A method and system for demand coordinated control under abnormal working conditions of cement plant load

Abstract

The application discloses a kind of cement plant load abnormal condition under demand collaborative control method and system, belong to power system demand control and electric energy management technical field, method includes: acquisition operation data and calculate plant equivalent production load;Data is preprocessed and sliced, and extract load event fragment;The working condition category and confidence of current load event are identified by CNN-LSTM model;Transformer available demand margin is calculated in combination with working condition category and confidence;According to working condition category, confidence and available demand margin, generate energy storage charging power control instruction and execute;After abnormality is removed, restore charging power according to climbing rate.This application can identify load abnormality in advance under the complex working condition of cement plant and actively correct energy storage charging power, avoid transformer demand overrun, improve the accuracy and initiative of demand control.

Description

Technical Field

[0001] This invention relates to the field of power system demand control and power management technology, and in particular to a method and system for coordinated demand control under abnormal load conditions in a cement plant. Background Technology

[0002] With the development of automated production and integrated energy systems in cement plants, a large number of high-power electromechanical equipment, such as raw material mills, cement mills, coal mills, vertical mills, roller presses, fans, belt conveyors, bucket elevators, air compressors, and rotary kiln main drives, are typically present simultaneously within the plant area. During startup, shutdown, loading, unloading, or abnormal operation of these devices, significant changes occur in parameters such as current, voltage, active power, reactive power, and power factor in the plant's power distribution system. The load change process differs between different equipment or under different operating conditions, exhibiting variations in load waveforms in terms of peak starting current, number of peaks, inter-peak time intervals, power surge amplitude, load change rate, duration, and stabilization process. For example, during the startup of a mill, there may be two distinct starting current peaks with a certain time interval between them; during the startup of a blower, there is usually one distinct starting current peak, followed by a gradual stabilization of current and power; when heavy-duty equipment starts with material, the starting current peak may be large, last for a long time, and the power may drop slowly; when multiple devices start continuously, multiple current peaks or power peaks may appear consecutively in a short period of time, accompanied by a step-like increase in load; when conveyor equipment is blocked or jammed, the current may gradually increase and remain at a high level; when equipment trips or is interlocked and shuts down, the power may suddenly drop, accompanied by a sequential change in the load of multiple devices.

[0003] For energy-intensive enterprises such as cement plants, transformer demand management is a crucial aspect of plant energy management. If the plant's production load increases significantly within a short period, and the demand control system cannot identify the specific cause and type of operating condition, control lag, misjudgment, or over-control can easily occur. For example, when the system only detects an increase in transformer power, it cannot determine whether this increase is caused by normal load fluctuations, mill start-up, fan start-up, heavy load start-up, continuous start-up of multiple devices, material blockage in conveyor equipment, or output fluctuations from distributed power sources such as photovoltaic or waste heat power generation. Without accurate identification of different abnormal load conditions, the demand control strategy cannot be adjusted in a targeted manner according to the actual production conditions.

[0004] Existing methods for load monitoring and demand control in cement plants largely rely on manual experience, operator observation, equipment protection signals, or simple threshold judgments. For example, when the current or active power exceeds a set threshold, the system determines that the equipment may be in an abnormal state; when the power on the transformer side increases significantly in a short period of time, the system determines that the plant load has surged; when the transformer power approaches the demand limit, the system limits the energy storage charging power or some adjustable loads, or issues a demand over-limit alarm. Summary of the Invention

[0005] The purpose of this invention is to provide a demand coordination control method and system for cement plants under abnormal load conditions. The aim is to use a CNN-LSTM model to identify typical abnormal load conditions in cement plants in real time, such as mill start-up, heavy load start-up, material blockage, and shutdown. The system will combine the identification results with the demand margin to actively adjust the energy storage charging power, thereby improving the initiative and accuracy of demand control, reducing the risk of transformer demand exceeding limits, and reducing false alarms, missed alarms, and over-control.

[0006] To achieve the above objectives, in a first aspect, the present invention provides a method for demand coordination control under abnormal load conditions in a cement plant, comprising the following steps: S1. Collect operating data of the power supply and distribution system and production equipment of the cement plant. The operating data shall include at least three-phase current, three-phase voltage, active power, reactive power, power factor, real-time power on the transformer side, energy storage charging and discharging power, photovoltaic power, waste heat power generation power and operating status signals of production equipment. S2. Calculate the equivalent production load of the plant area based on the collected operating data, and preprocess the operating data to form multi-channel time series data; S3. Use a sliding time window or event-triggered method to slice multi-channel time series data and extract load event segments; S4. Input the load event fragment into the pre-trained load anomaly identification model to identify the current load event's condition category and corresponding identification confidence level. The condition categories include at least normal load fluctuation, mill start-up, fan start-up, heavy load start-up, continuous start-up of multiple devices, material blockage in conveying equipment, and equipment shutdown. S5. Calculate the current real-time demand of the transformer based on the real-time power of the transformer side, and calculate the available demand margin of the transformer within the predicted time range based on the transformer demand limit, the planned charging power of the energy storage, the operating condition category, and the identification confidence level; wherein, the planned charging power of the energy storage is the preset charging power. S6. Generate energy storage charging power control commands based on the operating condition category, identification confidence level, and transformer available demand margin. S7. Execute the energy storage charging power control command. After the abnormal load condition is resolved and the transformer's available demand margin is restored, gradually restore the energy storage charging power to the planned energy storage charging power according to the set ramp rate.

[0007] Preferably, the formula for calculating the equivalent production load of the plant area in S2 is: ; in, Indicates the sampling time. This represents the real-time power on the transformer side. For energy storage charging and discharging power, For photovoltaic power, It generates electricity from waste heat, and during energy storage charging... During energy storage and discharge .

[0008] Preferably, the preprocessing in S2 includes time synchronization, missing value completion, outlier correction, filtering and denoising, and normalization of the running data; The formula for normalization is: ; in, This is the original data. This is the historical average. For historical standard deviation, This is the normalized data.

[0009] Preferably, the event triggering method in S3 is as follows: when the rate of change of the equivalent production load or the rate of change of the current exceeds a set threshold, an event slice is triggered, and the triggering condition is: ; or ; in, Indicates the sampling time. For the equivalent production load of the plant area, This refers to the effective value of the three-phase current or the combined value of the three-phase current. The power change rate threshold, The threshold for the rate of change of current. The sampling interval is denoted as .

[0010] Preferably, the abnormal load condition identification model in S4 is a CNN-LSTM model, including an input layer, a convolutional feature extraction layer, a pooling layer, an LSTM temporal feature learning layer, a fully connected layer, and a classification output layer; the classification output layer uses the Softmax function to output the probability of each type of load condition. ; in, The load event belongs to the first The probability of such working conditions For the first The classification score corresponding to each type of working condition This represents the total number of operating condition categories. Index for operating condition categories; Operating conditions and identification confidence They are respectively: ; .

[0011] Preferably, the transformer described in S5 has a demand margin. The calculation method is as follows: The predicted equivalent production load is adjusted based on the operating condition category and identification confidence level, with the adjustment amount being... When the operating condition is normal load fluctuation, Set to zero; when the operating condition is heavy load start, continuous start of multiple devices, or material blockage in the conveyor equipment. Take a positive value; when the operating condition is equipment trip, Take negative values, where, Indicates the sampling time. The sampling interval is... Indicates the prediction step size index; The predicted transformer-side power is calculated by combining the predicted photovoltaic power and the predicted waste heat power, and the maximum predicted demand within the predicted time range is calculated based on the predicted transformer-side power. The available demand margin for transformers is: ; in, This is the transformer demand limit. For safety margin.

[0012] Preferably, the method for generating the energy storage charging power control command in S6 is as follows: S61. When no abnormal load conditions are identified and the transformer has available demand margin At that time, the energy storage charging power control command is the planned charging power of the energy storage. ; S62. When abnormal load conditions are identified and the transformer has available demand margin At that time, the energy storage charging power control command is the planned charging power of the energy storage. It also records the current operating condition category, the time of occurrence, and the identification confidence level; S63. When abnormal load conditions are identified and the transformer has available demand margin Time: Calculate the allowable charging power under demand constraints based on the transformer's available demand margin. And set the working condition correction factor according to the working condition category. Operating condition correction factor The range of values ​​is Furthermore, the working condition correction factor decreases as the severity of the working condition increases; S64, Energy Storage Plan Charging Power Allowable charging power under demand constraints The product of the operating condition correction factor and the planned charging power of the energy storage The minimum value of the three is used as the initial charging power command. : ; S65, Based on the upper limit of the energy storage state of charge. Current state of charge of energy storage Rated energy storage capacity and charging efficiency Calculate the maximum charging power under state-of-charge constraints: ; S66, Obtain Preliminary Charging Power Command Maximum charging power under state of charge constraints The smaller value in the middle is used as the final energy storage charging power control command. : ; S67. Permissible charging power under demand constraints Or operating condition correction factor At that time, the final energy storage charging power control command is Temporarily suspend energy storage charging; S68. When the abnormal load condition is resolved and the transformer's available demand margin is restored, the final energy storage charging power control command will gradually restore the planned energy storage charging power according to the set ramp rate. The restoration method is as follows: ; in, To restore the ramp rate of energy storage charging power.

[0013] Preferably, the convolutional feature extraction layer contains multiple convolutional kernels of different sizes, used to extract waveform features of load event segments at different time scales. The waveform features include at least the peak value of the starting current, the number of peak values, the time interval between peaks, the power change amplitude, and the load change rate. The pooling layer employs max pooling to preserve the peak startup current and power surge amplitude characteristics. The local feature sequences extracted by the convolutional feature extraction layer and the pooling layer are input into the LSTM temporal feature learning layer. The LSTM temporal feature learning layer is used to identify the change pattern of the equipment from before startup, startup process, operation transition to stable state or abnormal state according to the order of local feature sequences in load event segments.

[0014] Secondly, the present invention also provides a demand coordination control system for cement plants under abnormal load conditions, comprising: The operation data acquisition unit is used to collect operating data of the cement plant's power supply and distribution system and production equipment; The load anomaly identifier is connected to the operation data acquisition unit. It is used to calculate the equivalent production load of the plant area based on the operation data, preprocess the operation data and extract load event fragments, and identify the operation category and corresponding identification confidence level of the current load event through a pre-trained load anomaly operation condition identification model. The demand margin estimator is connected to the operation data acquisition unit and the load anomaly identifier. It is used to calculate the real-time demand of the transformer based on the real-time power of the transformer side, and to calculate the available demand margin of the transformer within the predicted time range based on the transformer demand limit, the energy storage plan charging power, the operating condition category and the identification confidence level. The load anomaly controller, connected to the demand margin evaluator and load anomaly identifier, is used to generate energy storage charging power control commands based on the operating condition category, identification confidence level, and available demand margin of the transformer. The control switching unit, connected to the load anomaly controller, is used to select and execute either the planned energy storage charging power or the energy storage charging power control command generated by the load anomaly controller, based on the identification results of the load anomaly identifier and the evaluation results of the demand margin estimator.

[0015] Preferably, the load anomaly identifier includes a data preprocessing module, an event slicing module, and a CNN-LSTM recognition module; After the load anomaly controller outputs a control command to suspend energy storage charging, the control switching unit controls the energy storage charging power to gradually recover to the planned energy storage charging power according to the set ramp rate when the transformer's available demand margin is restored.

[0016] Therefore, compared with the prior art, the present invention, which adopts the above-mentioned demand coordination control method and system under abnormal load conditions in cement plants, has the following beneficial effects: (1) By collecting multi-source time-series data and using the CNN-LSTM model to extract waveform features such as peak current, peak frequency, inter-peak time interval, power fluctuation amplitude, and load change rate, this invention can accurately identify typical working conditions such as mill start-up, fan start-up, heavy load start-up, continuous start-up of multiple devices, material blockage of conveying equipment, and equipment shutdown, thus overcoming the shortcomings of existing technologies that can only determine "whether the power increases" but cannot identify the specific working condition type.

[0017] (2) The present invention introduces the results of abnormal load conditions into demand control. When an abnormal condition that may lead to a rapid increase in demand is identified, the energy storage charging power is adjusted in advance in combination with the demand margin, so that demand control changes from the traditional "passive response" to "active coordination", which significantly improves the accuracy and real-time performance of control and effectively reduces the risk of transformer demand exceeding the limit.

[0018] (3) This invention makes full use of the temporal characteristics of load changes in cement plants and uses a CNN-LSTM model to extract local features of load waveforms and temporal dependencies. Compared with manual experience, fixed thresholds or simple rate of change judgment, it is more suitable for identifying load anomalies under complex production conditions in cement plants and has stronger adaptive ability and generalization performance.

[0019] (4) While actively avoiding the risk of demand exceeding the limit, this invention avoids the problems of false alarms, missed alarms and excessive restrictions caused by fixed threshold control, takes into account the safety of power supply and distribution system and the economic efficiency of energy storage utilization, and provides reliable technical support for production operation monitoring, equipment status judgment and abnormal alarm.

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a method for coordinated demand control under abnormal load conditions in a cement plant, according to the present invention. Figure 2 This is a schematic diagram of the architecture of a demand coordination control system for cement plants under abnormal load conditions according to the present invention. Detailed Implementation

[0022] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example like Figures 1-2 As shown, the present invention provides a demand-balanced control system for cement plants under abnormal load conditions, including an operating data acquisition unit, a load anomaly identifier, a demand margin evaluator, a load anomaly controller, and a control switching unit.

[0025] The operation data acquisition unit is used to collect operation data of the cement plant's power supply and distribution system and production equipment.

[0026] The load anomaly identifier is connected to the operation data acquisition unit and is used to calculate the equivalent production load of the plant area based on the operation data, preprocess the operation data and extract load event fragments, and identify the operation category and corresponding identification confidence level of the current load event through a pre-trained load anomaly condition identification model.

[0027] The demand margin estimator is connected to the operation data acquisition unit and the load anomaly identifier. It is used to calculate the real-time demand of the transformer based on the real-time power of the transformer side, and to calculate the available demand margin of the transformer within the predicted time range based on the transformer demand limit, the energy storage plan charging power, the operating condition category and the identification confidence level.

[0028] The load anomaly controller is connected to the demand margin evaluator and the load anomaly identifier to generate energy storage charging power control commands based on the operating condition category, identification confidence level, and the available demand margin of the transformer.

[0029] The control switching unit is connected to the load anomaly controller and is used to select to execute the energy storage planned charging power or the energy storage charging power control command generated by the load anomaly controller based on the identification results of the load anomaly identifier and the evaluation results of the demand margin estimator.

[0030] The load anomaly detector includes a data preprocessing module, an event slicing module, and a CNN-LSTM recognition module.

[0031] The present invention will now be described in further detail with reference to the method steps.

[0032] S1. Collect operational data Collect operational data of the power supply and distribution system and production equipment of the cement plant. The operational data shall include at least three-phase current, three-phase voltage, active power, reactive power, power factor, real-time power on the transformer side, energy storage charging and discharging power, photovoltaic power, waste heat power generation power, and operating status signals of production equipment.

[0033] The system collects the above data according to the set sampling period. The sampling time is... The sampling interval is The data collected specifically includes: , , Three-phase current; , , Three-phase voltage; Active power; Reactive power; Power factor; Real-time power on the transformer side; Real-time demand of transformers; Energy storage charging and discharging power; Energy storage state of charge; Photovoltaic power; Waste heat power generation capacity; : Main production equipment operating status signals.

[0034] S2. Calculation of Equivalent Production Load and Data Preprocessing (a) Calculate the equivalent production load of the plant area The equivalent production load of the plant area is calculated based on the collected operational data. The formula for calculating the equivalent production load of the plant area in S2 is: ; in, Indicates the sampling time. This represents the real-time power on the transformer side. For energy storage charging and discharging power, For photovoltaic power, It generates electricity from waste heat, and during energy storage charging... During energy storage and discharge .

[0035] Real-time power on transformer side In reality, it is the equivalent production load of the factory area. Subtract photovoltaic power and waste heat power generation In addition, energy storage charging and discharging power The result (with positive power absorption from the grid during charging) is rearranged to obtain the above formula. This formula can isolate the impact of photovoltaic, waste heat power generation, and energy storage charging and discharging on the transformer-side power, obtaining load change data that more closely approximates actual production conditions.

[0036] (ii) Data preprocessing The running data is preprocessed to form multi-channel time-series data. Preprocessing in S2 includes time synchronization, missing value completion, outlier correction, filtering and denoising, and normalization of the running data.

[0037] Time synchronization: Time synchronization of data from smart meters, PLCs, frequency converters, energy storage systems, photovoltaic systems, and waste heat power generation systems, so that all types of data have a unified timestamp.

[0038] Missing value imputation: This involves processing missing values ​​and communication anomalies in the original data. For short-term missing data, interpolation between adjacent time points or moving averages can be used for imputation.

[0039] Outlier correction: This involves processing obvious errors in the raw data. For outliers that exceed the equipment's range or clearly do not conform to physical laws, amplitude limiting correction or median filtering methods can be used.

[0040] Filtering and noise reduction: Time-series data such as current, voltage, active power, reactive power, power factor, and equivalent production load are filtered to reduce the impact of sampling noise on the identification results. Filtering methods can include moving average filtering, median filtering, or low-pass filtering, but key waveform characteristics during equipment startup, load surges, and abnormal changes should be preserved.

[0041] Normalization: The formula for normalization is as follows: ; in, This is the original data. This is the historical average. For historical standard deviation, This is the normalized data.

[0042] The purpose of normalization is to enable different types of data, such as current, voltage, power, power factor, and device status, to be used as a unified multi-channel time-series input, so as to avoid some features being submerged during model training due to differences in the dimensions of different physical quantities. and The mean and standard deviation of the data channel over a longer historical period can be used, or the mean and standard deviation of the entire training set can be used.

[0043] S3, Load Event Slicing Multi-channel time-series data is sliced ​​using a sliding time window or event-triggered method to extract load event segments.

[0044] In the sliding time window method, the window length is Window step size is Then at the current moment The corresponding load event segment is: ; in, This contains the multi-channel load time series data within the current window.

[0045] In S3, the event triggering method is as follows: an event slice is triggered when the rate of change of the equivalent production load or the rate of change of the current exceeds a set threshold. The triggering condition is: ; or ; in, Indicates the sampling time. For the equivalent production load of the plant area, This refers to the effective value of the three-phase current or the combined value of the three-phase current. The power change rate threshold, The threshold for the rate of change of current. The sampling interval is denoted as .

[0046] The combined three-phase current can be expressed as the root mean square value of the three-phase current, the vector sum of the three-phase current, or other forms of combined value. Power change rate threshold. and current change rate threshold It can be set according to the actual operating conditions of the cement plant.

[0047] When the above triggering conditions are met, the system determines that a load event may occur. Centered on the triggering time, it extracts data within a certain time range before, during, and after startup to form a complete load event segment. Through this method, load event samples encompassing processes such as equipment startup, heavy-load startup, sudden load changes, material blockage development, and shutdown changes can be obtained.

[0048] S4. Abnormal Load Condition Identification Input load event fragments into a pre-trained load anomaly identification model to identify the current load event's condition category and corresponding identification confidence level. The condition categories include at least normal load fluctuations, mill start-up, fan start-up, heavy load start-up, continuous start-up of multiple devices, material blockage in conveying equipment, and equipment shutdown.

[0049] The load anomaly identification model in S4 is a CNN-LSTM model, which includes an input layer, a convolutional feature extraction layer, a pooling layer, an LSTM temporal feature learning layer, a fully connected layer, and a classification output layer.

[0050] For each load event segment, the system constructs a multi-channel input matrix from various electrical quantities in chronological order: ; in, The number of sampling points within the time window. This refers to the number of input channels. Input channels can include three-phase current, three-phase voltage, active power, reactive power, power factor, equivalent production load, transformer-side power, energy storage power, photovoltaic power, waste heat power generation, and equipment operating status, etc.

[0051] Convolutional Feature Extraction Layer: This layer contains multiple convolutional kernels of varying sizes to extract waveform features of load event segments at different time scales. These waveform features include at least the peak initiation current, the number of peaks, the time interval between peaks, the magnitude of power fluctuations, and the load change rate. The output of the convolutional layer can be represented as: ; in, For convolution kernel weights, For bias terms, This represents the convolution operation. This represents the activation function. This represents the output features of the convolutional layer.

[0052] Pooling Layer: The pooling layer employs max pooling to preserve the peak startup current and power surge amplitude characteristics. Average pooling can also be used, with max pooling being suitable for highlighting significant features such as peak startup current, peak power, and local surges.

[0053] The LSTM temporal feature learning layer: Local feature sequences extracted by the convolutional feature extraction layer and pooling layer are input into the LSTM temporal feature learning layer. This layer identifies the changing patterns of equipment from pre-startup, startup, and transition to a stable or abnormal state based on the chronological order of these local feature sequences within load event segments. LSTM can identify continuous changes that are difficult to express with a single-moment threshold, such as the slow power drop after heavy-load startup, the continuous current increase caused by material blockage in conveyor equipment, the sudden power drop after equipment shutdown, and subsequent interlocking changes.

[0054] The classification output layer uses the Softmax function to output the probabilities of each working condition: ; in, The load event belongs to the first The probability of such working conditions For the first The classification score corresponding to each type of working condition This represents the total number of operating condition categories. Index for operating condition categories; Operating conditions and identification confidence They are respectively: ; .

[0055] Indicates taking The largest The value serves as the output condition category; Indicates taking all The maximum value in the range is used as the identification confidence level.

[0056] when Greater than the set confidence threshold ,and When the load fluctuation does not fall under the category of normal load fluctuation, the system outputs a load anomaly flag. ;when Not greater than the set confidence threshold ,or When the load fluctuates within the normal range, the system output is... .in, This indicates that an abnormal load condition has been detected. This indicates that no abnormal load conditions were detected.

[0057] S5. Calculate the available demand margin for the transformer. The real-time demand of the transformer is calculated based on the real-time power on the transformer side. Based on the transformer demand limit, the planned charging power of the energy storage, the operating condition category, and the identification confidence level, the available demand margin of the transformer within the predicted time range is calculated. The planned charging power of the energy storage is the preset charging power.

[0058] (a) Calculate the real-time demand of the transformer The real-time demand of a transformer can be expressed as the average power over a certain time window: ; in, The number of sampling points within the demand calculation window. This indicates the real-time demand of the transformer.

[0059] (ii) Correction of predicted equivalent production load Available demand margin for transformers in S5 The calculation method is as follows: The predicted equivalent production load is adjusted based on the operating condition category and identification confidence level, with the adjustment amount being... When the operating condition is normal load fluctuation, Take zero; When the operating condition is heavy load start, continuous start of multiple devices, or material blockage in the conveyor equipment... Take positive values; When the operating condition is equipment tripped Take negative values, where, Indicates the sampling time. The sampling interval is... This indicates the prediction step size index.

[0060] The predicted equivalent production load can be expressed as: ; in, It can be determined based on historical samples of similar operating conditions, equipment startup characteristics, load change rate, or load prediction model.

[0061] (III) Calculate and predict the transformer-side power The predicted transformer-side power is calculated by combining the predicted photovoltaic power and the predicted waste heat power. When the energy storage system continues charging at the planned power, the predicted time is set to [time value missing]. Then the predicted transformer-side power can be expressed as: ; in, To predict transformer-side power, To predict equivalent production load, Charging power for energy storage programs, To predict photovoltaic power, To predict waste heat power generation, if photovoltaic power and waste heat power generation prediction results are unavailable in the short term, the current photovoltaic power and waste heat power generation can be used as approximations.

[0062] (iv) Calculate the average demand for the forecast window To avoid using historical power data that has already occurred as predicted values, the predicted demand is calculated using the method of "historical measured power + future predicted power". Regarding the prediction step size... Forecast window average demand It can be represented as: ; in, The measured transformer-side power that has already occurred within the demand window (total) (one measured point) Forecast transformer-side power at future times (total) (predicted points) This represents the total number of sampling points within the demand calculation window.

[0063] (v) Calculate the maximum predicted demand and the available demand margin. Based on the average demand within the forecast window, calculate the maximum forecast demand within the forecast time range. ; in, To predict the time range, express Predicting transformer demand at any given time.

[0064] The available demand margin for the transformer is: ; in, This is the transformer demand limit. For safety margin.

[0065] Safety margin This is a power margin reserved to prevent unexpected over-limit demand due to forecasting errors; its value can be set according to transformer capacity and system reliability requirements. When, it indicates that under the current operating conditions and within the predicted time frame, the energy storage system still has a demand margin when charging according to the planned power; when When this occurs, it indicates that the energy storage system continuing to charge at the planned power may cause the transformer demand to exceed the limit, requiring the entry into the demand control correction stage based on the identification results.

[0066] S6. Generate energy storage charging power control command Based on the operating condition category, identification confidence level, and transformer available demand margin, energy storage charging power control commands are generated. The method for generating energy storage charging power control commands in S6 is as follows: S61. When no abnormal load conditions are identified and the transformer has available demand margin At that time, the energy storage charging power control command is the planned charging power of the energy storage. ; S62. When abnormal load conditions are identified and the transformer has available demand margin At that time, the energy storage charging power control command is the planned charging power of the energy storage. It also records the current operating condition category, the time of occurrence, and the identification confidence level; S63. When abnormal load conditions are identified and the transformer has available demand margin Time: Calculate the allowable charging power under demand constraints based on the transformer's available demand margin. And set the working condition correction factor according to the working condition category. Operating condition correction factor The range of values ​​is Furthermore, the working condition correction factor decreases as the severity of the working condition increases; For normal load fluctuation conditions A value close to 1 is acceptable, indicating that there is no need to restrict charging excessively. However, for conditions such as heavy-load startup, continuous startup of multiple devices, and material blockage in conveyor equipment, which may lead to a further increase in load, A smaller value can be chosen; for severe abnormal operating conditions (such as equipment shutdown causing a sudden drop in load) or operating conditions where it is necessary to immediately suppress the increase in demand, 0 is acceptable.

[0067] S64, Energy Storage Plan Charging Power Allowable charging power under demand constraints The product of the operating condition correction factor and the planned charging power of the energy storage The minimum value of the three is used as the initial charging power command. : ; S65, Based on the upper limit of the energy storage state of charge. Current state of charge of energy storage Rated energy storage capacity and charging efficiency Calculate the maximum charging power under state-of-charge constraints: ; S66, Obtain Preliminary Charging Power Command Maximum charging power under state of charge constraints The smaller value in the middle is used as the final energy storage charging power control command. : ; S67. Permissible charging power under demand constraints Or operating condition correction factor At that time, the final energy storage charging power control command is Temporarily suspend energy storage charging; S68. When the abnormal load condition is resolved and the transformer's available demand margin is restored, the final energy storage charging power control command will gradually restore the planned energy storage charging power according to the set ramp rate. The restoration method is as follows: ; in, To restore the ramp rate of energy storage charging power.

[0068] Restore climbing rate The setting is to avoid the impact of sudden changes in energy storage charging power on the power grid, and its value can be set according to the performance of the energy storage system and the requirements of the power grid. After the load anomaly controller outputs a control command to suspend energy storage charging, the control switching unit controls the energy storage charging power to gradually restore to the planned energy storage charging power according to the set ramp rate when the available demand margin of the transformer is restored.

[0069] S7. Execute control commands and resume. Execute the energy storage charging power control command. When the abnormal load condition is resolved and the transformer's available demand margin is restored, gradually restore the energy storage charging power to the planned energy storage charging power according to the set ramp rate.

[0070] Specifically, when no abnormal load conditions are detected and the transformer demand margin meets requirements, the energy storage system charges according to the original planned power. When an abnormal load condition is detected, but the predicted transformer demand still does not reach the demand limit, the system can maintain the original planned charging while recording the current condition type, occurrence time, and identification confidence level. When an abnormal load condition is detected and the predicted transformer demand has a risk of exceeding the limit, the system adjusts the energy storage charging power according to the type of abnormal load condition and the demand margin, and performs reduced-power charging. When a severe abnormal condition is detected, or there is currently no available demand margin, the system suspends energy storage charging and outputs an abnormality alarm message. When the abnormal load condition is resolved and the transformer demand margin returns to normal, the energy storage system gradually restores to the original planned charging power according to the set ramp rate to avoid sudden changes in charging power.

[0071] In the above method steps, the CNN-LSTM model used in S4 is pre-trained in the following way.

[0072] During the model training phase, the system extracts load event segments under different operating conditions from the cement plant's historical operating data and labels them according to equipment operation logs, PLC start / stop records, manual confirmation results, or historical alarm records. Training samples include categories such as normal load fluctuations, mill start-up, fan start-up, heavy load start-up, continuous start-up of multiple devices, material blockage in conveyor equipment, and equipment shutdown. The labeled samples are divided into training, validation, and test sets.

[0073] The CNN-LSTM model uses the cross-entropy loss function during training: ; in, Label the actual working conditions (when the sample belongs to the first...) Class Time ,otherwise , For the model to predict belonging to the first The probability of such working conditions This represents the total number of operating condition categories.

[0074] By updating the parameters of the CNN convolutional layer, LSTM layer, and fully connected layer through the backpropagation algorithm, the model can automatically learn the waveform characteristics and temporal variation patterns corresponding to different abnormal load conditions.

[0075] In actual operation, the system can periodically add newly identified working condition samples, manual confirmation results, and abnormal alarm records to the sample library, and perform incremental training or periodic updates to the model, thereby improving the model's adaptability to changes in the actual production conditions of cement plants.

[0076] Application scenario examples The following example illustrates the complete control flow of the present invention to facilitate understanding.

[0077] Scenario setting: A cement plant is equipped with a main transformer, and the demand limit is... Safety margin Rated capacity of energy storage system Current state of charge Upper limit of state of charge Charging efficiency Planned charging power Current real-time demand for transformers Photovoltaic power Waste heat power generation The energy storage is currently charging. ).

[0078] S1 (Acquisition of Operational Data): The operational data acquisition unit acquires three-phase current, three-phase voltage, active power, and real-time power on the transformer side at a sampling period of 1 second. Energy storage charging and discharging power Photovoltaic power Waste heat power generation Operational data, such as operating status signals of production equipment.

[0079] S2 (Calculation of Equivalent Production Load and Pre-processing): The load anomaly detector uses the formula... Calculate the equivalent production load of the plant area.

[0080] Substitute the data: The collected data is processed through time synchronization, missing value completion, filtering and denoising, and normalization to form multi-channel time-series data.

[0081] S3 (Load Event Slice): Suppose that at a certain moment, a cement mill starts up. The load anomaly detector detects the rate of change of the equivalent production load. Exceeded the power change rate threshold The system triggers event slices. Centered on the trigger moment, the system extracts data from 5 seconds before startup to 30 seconds after startup to form a complete load event slice.

[0082] S4 (Abnormal Load Condition Identification): The above load event segments are input into a pre-trained CNN-LSTM model. The model extracts waveform features such as peak current, peak frequency, and inter-peak time interval through convolutional layers, learns the time-series changes from before to after startup through LSTM layers, and finally outputs the probabilities of various operating conditions through a softmax layer. Assume the model outputs the probability of mill start-up. Normal fluctuation probability If the combined probability of other operating conditions is 0.03, then the operating condition category is "mill operation". The confidence level is... .because (Assuming) )and This is not a normal load fluctuation; the system output load is abnormal. .

[0083] S5 (Calculate Transformer Available Demand Margin): The demand margin estimator corrects the predicted equivalent production load based on the identification results. Since a "mill start-up" condition was identified, based on historical samples of similar conditions, the load will increase by approximately 800kW after the mill starts and continue operating; therefore, the correction amount is taken. .

[0084] Predicted equivalent production load .

[0085] In this example, the demand calculation window is 15 minutes, and the sampling period is 1 second. The prediction time range is 15 minutes, therefore Assuming the historical transformer-side power within the current demand window is approximately 9200kW, then... .

[0086] Therefore, the predicted maximum demand is... Transformer available demand margin This indicates that if energy storage continues to charge as planned, it will lead to demand exceeding the limit and breach the demand control boundary after deducting the safety margin, posing a risk of demand exceeding the limit.

[0087] S6 (Generate Energy Storage Charging Power Control Command): Due to the detection of abnormal load conditions and Enter the abnormal control path.

[0088] S63: Permissible charging power under demand constraints Set the operating condition correction coefficient according to the operating condition category. For the mill start-up condition, Take 0.5.

[0089] S64: Preliminary Charging Power Command .

[0090] S65: Maximum charging power under state-of-charge constraints .

[0091] S66: Final energy storage charging power control command .

[0092] S7 (Execute Control Command and Restore): The system reduces the energy storage charging power from 500kW to 250kW. Once the mill startup is complete and the load stabilizes, the load anomaly detector identifies and resolves the abnormal load condition. The demand margin estimator recalculates the results. At this point, the energy storage charging power gradually recovers according to the set ramp rate. Assuming the ramp rate recovers... Then, it will increase by 50kW per minute until it returns to the planned charging power of 500kW.

[0093] Through the above process, this invention, under the abnormal load condition of mill start-up, identifies the operating condition type and predicts the demand change trend, thereby reducing the energy storage charging power in advance, effectively avoiding transformer demand exceeding the limit, and realizing the initiative and accuracy of demand control.

[0094] Therefore, this invention employs the aforementioned demand coordination control method and system for abnormal load conditions in cement plants. By collecting multi-source time-series data and utilizing a CNN-LSTM model to extract waveform features such as peak starting current, peak frequency, and inter-peak time interval, it can accurately distinguish different abnormal load conditions, transforming demand control from a passive threshold response to an active adaptive control. When an abnormal condition that may lead to demand exceeding limits is identified, the system proactively reduces or suspends energy storage charging power to release power space for production load; after the abnormality is resolved, it smoothly resumes at the ramp rate, avoiding secondary impacts. This invention effectively improves the accuracy and adaptability of demand control, reduces the risk of transformer demand exceeding limits penalties, ensures the safe operation of the power supply and distribution system, reduces unnecessary charging restrictions, and improves energy storage utilization efficiency.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for coordinated demand control under abnormal load conditions in a cement plant, characterized in that, Includes the following steps: S1. Collect operating data of the power supply and distribution system and production equipment of the cement plant. The operating data shall include at least three-phase current, three-phase voltage, active power, reactive power, power factor, real-time power on the transformer side, energy storage charging and discharging power, photovoltaic power, waste heat power generation power and operating status signals of production equipment. S2. Calculate the equivalent production load of the plant area based on the collected operating data, and preprocess the operating data to form multi-channel time series data; S3. Use a sliding time window or event-triggered method to slice multi-channel time series data and extract load event segments; S4. Input the load event fragment into the pre-trained load anomaly identification model to identify the current load event's condition category and corresponding identification confidence level. The condition categories include at least normal load fluctuation, mill start-up, fan start-up, heavy load start-up, continuous start-up of multiple devices, material blockage in conveying equipment, and equipment shutdown. S5. Calculate the current real-time demand of the transformer based on the real-time power of the transformer side, and calculate the available demand margin of the transformer within the predicted time range based on the transformer demand limit, the planned charging power of the energy storage, the operating condition category, and the identification confidence level; wherein, the planned charging power of the energy storage is the preset charging power. S6. Generate energy storage charging power control commands based on the operating condition category, identification confidence level, and transformer available demand margin. S7. Execute the energy storage charging power control command. After the abnormal load condition is resolved and the transformer's available demand margin is restored, gradually restore the energy storage charging power to the planned energy storage charging power according to the set ramp rate.

2. The method according to claim 1, characterized in that, The formula for calculating the equivalent production load of the plant area mentioned in S2 is as follows: ； in, Indicates the sampling time. This represents the real-time power on the transformer side. For energy storage charging and discharging power, For photovoltaic power, It generates electricity from waste heat, and during energy storage charging... During energy storage and discharge .

3. The method according to claim 1, characterized in that, The preprocessing described in S2 includes time synchronization, missing value completion, outlier correction, filtering and denoising, and normalization of the running data. The formula for normalization is: ； in, The original data, This is the historical average. For historical standard deviation, This is the normalized data.

4. The method according to claim 1, characterized in that, The event triggering method described in S3 is as follows: when the rate of change of equivalent production load or the rate of change of current exceeds a set threshold, an event slice is triggered. The triggering condition is: ； or ； in, Indicates the sampling time. For the equivalent production load of the plant area, This refers to the effective value of the three-phase current or the combined value of the three-phase current. The power change rate threshold, The threshold for the rate of change of current. The sampling interval is denoted as .

5. The method according to claim 1, characterized in that, The abnormal load condition identification model described in S4 is a CNN-LSTM model, which includes an input layer, a convolutional feature extraction layer, a pooling layer, an LSTM temporal feature learning layer, a fully connected layer, and a classification output layer; the classification output layer uses the Softmax function to output the probability of each type of load condition. ； in, The load event belongs to the first The probability of such working conditions For the first The classification score corresponding to each type of working condition This represents the total number of operating condition categories. Index for operating condition categories; Operating conditions and identification confidence They are respectively: ； 。 6. The method according to claim 1, characterized in that, The available demand margin of the transformer described in S5 The calculation method is as follows: The predicted equivalent production load is adjusted based on the operating condition category and identification confidence level, with the adjustment amount being... When the operating condition is normal load fluctuation, Set to zero; when the operating condition is heavy load start, continuous start of multiple devices, or material blockage in the conveyor equipment. Take a positive value; when the operating condition is equipment trip, Take negative values, where, Indicates the sampling time. The sampling interval is... Indicates the prediction step size index; The predicted transformer-side power is calculated by combining the predicted photovoltaic power and the predicted waste heat power, and the maximum predicted demand within the predicted time range is calculated based on the predicted transformer-side power. The available demand margin for the transformer is: ； in, This is the transformer demand limit. For safety margin.

7. The method according to claim 1, characterized in that, The method for generating the energy storage charging power control command in S6 is as follows: S61. When no abnormal load conditions are identified and the transformer has available demand margin At that time, the energy storage charging power control command is the planned charging power of the energy storage. ; S62. When abnormal load conditions are identified and the transformer has available demand margin At that time, the energy storage charging power control command is the planned charging power of the energy storage. It also records the current operating condition category, the time of occurrence, and the identification confidence level; S63. When abnormal load conditions are identified and the transformer has available demand margin Time: Calculate the allowable charging power under demand constraints based on the transformer's available demand margin. And set the working condition correction factor according to the working condition category. Operating condition correction factor The range of values ​​is Furthermore, the working condition correction factor decreases as the severity of the working condition increases; S64, Energy Storage Plan Charging Power Allowable charging power under demand constraints The product of the operating condition correction factor and the planned charging power of the energy storage The minimum value of the three is used as the initial charging power command. : ； S65, Based on the upper limit of the energy storage state of charge. Current state of charge of energy storage Rated energy storage capacity and charging efficiency Calculate the maximum charging power under state-of-charge constraints: ； S66, Obtain Preliminary Charging Power Command Maximum charging power under state of charge constraints The smaller value in the middle is used as the final energy storage charging power control command. : ； S67. Permissible charging power under demand constraints Or operating condition correction factor At that time, the final energy storage charging power control command is Temporarily suspend energy storage charging; S68. When the abnormal load condition is resolved and the transformer's available demand margin is restored, the final energy storage charging power control command will gradually restore the planned energy storage charging power according to the set ramp rate. The restoration method is as follows: ； in, To restore the ramp rate of energy storage charging power.

8. The method according to claim 5, characterized in that, The convolutional feature extraction layer contains multiple convolutional kernels of different sizes, used to extract waveform features of load event segments at different time scales. The waveform features include at least the peak value of the starting current, the number of peak values, the time interval between peaks, the power change amplitude, and the load change rate. The pooling layer employs max pooling to preserve the peak startup current and power surge amplitude characteristics. The local feature sequences extracted by the convolutional feature extraction layer and the pooling layer are input into the LSTM temporal feature learning layer. The LSTM temporal feature learning layer is used to identify the change pattern of the equipment from before startup, startup process, operation transition to stable state or abnormal state according to the order of local feature sequences in load event segments.

9. A demand coordination control system for cement plants under abnormal load conditions, used to execute the method according to any one of claims 1-8, characterized in that, include: The operation data acquisition unit is used to collect operating data of the cement plant's power supply and distribution system and production equipment; The load anomaly identifier is connected to the operation data acquisition unit. It is used to calculate the equivalent production load of the plant area based on the operation data, preprocess the operation data and extract load event fragments, and identify the operation category and corresponding identification confidence level of the current load event through a pre-trained load anomaly operation condition identification model. The demand margin estimator is connected to the operation data acquisition unit and the load anomaly identifier. It is used to calculate the real-time demand of the transformer based on the real-time power of the transformer side, and to calculate the available demand margin of the transformer within the predicted time range based on the transformer demand limit, the energy storage plan charging power, the operating condition category and the identification confidence level. The load anomaly controller, connected to the demand margin evaluator and load anomaly identifier, is used to generate energy storage charging power control commands based on the operating condition category, identification confidence level, and available demand margin of the transformer. The control switching unit, connected to the load anomaly controller, is used to select and execute either the planned energy storage charging power or the energy storage charging power control command generated by the load anomaly controller, based on the identification results of the load anomaly identifier and the evaluation results of the demand margin estimator.

10. The system according to claim 9, characterized in that, The load anomaly detector includes a data preprocessing module, an event slicing module, and a CNN-LSTM recognition module; After the load anomaly controller outputs a control command to suspend energy storage charging, the control switching unit controls the energy storage charging power to gradually recover to the planned energy storage charging power according to the set ramp rate when the transformer's available demand margin is restored.

Images