Port power load boundary prediction system and regulation method based on two-stage model

Abstract

The application discloses a port power load boundary prediction system and regulation method based on a two-stage model, which comprises the following steps: a real-time data acquisition module is mainly used for acquiring the running state data of port equipment in real time, collecting meteorological sensor environment monitoring data and power distribution station power grid load data; a data preprocessing module is electrically connected with the data acquisition module and is mainly used for standardizing the multi-source heterogeneous data; a two-stage prediction module is mainly used for firstly performing short-term load prediction and then combining long-term trend prediction to provide accurate port power load prediction results; an intelligent regulation module is mainly used for generating equipment regulation strategies by using a model prediction control algorithm with constraints; and an execution feedback module is mainly used for realizing closed-loop execution of the regulation strategies and collecting running feedback data. The application has the advantages that through real-time data acquisition and two-stage load prediction, short-term load prediction is firstly performed and then long-term trend prediction is combined, so that short-term fluctuations and long-term changes of power load can be accurately coped with.

Description

TECHNICAL FIELD

[0001] The present application relates to load boundary prediction, in particular to a port power load boundary prediction system and regulation method based on a two-stage model. BACKGROUND

[0002] The port power load boundary prediction system and regulation method is an advanced technology proposed to solve the problems of large fluctuation of port power demand and difficult load management. With the continuous expansion of the port scale and the increasing busy of logistics activities, the port power load shows significant volatility, especially during peak periods, the load demand increases dramatically, which easily leads to power supply and demand imbalance, and even may cause power grid failure or power interruption. This makes it difficult for traditional power management methods to adapt to the increasingly complex changes in port power demand.

[0003] The port power load boundary prediction system and regulation method currently on the market only relies on a single short-term or long-term prediction model when predicting load, and cannot take into account both short-term fluctuations and long-term trends. This makes them often ignore the sharp fluctuations in the short term when dealing with port power load, resulting in inaccurate prediction results, especially in the face of unexpected events or equipment failures, they cannot respond quickly. In addition, existing systems have deficiencies in data processing and fusion, usually only relying on a single source or relatively simple data input, making it difficult to effectively integrate various heterogeneous data such as port equipment, weather data, and power grid load, thereby affecting the overall prediction accuracy and regulation effect. Furthermore, the regulation strategies of many systems are relatively fixed, lacking flexible adaptive ability, and unable to adjust and optimize the scheme in a timely manner according to the actual load changes, which may lead to waste of power resources or equipment overload. SUMMARY

[0004] In order to improve the existing port power load boundary prediction system and regulation method, a port power load boundary prediction system and regulation method based on a two-stage model is provided, which first performs short-term load prediction and then combines long-term trend prediction to accurately respond to short-term fluctuations and long-term changes in power load. This strategy effectively improves the regulation accuracy and stability of the port power system, while optimizing energy utilization efficiency.

[0005] To achieve the above purpose, the technical solution adopted by the present application is:

[0006] The port power load boundary prediction system based on a two-stage model comprises:

[0007] A real-time data acquisition module is provided for acquiring real-time operation state data of port shore cranes, yard cranes, and cold box yard equipment, and simultaneously acquiring environmental monitoring data of weather sensors and power grid load data of distribution stations.

[0008] The data preprocessing module is electrically connected with the data acquisition module and is mainly used for standardizing processing of multi-source heterogeneous data.

[0009] The two-stage prediction module is mainly used for short-term load prediction and long-term trend prediction to provide accurate port power load prediction results.

[0010] The intelligent regulation and control module is mainly used for generating a device regulation and control strategy by using a model predictive control algorithm with a constraint condition based on the load boundary prediction result.

[0011] The execution feedback module is mainly used for realizing closed-loop execution of the regulation and control strategy and collecting running feedback data.

[0012] The database module is mainly used for storing historical port power load data, model parameter data and prediction data.

[0013] Preferably, the real-time data acquisition module specifically comprises:

[0014] The port electrical equipment monitoring unit collects shore-based power supply connection state, instantaneous power and cumulative power consumption.

[0015] The crane operation unit is mainly used for recording start-stop time, working cycle and energy consumption data of devices such as bridge cranes and gantry cranes in real time.

[0016] The power quality analysis unit is mainly used for monitoring voltage fluctuation rate, harmonic distortion rate and three-phase imbalance degree index.

[0017] Preferably, the data preprocessing module specifically comprises:

[0018] The device vibration signal is subjected to wavelet threshold denoising processing.

[0019] The power grid load data is subjected to multiple missing value interpolation and abnormal value sliding window detection.

[0020] The meteorological data is subjected to discretization coding standardization.

[0021] Preferably, the two-stage prediction module specifically comprises:

[0022] The first stage unit is mainly used for load trend prediction, adopts a multi-level LSTM network architecture, extracts load time sequence features through a gating mechanism, and generates a load prediction curve containing short-term fluctuation features and long-term trend components.

[0023] The second stage unit is mainly used for boundary interval generation, and based on the load change law obtained by the first stage unit, a dynamic load boundary interval with time-varying characteristics is constructed according to a quantile regression model.

[0024] Preferably, the first stage unit is mainly used for load trend prediction, adopts a multi-level LSTM network architecture, extracts load time sequence features through a gating mechanism, and generates a basic load prediction curve containing short-term fluctuation characteristics and long-term trend components, specifically including:

[0025] Based on the preprocessed data acquisition module unit data, the sliding mean and variance data features of each unit data are extracted through sliding window statistical calculation to construct a short-term fluctuation prediction data set;

[0026] The long-term trend, seasonality and residual term of the load are separated through STL to perform periodic decomposition, and a long-term fluctuation prediction data set is constructed;

[0027] Based on the processed unit data, the short-term fluctuation prediction data set is preferentially processed, a multivariate time sequence matrix is constructed according to timestamp alignment;

[0028] Based on the multi-level LSTM model, the historical time sequence data is processed through the first layer LSTM, the second layer LSTM, the Dropout layer and the full connection layer, and the load prediction value in the future short-term time is outputted;

[0029] Based on the constructed long-term fluctuation prediction data set, the multi-level LSTM model is further trained and processed, and the load prediction value in the future long-term time is outputted;

[0030] By minimizing the loss function and optimizing the hyperparameters, the load prediction value with the minimum error is obtained, and a load prediction curve containing short-term fluctuation and long-term trend components is outputted.

[0031] Preferably, the second stage unit is mainly used for boundary interval generation, and based on the load change law obtained by the first stage unit, a dynamic load boundary interval with time-varying characteristics is constructed according to a quantile regression model, specifically including:

[0032] Based on the load prediction curve obtained by the first stage unit, the residual sequence of the load prediction curve and the actual value is calculated, and a quantile regression model is established to quantify the uncertainty of the load prediction error;

[0033] Based on the time-varying volatility rate estimation and the quantile regression result, the dynamically adjusted power load upper and lower boundaries are calculated;

[0034] Based on the device physical limit and real-time events, the safety threshold of the boundary interval is corrected;

[0035] The coverage and width of the interval are evaluated based on a multi-objective loss function, and the safety threshold effectiveness is determined by the coverage probability and average width indicators.

[0036] Preferably, the intelligent regulation module specifically comprises:

[0037] Based on the load boundary prediction result, an optimal model is established with the minimization of power consumption cost as an objective function;

[0038] The rolling horizon problem is optimized by the interior point method, and a regulation strategy sequence with a 15-minute interval is generated.

[0039] Further, the port power load boundary prediction and regulation method based on the two-stage model comprises:

[0040] The data acquisition module sensor is used to collect the port equipment operation data, environmental parameters and power grid load data in real time;

[0041] The original data are preprocessed, and a training sample set containing time domain features and frequency domain features is constructed;

[0042] The first-order multi-level LSTM network model is used for short-term and long-term power load prediction, and a load prediction curve is generated;

[0043] In the second stage, a dynamic load boundary interval with time-varying characteristics is constructed according to the quantile regression model;

[0044] Based on the load boundary constraint, a multi-objective optimization model is established to intelligently regulate the port equipment;

[0045] The regulation instruction is sent to the port operation equipment, and the prediction model parameters are updated based on the execution effect feedback data.

[0046] Compared with the prior art, the advantages of the present application are:

[0047] By first conducting short-term load forecasting and then combining long-term trend forecasting, the fluctuation of power load in the short term and the trend of long-term changes can be fully considered. The short-term load forecasting module mainly relies on real-time collection of port equipment operating status, weather data, and power grid load information, and can quickly respond and accurately predict the load demand in the short term. The prediction accuracy in this stage is high, which is suitable for real-time scheduling and short-term decision-making, ensuring the stability and reliability of the power system in the short term. Combined with long-term trend forecasting, the system can predict the power load changes in a long period of time in the future according to historical data and long-term trends, helping the port power system to foresee possible load growth or decline and make long-term power scheduling and optimization planning. The two-stage model effectively combines the high timeliness of short-term prediction and the comprehensiveness of long-term trend prediction, ensuring accurate grasp of current load changes and making full preparations for future power demand. Overall, this method not only improves the accuracy of load forecasting, but also enhances the intelligent control capability of the port power system, thereby improving energy use efficiency and system reliability. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 System diagram for the present invention;

[0049] Figure 2 Method diagram for the present invention;

[0050] Figure 3 Data acquisition module diagram for the present invention;

[0051] Figure 4 Data preprocessing flowchart for the present invention;

[0052] Figure 5 Two-stage prediction module diagram for the present invention;

[0053] Figure 6 First-stage unit workflow diagram for the present invention;

[0054] Figure 7 Second-stage unit workflow diagram for the present invention;

[0055] Figure 8 Intelligent control module workflow diagram for the present invention;

[0056] Figure 9 Architecture diagram of the electronic device in the present scheme;

[0057] Figure 10 Structure diagram of the computer readable storage medium in the present scheme. DETAILED DESCRIPTION

[0058] The following description is used to disclose the present application to enable a person skilled in the art to implement the present application. The preferred embodiments in the following description are only as examples, and other obvious modifications can be thought of by those skilled in the art.

[0059] Referring to Figure 1 As shown in the figure, the port power load boundary prediction system based on a two-stage model comprises:

[0060] A real-time data acquisition module is mainly used for acquiring the running state data of port shore cranes, yard cranes and cold box yard equipment in real time, and synchronously acquiring environmental monitoring data of weather sensors and power grid load data of distribution stations;

[0061] A data preprocessing module is electrically connected with the data acquisition module and is mainly used for standardizing processing of multi-source heterogeneous data;

[0062] A two-stage prediction module is mainly used for firstly performing short-term load prediction and then combining long-term trend prediction to provide accurate port power load prediction results;

[0063] An intelligent control module is mainly used for generating equipment control strategies by using a model predictive control algorithm with constraint conditions based on the load boundary prediction results;

[0064] An execution feedback module is mainly used for realizing closed-loop execution of control strategies and collecting running feedback data;

[0065] A database module is mainly used for storing historical port power load data, model parameter data and prediction data.

[0066] Referring to Figure 2 As shown in the figure, the port power load boundary prediction and control method based on a two-stage model comprises:

[0067] Through a data acquisition module sensor, port equipment running data, environmental parameters and power grid load data are collected in real time;

[0068] The original data are preprocessed to construct a training sample set containing time domain features and frequency domain features;

[0069] A first-stage multi-level LSTM network model is used for short-term and long-term power load prediction to generate a load prediction curve;

[0070] In a second stage, a quantile regression model is used to construct a dynamic load boundary interval with time-varying characteristics;

[0071] Based on the load boundary constraint, a multi-objective optimization model is established to intelligently control the port equipment;

[0072] Control commands are issued to port operating equipment, and the algorithm is optimized and the prediction model parameters are updated based on the feedback data of the execution effect.

[0073] See Figure 3 As shown, the real-time data acquisition module specifically includes:

[0074] Port Area Electrical Equipment Monitoring Unit: The port area electrical equipment monitoring unit collects the connection status, instantaneous power, and cumulative power consumption of the shore-based power supply;

[0075] Crane Operation Unit: The crane operation unit is mainly used to record the start-up and stop times, working cycles, and energy consumption data of equipment such as bridge cranes and gantry cranes in real time;

[0076] Power quality analysis unit: The power quality analysis unit is mainly used to monitor voltage fluctuation rate, harmonic distortion rate and three-phase imbalance index.

[0077] Specifically, the connection status of the power supply is determined by monitoring the voltage and current signals of the shore power interface. If the voltage input is greater than the preset threshold, the power supply is connected normally; otherwise, it is disconnected.

[0078] Instantaneous power can be calculated using real-time voltage and real-time current, using the following formula:

[0079] P(t)=V(t)×I(t)×cos(θ) where P(t) is the instantaneous power, V(t) is the instantaneous voltage, I(t) is the instantaneous current, θ is the phase angle between the voltage and the current, and cos(θ) is the power factor;

[0080] Cumulative electricity consumption refers to the total energy consumption over a period of time, which can be obtained by summing instantaneous power consumption. The formula is:

[0081]

[0082] The crane operating unit uses sensors or control systems to monitor the crane's start and stop events, determines whether the equipment is in use based on the equipment's operating signals, and calculates the equipment's working cycle and energy consumption based on the instantaneous power acquisition results.

[0083] The power quality analysis unit monitors relevant power quality indicators, including voltage fluctuation rate, harmonic distortion rate, and three-phase imbalance. Voltage fluctuation rate is the relative amplitude of voltage fluctuations, usually calculated by measuring the range of voltage variation. The formula is:

[0084]

[0085] Among them, V max V min V represents the maximum and minimum values ​​of voltage fluctuation. nom Rated voltage;

[0086] Harmonic distortion rate is the proportion of high-order harmonics in power system, which affects power quality. The frequency spectrum of voltage or current waveform is obtained by Fourier transform, and the harmonic distortion is calculated, the formula is:

[0087]

[0088] Among them, V n is the amplitude of the nth harmonic, V1 is the amplitude of the fundamental wave, and THD represents the degree of harmonic distortion;

[0089] Three-phase unbalance degree measures the unbalance degree of three-phase current or voltage. It is evaluated by calculating the relative difference of three-phase voltage or current.

[0090] Referring to Figure 4 , the data preprocessing module specifically includes:

[0091] Wavelet threshold denoising processing is performed on the equipment vibration signal;

[0092] Multiple imputation of missing values and sliding window detection of outliers are performed on the power grid load data;

[0093] Discrete coding standards are established for meteorological data.

[0094] Specifically, through wavelet decomposition, the vibration signal is decomposed into low-frequency parts (approximation coefficients) and high-frequency parts (detail coefficients) of different scales. The high-frequency detail coefficients obtained by decomposition are processed by threshold denoising, including soft threshold and hard threshold methods. The coefficients after thresholding are used for wavelet reconstruction to obtain the denoised signal, and the formula is:

[0095]

[0096] Among them, is the wavelet basis function, is the decomposition coefficient;

[0097] For missing values in power grid load data, multiple imputation techniques are used to fill in missing data, including the following steps:

[0098] A regression model is established according to the historical load data of the data;

[0099] The missing values are predicted and filled in, and the imputation process is repeated multiple times;

[0100] Generate multiple possible filled data sets;

[0101] Multiple imputed data sets are analyzed and the results are combined.

[0102] Anomalies are detected by a sliding window method, which identifies outliers by calculating statistical features (such as mean, standard deviation) within a window of data, including:

[0103] Setting the window size;

[0104] Calculating the mean and standard deviation of the data within each window;

[0105] If a data point deviates from the mean by more than a certain threshold, it is determined to be an outlier.

[0106] The outlier determination formula is:

[0107] Outlier = |x i - μ | > k x σ

[0108] Where x i is the i-th data point within the window, μ is the mean of the data within the window, σ is the standard deviation, and k is a constant that controls the threshold;

[0109] Meteorological data usually includes continuous data (such as temperature, humidity, wind speed, etc.), and discretization converts continuous data into discrete values, such as using integer numbers. One-Hot encoding or label encoding can be used for conversion, making it easier to analyze and model.

[0110] Referring to Figure 5 , the two-stage prediction module specifically includes:

[0111] First stage unit: the first stage unit is mainly used for load trend prediction, adopts a multi-level LSTM network architecture, extracts load time series features through a gating mechanism, and generates a load prediction curve containing short-term fluctuation features and long-term trend components;

[0112] Second stage unit: the second stage unit is mainly used for boundary interval generation, based on the load variation law obtained by the first stage unit, and according to the quantile regression model, a dynamic load boundary interval with time-varying characteristics is constructed.

[0113] Referring to Figure 6 , the first stage unit is mainly used for load trend prediction, adopts a multi-level LSTM network architecture, extracts load time series features through a gating mechanism, and generates a basic load prediction curve containing short-term fluctuation features and long-term trend components, specifically including:

[0114] Based on the pre-processed data acquisition module unit data, the sliding mean and variance data features of each unit data are extracted through sliding window statistical calculation, and a short-term fluctuation prediction data set is constructed;

[0115] The long-term trend, seasonality and residual term of the load are separated by STL to perform periodic decomposition, and a long-term fluctuation prediction data set is constructed;

[0116] Based on the processed unit data, the short-term fluctuation prediction data set is preferentially processed, and a multivariate time sequence matrix is constructed according to timestamp alignment;

[0117] Based on the multilevel LSTM model, the historical time sequence data is processed through the first layer LSTM, the second layer LSTM, the Dropout layer and the full connection layer, and the load prediction value in the future short-term time is output;

[0118] Based on the constructed long-term fluctuation prediction data set, the multilevel LSTM model is further trained and processed, and the load prediction value in the future long-term time is output;

[0119] By minimizing the loss function and optimizing the hyperparameters, the load prediction value with the minimum error is obtained, and the load prediction curve containing the short-term fluctuation and long-term trend component is output.

[0120] Specifically, the electric load data is seasonally decomposed, the STL (seasonal and trend decomposition) method is used to separate the long-term trend, seasonality and residual term of the data, and the formula is:

[0121] x t =T t +S t +R t

[0122] Wherein, T t is the long-term trend, S t is the periodic seasonal component, and R t is the residual component;

[0123] Through the processed data, including the sliding mean, variance characteristics, long-term trend, seasonality and residual term, they are aligned according to the timestamp, and a multivariate time sequence matrix is constructed. Assuming that there are n characteristics (sliding mean, variance, etc.), then at time step t, the time sequence matrix can be represented as:

[0124] X t =[Mean t ,Variance t ,T t ,S t ,R t ]

[0125] Wherein, Mean t , Variance t are short-term characteristics calculated by sliding window, T t , S t , R t are long-term trend, seasonality and residual term obtained by STL decomposition;

[0126] These features and target variables (e.g. future load values) are inputted as training data into the LSTM model for training;

[0127] Short-term volatility prediction based on multi-level LSTM model includes:

[0128] Input layer: input multivariate time series matrix as input of LSTM;

[0129] First layer LSTM: input historical time series data into the first layer LSTM network, set input as X t , LSTM layer processes time series data through internal state memory: h t = LSTM(X t , W h , b h ) where h t is the output of the LSTM layer, W h is the weight of the LSTM layer, and b h is the bias term;

[0130] Second layer LSTM: input h t of the first layer as input of the second layer LSTM, further extract more complex features. h t 2 = LSTM(X t , W h , b h ), where h t 2 is the output of the second layer LSTM;

[0131] Dropout layer: apply Dropout layer to avoid overfitting. By discarding a portion of the activation values of neurons, the generalization ability of the model is improved;

[0132] Fully connected layer: pass the LSTM output to the fully connected layer to get the predicted value: where y is the predicted load value, W out is the weight of the fully connected layer, and b out is the bias term;

[0133] Output layer: output the load prediction value in the future short-term time.

[0134] Long-term volatility prediction based on multi-level LSTM model includes:

[0135] Use long-term volatility prediction dataset (contains long-term trend T t and seasonal component S t ) to construct time series matrix;

[0136] The same LSTM architecture (including two layers of LSTM, a Dropout layer, and a fully connected layer) is used for training and prediction.

[0137] The long-term fluctuations are predicted by the LSTM network, and the load prediction value in the future long-term time is output.

[0138] The results of short-term fluctuation prediction and long-term trend prediction are integrated to obtain the final load prediction curve.

[0139] Referring to Figure 7 As shown in the figure, the second stage unit is mainly used for boundary interval generation, based on the load variation law obtained by the first stage unit, a dynamic load boundary interval with time-varying characteristics is constructed according to the quantile regression model, which specifically includes:

[0140] Based on the load prediction curve obtained by the first stage unit, the residual sequence of the actual value is calculated, and a quantile regression model is established to quantify the uncertainty of the load prediction error.

[0141] Based on the time-varying volatility estimation and quantile regression results, the dynamically adjusted power load upper and lower boundaries are calculated.

[0142] Based on the physical limit of the device and real-time events, the safety threshold of the boundary interval is corrected.

[0143] Based on the multi-objective loss function, the coverage rate and width of the interval are evaluated, and the safety threshold effectiveness is judged by the coverage probability and average width indicators.

[0144] Specifically, quantile regression quantifies the uncertainty of load prediction error by predicting different quantiles (such as 0.05, 0.5, 0.95), in order to quantify the error of load prediction, a quantile regression model based on residual is constructed. Set a target quantile τ∈[0,1], the goal of the quantile regression model is to predict the prediction value corresponding to each quantile;

[0145] The time-varying volatility reflects the uncertainty of load fluctuation, and the volatility is estimated by the GARCH model, the formula is:

[0146]

[0147] Where α0, α1, β1 are GARCH model parameters, is the estimated value of time-varying volatility;

[0148] Using time-varying volatility and quantile regression results, the upper and lower boundaries of power load can be calculated, based on the physical limit of the device and real-time events (such as device failure, weather change, etc.), the upper and lower boundaries of power load can be corrected to ensure safe operation of the power grid.

[0149] The coverage and width indicators are calculated by a multi-objective loss function, the coverage represents the proportion of actual load values within the predicted upper and lower boundary intervals, and the width represents the width of the load upper and lower boundary interval.

[0150] Referring to Figure 8 As shown in the figure, the intelligent regulation and control module specifically includes:

[0151] Based on the load boundary prediction result, an optimal model is established with the minimization of electricity cost as the objective function.

[0152] The interior point method is used to optimize the rolling horizon problem, and a 15-minute interval regulation strategy sequence is generated.

[0153] Specifically, based on the load boundary prediction result as a constraint condition, the target of electricity cost is to minimize the following objective function:

[0154]

[0155] Wherein, p t is the electricity price at time t, x t is the power consumption at time t, and T is the total number of time periods in the prediction time domain.

[0156] When the interior point method is used to optimize the rolling horizon problem, it is modeled as a linear programming problem or a quadratic programming problem, and the specific form is as follows:

[0157]

[0158] The interior point method is used to solve the optimization problem, and the specific steps include:

[0159] Initialize all variables, and set the step size and convergence accuracy;

[0160] Obtain the optimal solution of each step by solving the linear system;

[0161] In each iteration, update the decision variables and correct the dual variables to meet the constraint conditions.

[0162] Until the predetermined accuracy is reached or the maximum number of iterations is reached.

[0163] Further, the method according to the embodiments of the present application can also be implemented by means of Figure 9 The architecture of the electronic device is shown in the figure. As Figure 9As shown, the electronic device 500 can include a bus 501, one or more CPUs 502, a read-only memory (ROM) 503, a random access memory (RAM) 504, a communication port connected to a network 505, an input / output component 506, a hard disk 507, etc. The storage device in the electronic device 500, such as the ROM 503 or the hard disk 507, can store the port power load boundary prediction system and regulation method based on the two-stage model provided in the present application. The electronic device 500 can also include a terminal interface 508. Of course, Figure 9 The architecture shown is only exemplary, and in implementing different devices, according to actual needs, some of the components shown can be omitted Figure 9 One or more components in the electronic device shown.

[0164] Figure 10 is a computer readable storage medium structure diagram provided by an embodiment of the present application. As shown in Figure 10 As shown, the computer readable storage medium 600 according to an embodiment of the present application. The computer readable storage medium 600 stores computer readable instructions. When the computer readable instructions are run by the processor, the port power load boundary prediction system and regulation method based on the two-stage model according to the embodiments of the present application described with reference to the above figures can be executed. The storage medium 600 includes, but is not limited to, for example, volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and cache memory, etc. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0165] It should be noted that the above-mentioned order of the embodiments of the present application is only for description, not representing the advantages and disadvantages of the embodiments. And the above describes the specific embodiments of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0166] Each embodiment in the present specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other, and each embodiment focuses on the differences from other embodiments.

[0167] The above is only the preferred embodiment of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A port power load boundary prediction system based on a two-stage model, characterized by: including: a real-time data acquisition module: the real-time data acquisition module is used to acquire the running state data of the port shore crane, yard crane and cold box yard equipment in real time, and synchronously acquire the environmental monitoring data of the weather sensor and the power grid load data of the power distribution station; a data preprocessing module: the data preprocessing module is electrically connected with the data acquisition module, and is used for standardizing the multi-source heterogeneous data; a two-stage prediction module: the two-stage prediction module is used for short-term load prediction, and then combined with long-term trend prediction to provide accurate port power load prediction results; the two-stage prediction module includes: a first stage unit: the first stage unit is used for load trend prediction, adopts a multi-level LSTM network architecture, extracts load time sequence features through a gating mechanism, and generates a load prediction curve containing short-term fluctuation characteristics and long-term trend components; a second stage unit: the second stage unit is used for boundary interval generation, and based on the load variation law obtained by the first stage unit, a dynamic load boundary interval with time-varying characteristics is constructed according to a quantile regression model; the first stage unit is used for load trend prediction, adopts a multi-level LSTM network architecture, extracts load time sequence features through a gating mechanism, and generates a basic load prediction curve containing short-term fluctuation characteristics and long-term trend components, including: based on the preprocessed data acquisition module unit data, the sliding mean and variance data features of each unit data are extracted through sliding window statistical calculation to construct a short-term fluctuation prediction data set; the long-term trend, seasonality and residual term of the load are separated through STL to perform periodical decomposition, and a long-term fluctuation prediction data set is constructed; based on the processed unit data, the short-term fluctuation prediction data set is preferentially processed, and a multivariate time sequence matrix is constructed according to the timestamp alignment; based on the multi-level LSTM model, the historical time sequence data is processed through the first layer LSTM, the second layer LSTM, the Dropout layer and the full connection layer, and the load prediction value in the future short-term time is outputted; based on the constructed long-term fluctuation prediction data set, the multi-level LSTM model is further trained and processed, and the load prediction value in the future long-term time is outputted; by minimizing the loss function and optimizing the hyperparameters, the load prediction value with the minimum error is obtained, and a load prediction curve containing short-term fluctuation and long-term trend components is outputted; the second stage unit is used for boundary interval generation, and based on the load variation law obtained by the first stage unit, a dynamic load boundary interval with time-varying characteristics is constructed according to a quantile regression model, including: based on the load prediction curve obtained by the first stage unit, the residual sequence of the load prediction curve and the actual value is calculated, and a quantile regression model is established to quantify the uncertainty of the load prediction error; based on the time-varying volatility rate estimation and the quantile regression result, the dynamically adjusted power load upper and lower boundaries are calculated; based on the device physical limit and the real-time event, the safety threshold of the boundary interval is corrected; based on the multi-objective loss function, the coverage rate and the width of the interval are evaluated, and the safety threshold effectiveness is judged through the coverage probability and the average width index; The intelligent regulation module is configured for generating a device regulation strategy by using a model predictive control algorithm with a constraint condition based on the load boundary prediction result. The execution feedback module is configured for realizing closed-loop execution of the regulation strategy and collecting running feedback data. The database module is configured for storing historical power load data, model parameter data, and prediction data of the port area.

2. The port area power load boundary prediction system based on a two-stage model according to claim 1, characterized in that: The real-time data collection module comprises: The port area electrical equipment monitoring unit is configured to collect shore-based power supply connection states, instantaneous power, and cumulative power consumption. The crane running unit is configured to record start-stop time, working cycle, and energy consumption data of the bridge crane and the gantry crane in real time. The power quality analysis unit is configured to monitor voltage fluctuation rate, harmonic distortion rate, and three-phase imbalance degree indicators.

3. The port area power load boundary prediction system based on a two-stage model according to claim 1, characterized in that: The data preprocessing module comprises: Wavelet threshold denoising processing is performed on the device vibration signal. Multiple imputation of missing values and sliding window detection of outliers are performed on the power grid load data. A discretization coding standard is established for the meteorological data.

4. The port area power load boundary prediction system based on a two-stage model according to claim 1, characterized in that: The intelligent regulation module comprises: Based on the load boundary prediction result, an optimal model with a minimum power consumption cost as an objective function is established. The interior point method is used to optimize the rolling horizon problem, and a regulation strategy sequence with an interval of 15 minutes is generated.

5. A port area power load boundary prediction and regulation method based on a two-stage model, which uses the port area power load boundary prediction system based on a two-stage model according to any one of claims 1-4, characterized in that: It comprises: Port equipment running data, environmental parameters, and power grid load data are collected in real time by the data collection module sensors. The original data are preprocessed to construct a training sample set containing time domain features and frequency domain features. In the first stage, a multi-level LSTM network model is used to perform short-term and long-term power load prediction, respectively, to generate a load prediction curve. In the second stage, a quantile regression model is used to construct a dynamic load boundary interval with time-varying characteristics. Based on the load boundary constraint, a multi-objective optimization model is established to intelligently regulate the port area devices. The regulation instructions are sent to the port operation equipment, and the prediction model parameters are updated based on the execution effect feedback data.

6. An electronic device, characterized in that: It comprises: At least one processor; and a memory connected in communication with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the port area power load boundary prediction and regulation method based on a two-stage model according to claim 5.

7. A computer-readable storage medium storing computer-readable instructions, characterized in that: The computer readable instructions, when executed by the processor, implement the port power load boundary prediction and regulation method based on the two-stage model in claim 5.

Images