Port electrical load coupling photovoltaic generating capacity prediction method, recording medium and system

Abstract

The invention belongs to the technical field of distributed power supply, and particularly relates to a prediction method for port power load coupling photovoltaic power generation capacity, which comprises the following steps of: integrating two groups of time sequence sequences of port photovoltaic power generation capacity and port power consumption into a composite time sequence according to a causal relationship, and training and learning through multi-layer neural networks such as TCN, GRU and an attention layer; the prediction value in the future 24 hours is obtained, the process that a training model must obtain associated features through spontaneous learning of a large amount of original random data is reduced, the burden on the calculation power of the model is reduced, and the calculation efficiency and the accuracy of the prediction result are improved. The invention further provides a non-transient readable recording medium storing the program of the method and a system comprising the medium, and the program can be called through a processing circuit to execute the method.

Description

Technical Field

[0001] The invention belongs to the technical field of distributed power supply and discloses a prediction method, recording medium and system for a port power load coupled photovoltaic power generation. Background Art

[0002] As port operations continue to expand and upgrade, their energy structure is becoming increasingly diversified and complex. The introduction of clean energy such as photovoltaics and wind power has provided ports with environmentally friendly and sustainable energy options, but it has also brought about instability in energy supply and increased management difficulties. In addition, the diversity and dynamics of energy storage systems, shore power systems, and port area loads have also put forward higher requirements for port energy management.

[0003] As an advanced energy management technology, microgrid technology can organically combine distributed power sources, energy storage devices, loads, etc. through power electronic devices and control strategies to form a local power grid system that can operate independently or be connected to the grid. The application of microgrid technology in ports can achieve efficient integration and optimal utilization of clean energy, while improving the reliability and flexibility of port energy systems.

[0004] To achieve port energy management, it is essential to predict the port power load demand and photovoltaic power generation output, and the accuracy of its prediction rate directly affects the economic efficiency of port energy use and the consumption of renewable energy. Existing short-term prediction of microgrid load often adopts data-driven methods. Currently, artificial intelligence algorithms such as long short-term memory neural networks are more commonly used, but their computational complexity is high and they are easily affected by the strong volatility of port power load and photovoltaic power generation, and the prediction accuracy is not high.

[0005] Existing prediction methods often treat port power load and photovoltaic power generation output as two sets of independent data that are unrelated to each other and predict them separately. However, in actual port energy management, managers will adjust port business according to the grid time-of-use electricity price and photovoltaic power generation output, thereby changing the port power load. When photovoltaic power generation is in full swing, the port power load will increase relatively; and when photovoltaic power generation output is insufficient (rainy weather), the general port power load will also take demand-side response measures to control the power load. Existing prediction models ignore the correlation between the two sets of data. In order to obtain accurate prediction values, they have to expand the sample size of the original input or increase the number of training layers of the network, resulting in tight computing power and delayed prediction. If composite prediction is carried out, how to organize the two types of data, how to determine whether the output prediction value for a certain day is port power consumption or photovoltaic power generation, and how to use artificial intelligence technology to efficiently complete the processing of massive data and obtain relatively accurate results are all technical problems that technicians urgently need to solve. Summary of the invention

[0006] In view of the above problems, the present invention provides a method for predicting the amount of power generation coupled with photovoltaic power generation of a port, comprising the following steps:

[0007] The photovoltaic power generation of the port on that day is recorded at 12:00 every day, and the power consumption of the port on that day is recorded at 24:00 every day. The recorded data are arranged in chronological order to form a composite time series of the port power load coupled with photovoltaic power generation;

[0008] Construct a two-layer one-dimensional time domain convolutional neural network, take the composite time series as input, and extract the composite time series feature value Y';

[0009] The composite time series feature value Y' is input into the gated recurrent neural network layer for training, and the trained output value is introduced into the attention mechanism layer to generate a weighted output. After filtering out the data exceeding the set threshold, it is flattened and deconvolved into a one-dimensional array, and then mapped to an output time series through a linear activation function. The data volume and time interval of the output time series are the same as those of the composite time series, but are delayed by one day; of the two output values ​​within the last 24 hours, the output value corresponding to the time point of 12 o'clock is the predicted port photovoltaic power generation at that time, and the output value corresponding to the time point of 24 o'clock is the predicted port electricity consumption at that time.

[0010] Preferably, the extended convolution field of view is used to increase the receptive field in the double-layer one-dimensional time-domain convolutional neural network, and the convolution kernel size is set to 3, the first layer expansion coefficient is 1, and the second layer expansion coefficient is 2.

[0011] Preferably, the learning rate of the gated recurrent unit is set to 0.001, the deactivation rate is set to 0.1, and the number of hidden layer units is set to 128.

[0012] Preferably, the attention mechanism layer corresponds to the output at time k Among them, the attention mechanism layer corresponds to the output at time k Among them, h k is the output of the gated recurrent neural network layer at time k, a k Yes k Normalized weight; s k Represents the output of the attention mechanism layer at time k.

[0013] Another embodiment of the present invention is to provide a non-transitory readable recording medium for storing one or more programs including a plurality of instructions. When the instructions are executed, the processing circuit will execute the above-mentioned method for predicting the coupling of port power load with photovoltaic power generation.

[0014] Another embodiment of the present invention is to provide a prediction system for port power load coupled with photovoltaic power generation, including a processing circuit and a memory electrically coupled thereto, wherein the memory is configured to store at least one program, wherein the program includes a plurality of instructions, and the processing circuit runs the program to execute the above-mentioned prediction method for port power load coupled with photovoltaic power generation.

[0015] Compared with the prior art, the prediction method, recording medium and system of the port power load coupled photovoltaic power generation provided by the present invention have the following beneficial effects: by integrating the two groups of time series of port photovoltaic power generation and port power consumption into a composite time series according to the causal relationship and extracting features through a neural network, the data of port photovoltaic power generation is associated with the data of port power consumption during the training process, which reduces the process that the training model must spontaneously learn from a large amount of original random data to obtain related features, reduces the computing power of the model, and improves the computing efficiency and the accuracy of the prediction results. Among them, the structure of the gated recurrent unit can retain feature information to the maximum extent. When important feature information is transmitted in the gated recurrent unit, an attention mechanism is introduced after the output layer of the gated recurrent unit to perform weighted processing on the important features output by the gated recurrent unit, avoiding the deviation of important information at all levels caused by long time and long sequence, thereby ensuring the reliability and validity of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of a method for predicting port power load coupled with photovoltaic power generation in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be described below in conjunction with the drawings in the embodiments of the present invention. The described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making innovative efforts are within the scope of protection of the present invention.

[0018] See also Figure 1 The embodiment of the present invention provides a unified prediction method for port power load and photovoltaic power generation based on deep learning, comprising the following steps:

[0019] Step 1: Record the photovoltaic power generation of the port at 12:00 every day, and record the power consumption of the port at 24:00 every day, and arrange the recorded data in chronological order to form a composite time series Y of the port power load coupled with photovoltaic power generation;

[0020] Step 2: Construct a two-layer one-dimensional time-domain convolutional neural network (TCN), take the composite time series Y as input, and extract the composite time series feature value Y'. Specifically, the proposed one-dimensional convolutional neural network processes the fused input data Y through two rounds of convolution. The first layer of convolution performs convolution processing on Y, using a convolution kernel size of 3 and a dilation factor of 1 to obtain a unified predicted feature map; the second layer of convolutional neural network implements further convolution processing on the feature map after the first layer of convolution processing, using a convolution kernel size of 3 and a dilation factor of 2, and finally sends the output result Y' to the next layer of gated recurrent unit layer.

[0021] The method for predicting the amount of photovoltaic power generation coupled with port power load according to claim 2 is characterized in that the learning rate of the gated recurrent unit (GRU) is set to 0.001, the deactivation rate is set to 0.1, and the number of hidden layer units is set to 128.

[0022] Step 3: Input the composite time series feature value Y' into the gated recurrent unit layer training. Taking advantage of the simple structure, few parameters and fast execution of the gated recurrent unit, the fusion processing result Y' of the power load and photovoltaic power generation after two series-connected one-dimensional convolutional neural networks is used to predict the 24 hours before the day. Specifically, the learning rate of the gated recurrent unit is set to 0.001, the deactivation rate is set to 0.1, and the number of hidden layer units is set to 128.

[0023] Step 4: Utilize the information filtering function of the attention mechanism to suppress the influence of noise, especially the low-correlation or irrelevant features caused by noise, so as to further improve the accuracy of port power load and photovoltaic power generation prediction.

[0024] Specifically, the data processed in step 3 is analyzed and processed by the attention layer to generate weighted output. The output of the attention mechanism layer at time k Among them, h k is the output of the gated recurrent neural network layer at time k, a k Yes k Normalized weight; s k Represents the output of the attention mechanism layer at time k.

[0025] The attention layer then filters out the data that exceeds the set threshold, flattens it, and deconvolves it into a one-dimensional array, which is then mapped to an output time series through a linear activation function. The data volume and time interval of the output time series are the same as those of the composite time series, but are delayed by one day. The output value corresponding to the time point 12 o'clock in the last two output values ​​within 24 hours is the predicted photovoltaic power generation of the port at that time, and the output value corresponding to the time point 24 o'clock is the predicted power consumption of the port at that time.

[0026] We use the actual power load data and photovoltaic data of a port from January 1, 2021 to January 30, 2022 for training, and predict the actual power load data of the port on January 31, 2022. TGA represents the method proposed in this patent, TG represents a two-layer fusion load prediction algorithm that uses only one-dimensional convolutional networks and gated recurrent units, GRU represents a load prediction algorithm that uses only gated recurrent units, and BGA represents a load prediction algorithm that uses a bidirectional gated recurrent unit and an attention mechanism. The evaluation index determination coefficient (R 2 ), mean absolute error (MAE), and root mean square error (RMSE). 2 It indicates the degree of fit between the model prediction value and the original data. The formula is The closer its value is to 1, the better the model fit is; RMSE and MAE are used to indicate the size of the model prediction error, and the formulas are and The smaller the value, the better the prediction effect of the model. The comparison of prediction accuracy of several methods is shown in the following table.

[0027]

[0028]

[0029] The proposed method achieves unified prediction of port power load and photovoltaic output, and the comparison results show that the prediction error of the proposed algorithm (TGA) in predicting load demand and renewable energy output is significantly smaller than that of TG (a two-layer fusion load prediction algorithm using only one-dimensional convolutional network and gated recurrent unit) and GRU (a load prediction algorithm using only gated recurrent unit), especially in the load demand forecast on January 31, compared with the original GRU model. The RMSE of TGA is reduced by 57.02% and the MAE is reduced by 56.31%. Although the prediction error of TGA in renewable energy output is not much different from that of BGA, its prediction error in predicting load demand is significantly smaller than that of BGA. In summary, TGA has higher prediction accuracy in the two data sets.

[0030] Compiling the above method steps into a program and then storing it on a hard disk or other non-transitory storage medium constitutes an embodiment of "a non-transitory readable recording medium" of the present invention; and electrically connecting the storage medium to a computer processor and completing the prediction of the port power load coupled with photovoltaic power generation through data processing constitutes an embodiment of "a prediction system for port power load coupled with photovoltaic power generation" of the present invention.

[0031] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computers, available storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes.

[0032] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0033] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0034] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0035] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting port power consumption coupled with photovoltaic power generation, characterized in that: The following steps are involved: The photovoltaic power generation of the port on that day is recorded at 12:00 every day, and the power consumption of the port on that day is recorded at 24:00 every day. The recorded data are arranged in chronological order to form a composite time series of the port power load coupled with photovoltaic power generation; Construct a two-layer one-dimensional time domain convolutional neural network, take the composite time series as input, and extract the composite time series feature value Y'; The composite time series feature value Y' is input into the gated recurrent neural network layer for training, and the trained output value is introduced into the attention mechanism layer to generate a weighted output. After filtering out the data exceeding the set threshold, it is flattened and deconvolved into a one-dimensional array, and then mapped to an output time series through a linear activation function. The data volume and time interval of the output time series are the same as those of the composite time series, but are delayed by one day; of the two output values ​​within the last 24 hours, the output value corresponding to the time point of 12 o'clock is the predicted port photovoltaic power generation at that time, and the output value corresponding to the time point of 24 o'clock is the predicted port electricity consumption at that time.

2. A method for predicting port power load coupled with photovoltaic power generation according to claim 1, characterized in that: In the two-layer one-dimensional time-domain convolutional neural network, the extended convolution field of view is used to increase the receptive field. The convolution kernel size is set to 3, the expansion coefficient of the first layer is 1, and the expansion coefficient of the second layer is 2.

3. A method for predicting port power load coupled with photovoltaic power generation according to claim 2, characterized in that: The learning rate of the gated recurrent unit is set to 0.001, the inactivation rate is set to 0.1, and the number of hidden layer units is set to 128.

4. A method for predicting port power load coupled with photovoltaic power generation according to claim 3, characterized in that: The output of the attention mechanism layer corresponding to time k Among them, h k is the output of the gated recurrent neural network layer at time k, a k Yes k Normalized weight; s k Represents the output of the attention mechanism layer at time k.

5. A non-transitory readable recording medium for storing one or more programs including a plurality of instructions, characterized in that: When the instruction is executed, the processing circuit will be caused to execute a method for predicting the amount of photovoltaic power generation coupled with port power load according to any one of claims 1 to 4.

6. A prediction system for port power load coupled with photovoltaic power generation, comprising a processing circuit and a memory electrically coupled thereto, characterized in that: The memory configuration stores at least one program, the program includes a plurality of instructions, and the processing circuit runs the program to execute a method for predicting the amount of photovoltaic power generation coupled with port power load according to any one of claims 1-4.

Images