demand control target is the larger value of the maximum demand that has occurred in the current month and the planned demand control target for the current month, which is more practical. The demand control target can be flexibly set and adjusted according to the production and maintenance plan and the actual situation.

[0111] Configuration of demand economic indicators: It includes the setting of electricity prices during peak, valley, and flat periods, and the setting of the prices of major energy sources such as purchased coal, which is convenient for the statistical calculation of electricity-related costs.

[0112] Configuration of the electricity consumption or demand threshold of power generation equipment: This parameter configuration module is mainly used for the real-time monitoring of the load or demand of key power generation equipment in the above-mentioned energy dashboard module. The threshold settings of this module can be flexibly set and adjusted according to the actual needs such as the equipment operation status and production rhythm.

[0113] Function of importing and entering production plans: This function is mainly used to import or enter the daily or weekly production and maintenance plans into the system for the long-term demand prediction function.

[0114] Configuration function of bus tie switching: The main equipment can be switched between different main substations, and two main substations supply power in a one-use-one-backup mode. When a certain main substation is under maintenance or a sudden power supply failure occurs in a certain main substation, the power supply can be switched to the power supply under another main substation to avoid equipment downtime affecting production or major losses caused by improper equipment shutdown. The configuration function of bus tie switching is to solve the problem of the impact of equipment switching under the main substation on the demand.

[0115] Abnormal event module

[0116] The abnormal event module realizes the alarm and recording of all abnormal events related to energy demand, mainly including the following: real-time demand exceeding the limit, real-time demand prediction exceeding the limit, data quality abnormality, key data instrument abnormality, blast furnace equipment abnormality, power generation equipment power generation abnormality, equipment abnormal events, etc.

[0117] Historical query module

[0118] The system supports historical energy demand queries and can view information such as real-time load, real-time demand, demand prediction, etc. for different time periods. At the same time, corresponding display components can be created according to needs. The historical query module mainly realizes the historical query of information such as historical demand, historical electricity load, historical power economic indicators, demand overlimit alarm records, demand prediction overlimit alarm records, regulation records, and regulation measures, which is convenient for quantitative tracking and post-event review.

[0119] Although the embodiments disclosed in this application are as above, the content described is only an embodiment adopted for the convenience of understanding this application and is not used to limit this application. Any person skilled in the art within the technical field to which this application pertains can make any modifications and changes in the form of implementation and details without departing from the spirit and scope disclosed in this application. However, the scope of patent protection of this application shall still be subject to the scope defined by the appended claims.

Claims

1. An intelligent management system for energy demand, characterized in that, it includes the following modules: A real-time demand prediction module, which establishes a real-time demand prediction model based on machine learning algorithms to achieve minute-level real-time demand prediction; A long-term demand prediction module, which realizes the energy demand prediction for the next week level according to the production plan, maintenance plan, bus-coupler switching status and the data of past electrical equipment; A production process anomaly monitoring module, which makes an anomaly judgment on the data quality concerned in the production process, and then detects the anomaly of the equipment operation status. When an anomaly is detected, it provides regulation suggestions; An overlimit cause analysis module, which automatically analyzes the cause of the energy demand overlimit and makes a manual feedback on whether the automatically analyzed overlimit cause is correct or not; An energy dashboard module, which can display the information on energy demand management; A parameter configuration module, which can configure the parameters related to energy demand; The real-time demand prediction module includes the following sub-modules: An equipment classification module, which divides the equipment types according to the characteristics of the equipment power consumption data, and classifies the equipment into impact large-load equipment and non-large-load equipment; A feature extraction module, which extracts features according to the equipment power consumption data, the key data affecting the equipment power consumption, and the key data representing the equipment state change respectively; A working condition identification module, which identifies different working conditions by identifying the bus-coupler state, establishes different models for different working conditions, and trains all the simulated working conditions respectively to obtain the prediction models for different working conditions; A model training module, which trains the prediction models for different working conditions by using machine learning algorithms for the data after feature extraction processing, optimizes and adjusts the parameters to obtain the prediction model for a preset future time; A real-time prediction module, which uses the data of a recent period of time in real time, constructs features through the feature extraction module, and selects the model trained by the model training module to achieve the prediction for the preset time; The real-time demand prediction module also includes a model online update module. When the prediction result of the real-time prediction module exceeds the deviation threshold, it uses the actual working condition data to retrain the model through the feature extraction module again; The long-term demand prediction module divides the energy demand of the prediction period into two parts: one part is the predicted demand of the equipment related to the steel grade power consumption, and this part of the demand is calculated through the historical data and production plan of each steel grade; the other part is the predicted demand of the other equipment except the equipment related to the steel grade power consumption, which is obtained by calculating the demand change of the equipment with inconsistent operation status between the reference period and the prediction period and the demand of the equipment without start-stop change; The production process anomaly monitoring module includes the following sub-modules: A data quality judgment module, which makes a preliminary judgment on the data quality of the production process collected by the sensor by using a statistical or rule-based method, and then uses a statistical method or a method based on an anomaly monitoring algorithm to judge whether the measured value deviates too much from the actual value; An equipment anomaly detection module, which judges whether the equipment operation status is abnormal by using rules or machine learning algorithms when the data quality is normal; The regulation rule suggestion module judges the types of equipment anomalies when the equipment anomaly detection module determines that the equipment has an anomaly and the data quality judgment module determines that the data quality is normal, and triggers regulation suggestions to provide reference for operators; The regulation rule suggestion module includes an anomaly judgment standard configuration unit and a regulation rule suggestion library unit; Among them, the method for establishing the regulation rule suggestion library includes the following steps: First, sort out the energy requirements of all equipment under various working conditions; then, divide the equipment regulation into a priority order, comprehensively consider the production process and the energy requirement plan of each equipment, form corresponding scheduling combinations by quantifying indicators and referring to actual regulation experience, and finally solidify them into the regulation rule suggestion library.

2. The intelligent management system according to claim 1, wherein, The predicted demand of the equipment related to steel grade power consumption is calculated based on the start-up and shutdown times of these equipment and the steel grades produced in the production and maintenance plan. The predicted demand of the equipment i related to steel grade power consumption at the predicted time t is: Wherein: and are respectively the statistical energy demand values of equipment i during the production of steel grade A and during shutdown maintenance, and set1 is the set of equipment related to the electricity consumption of steel grades.

3. The intelligent management system according to claim 2, wherein, For the predicted demand of other equipment, first, based on the total demand data P in the actual power consumption data for the entire reference period 总需量_Base and the power consumption data of equipment related to the power consumption of each steel grade calculate the reference data P Other_Base for the predicted demand of other equipment at the t-th moment: Compare the switch conditions of other equipment in the reference period and the predicted period, and calculate the demand of the equipment that is running in the reference period and not running in the predicted period and the demand of the equipment that is not running in the reference period and running in the predicted period at the t-th moment respectively, so as to obtain the sum of the predicted demands of other equipment at the t-th moment: Among them, are the statistical energy demand values when device j is operating and not operating, set2 (t) is the set of devices that operate in the reference period and do not operate in the prediction period at time t, set3 (t) is the set of devices that do not operate in the reference period and operate in the prediction period at time t; Obtain the total demand prediction value at the t-th moment of the prediction period 4. The intelligent management system according to claim 1, wherein, The data quality judgment module uses a statistics-based or rule-based method to preliminarily judge the quality of the production process data collected by the sensor, and then uses a statistics or anomaly monitoring algorithm to realize the quality judgment of the data.

5. The intelligent management system according to any one of claims 1-3, wherein, The operation method of the overlimit attribution analysis module includes the following steps: (1) After judging and obtaining the overlimit moment corresponding to the main substation, divide all equipment into multiple equipment groups, and then calculate the demand change amount of each equipment group in a certain period of the past respectively; (2) Set the weight coefficient of the demand change amount of each equipment group to obtain the sorting sequence of the weighted demand change amounts of each equipment group, and select the equipment group with the largest weighted demand change amount as the main cause of the overlimit; (3) Set the weight coefficient of the demand change amount of each equipment under the equipment group with the largest weighted demand change amount, and find out the equipment with the largest weighted demand change amount under the equipment group as the main cause of the demand overlimit.

6. The intelligent management system according to any one of claims 1-3, wherein, The intelligent management system further includes a parameter configuration module, which can perform functions such as setting the demand control target, configuring the demand economic index, configuring the power or demand threshold of the power consumption and power generation equipment, inputting and importing the production plan, and configuring the bus tie switch.

7. The intelligent management system according to claim 6, wherein, The intelligent management system further includes an abnormal event module, which can alarm and record all abnormal events related to the energy demand.

Images