Ultra-short-term prediction method and system for power load of offshore oilfield

Abstract

The invention discloses an offshore oilfield power load ultra-short-term prediction method and system. The method comprises the following steps: obtaining related data of an offshore oilfield in real time; preprocessing the related data of the offshore oilfield; according to a preset sliding window, dividing the preprocessed offshore oilfield related data into multiple segments of time series data; constructing a time sequence diagram based on the multiple segments of time sequence data; and inputting the time sequence diagram into the mobile pooling graph neural network, and outputting an offshore oilfield power load prediction value. According to the method, the time features of the related data of the offshore oilfield are extracted, the spatial dependency among different variables is captured through the graph neural network, the prediction performance of the model is remarkably improved through the multi-level feature extraction method, and complex dynamic changes can be better dealt with.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning technology, and more specifically to an ultra-short-term prediction method and system for offshore oil field power load. Background Art

[0002] Offshore oilfield power load forecasting is an important part of ensuring the safe and efficient operation of offshore platforms. Accurate power load forecasting can optimize the allocation of power resources, reduce energy waste, and prevent potential equipment failures. However, the power load of offshore platforms is affected by many factors, including environmental conditions (such as weather and sea conditions), platform production activities (such as drilling and water injection operations), equipment operating status and maintenance strategies. These factors not only have complex dynamic changes in time, but also show significant correlation in space. Therefore, how to comprehensively consider the impact of these multiple variables at different time points has become a key challenge to improve the accuracy of offshore oilfield power load forecasting.

[0003] The current offshore platform power load forecasting methods have the following main problems and shortcomings:

[0004] Single factor consideration: Many existing methods tend to consider only a few key factors (such as historical load data or a specific environmental variable), while ignoring other important influencing factors, resulting in incomplete and inaccurate prediction results.

[0005] Ignoring spatiotemporal dependencies: Existing methods generally fail to effectively capture spatiotemporal dependencies in multivariate time series data. Traditional time series analysis methods (such as ARIMA and simple regression models) mainly focus on changes in the time dimension, while ignoring the relationships between variables at different time points.

[0006] Insufficient data fusion capabilities: The fusion and comprehensive processing of multivariate data is a complex issue. Existing technologies often lack effective methods when processing and fusing multiple types of data, resulting in insufficient information utilization and affecting prediction performance.

[0007] Therefore, how to solve the problem that the impact of multiple factors at different time points cannot be comprehensively considered in offshore platform power load forecasting is an issue that technical personnel in this field urgently need to solve. Summary of the invention

[0008] In view of the above problems, the present invention provides an ultra-short-term prediction method and system for offshore oilfield power load, so as to at least solve some of the technical problems mentioned in the above background technology.

[0009] In order to achieve the above object, the present invention adopts the following technical solution:

[0010] On the one hand, the present invention provides an ultra-short-term prediction method for offshore oilfield power load, comprising the following steps:

[0011] S1. Obtain relevant data of offshore oil fields in real time;

[0012] S2. Preprocessing the offshore oil field related data;

[0013] S3, dividing the pre-processed offshore oilfield related data into multiple time series data segments according to a preset sliding window;

[0014] S4, constructing a time series diagram based on the multiple time series data;

[0015] S5. Input the time series diagram into the mobile pooling graph neural network, and output the predicted value of the offshore oil field power load.

[0016] Furthermore, the offshore oilfield related data includes environmental data, production activity data and equipment status data.

[0017] Furthermore, the preprocessing includes cleaning outliers, normalizing and filling missing data.

[0018] Furthermore, the step S3 specifically includes:

[0019] S31, encoding each segment of time series data to obtain variable node features;

[0020] S32, based on the variable node features, adding relative position information to the variables between each segment of time series data by position embedding, to obtain updated variable node features;

[0021] S33. Construct a time series diagram based on the updated variable node features.

[0022] Furthermore, each timing feature in the timing diagram is connected via a decreasing matrix; the decreasing matrix is ​​represented as:

[0023]

[0024] Where C represents a decreasing matrix; t represents the tth time series data; r represents the rth time series data; c tr,ij Represents the element in the i-th row and j-th column in the decreasing matrix C; D represents the dimension of the time series data; K represents a total of K segments of time series data; i represents the i-th dimension; j represents the j-th dimension.

[0025] Furthermore, the mobile pooling graph neural network includes a mobile window layer and a time pooling layer;

[0026] The moving window layer processes the time sequence diagram to obtain propagation characteristics of the neighboring node information in the process of being propagated to the central node;

[0027] Updating the propagation characteristics to obtain updated propagation characteristics;

[0028] Processing the update propagation features through the time pooling layer to obtain sensor features corresponding to each window;

[0029] The sensor features corresponding to each window are stacked to output the offshore oilfield power load forecast value.

[0030] Furthermore, it also includes:

[0031] S6. Displaying the offshore oil field power load forecast value through a visualization tool.

[0032] On the other hand, the present invention provides an ultra-short-term prediction system for offshore oilfield power load, applying the above method, the system comprises: a data acquisition module, a data processing module, a time sequence diagram construction module and a prediction module;

[0033] The data acquisition module is used to obtain relevant data of offshore oil fields in real time;

[0034] The data processing module is used to pre-process the offshore oilfield related data; and divide the pre-processed offshore oilfield related data into multiple time series data segments according to a preset sliding window;

[0035] The timing diagram construction module is used to construct a timing diagram based on the multiple segments of timing data;

[0036] The prediction module is used to input the time series diagram into a trained graph neural network and output a predicted value of offshore oil field power load.

[0037] Furthermore, it also includes: a display module;

[0038] The display module is used to display the offshore oil field power load forecast value through a visualization tool.

[0039] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method for ultra-short-term prediction of offshore oilfield power load, which has the following beneficial effects:

[0040] The present invention extracts the temporal characteristics of offshore oilfield related data and captures the spatial dependencies between different variables through graph neural networks. This multi-level feature extraction method significantly improves the predictive performance of the model and can better cope with complex dynamic changes.

[0041] When processing and fusing various types of data, the present invention adopts a data processing method of constructing a space-time graph, which can make full use of multivariate information; by constructing a time series graph in which variables are fully connected, a comprehensive interconnection between different time points and different variables is achieved, thereby improving the efficiency and effect of data fusion.

[0042] Other features and advantages of the present invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the present invention.

[0043] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0045] Figure 1 A schematic flow chart of an ultra-short-term forecasting method for offshore oilfield power load provided in an embodiment of the present invention.

[0046] Figure 2 A schematic diagram of an adjacency matrix provided by an embodiment of the present invention.

[0047] Figure 3 A schematic diagram of a decreasing matrix provided by an embodiment of the present invention.

[0048] Figure 4 A schematic diagram of a moving window layer provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] Embodiment 1:

[0051] See also Figure 1 As shown, the embodiment of the present invention discloses an ultra-short-term prediction method for offshore oil field power load, comprising the following steps:

[0052] S1. Obtain relevant data of offshore oil fields in real time;

[0053] S2. Preprocessing of offshore oilfield related data;

[0054] S3, dividing the pre-processed offshore oilfield related data into multiple time series data segments according to a preset sliding window;

[0055] S4. Construct a time series diagram based on multiple time series data;

[0056] S5. Input the time series diagram into the trained graph neural network and output the predicted value of offshore oil field power load.

[0057] The following is a detailed description of each of the above steps.

[0058] In the above step S1, offshore oil field related data is obtained in real time, including environmental data (such as temperature, humidity and wind speed, etc.), production activity data (such as drilling speed and water injection volume, etc.) and equipment status data (such as equipment operating time and power consumption, etc.);

[0059] In the above step S2, the offshore oilfield related data is preprocessed, specifically including cleaning outliers, normalizing and filling missing data, so as to make the related data more accurate and ensure the accuracy of subsequent predictions.

[0060] In the above step S3, the pre-processed offshore oilfield related data is divided into multiple time series data according to the preset sliding window; specifically: assuming that the data length of the offshore oilfield related data is N and the dimension is D, the offshore oilfield related data is divided into K segments according to the preset sliding window, t is the index of the segment, each time series data corresponds to a timestamp, and the length of each time series data Each time series data segment contains the segmented time series of N variables, that is, Among them, X t Represents a segment of time series data after segmentation; x t,i Represents the t-th segment of time series data X t The time series data of the i-th dimension in .

[0061] In the above step S4, a time series diagram is constructed based on multiple time series data segments, which specifically includes:

[0062] (1) Encode each segment of time series data to obtain time features; specifically:

[0063] Use encoder f(·|W c ) to process each time series data segment, the encoder learns the variable node features in each variable of each time series data segment and initializes the node features of the graph:

[0064] x t ' ,i =f(x t,i |W c)

[0065] Among them, x t ' ,i Represents the node features after initialization; W c represents the learnable weight matrix; f(·) represents the nonlinear transformation. By position embedding, the relative position information of the variables between each time series data segment is added, that is, the variable node feature is updated as: t,i =f p (t)+x t ' ,i ; where the position embedding is expressed as:

[0066]

[0067] Where m represents the mth timestamp of the variable; k represents the kth time series data; f p (t) represents the encoding function of the relative position at time t; w k represents the scaling factor, which makes sine and cosine waves of different dimensions have different frequencies. As k increases, the scaling factor will also become larger, resulting in higher-dimensional values ​​corresponding to lower-frequency sine and cosine waves.

[0068] (2) Based on the updated variable node features, a time series graph is constructed; specifically:

[0069] A time series graph is constructed based on the updated variable node features obtained above. By considering the correlation of different variables at different timestamps, all time points of all variables are interconnected. When constructing the time series graph, it is assumed that related variables should have similar properties and their features are close in the feature space. Here, a similarity metric is used to represent the correlation between different variables. The greater the similarity, the higher the correlation. In the embodiment of the present invention, a simple dot product is selected as the similarity metric. The correlation between different variables is expressed as:

[0070] e tr,ij =g s (z t,i )(g s (z r,j )) T

[0071] where t,r∈[1,K],i,j∈[1,D]; z t,i represents the updated variable node feature; z r,j Represents a time series data point after the position encoding is introduced;

[0072] Here we use the function g s (z)=zW s To enhance the expression ability, Ws is a learnable weight matrix. z generally refers to z r,j and zt,i ; At the same time, softmax limits the correlation to [0,1]. Finally, the time series graph G is derived; the time series graph G is expressed as:

[0073] G=(Z,E)

[0074] in, Z represents each time series data (1-K segments), and the set of the 1st to D dimensions of each time series data segment. E represents the adjacency matrix of the time series graph, and the elements in the adjacency matrix E represent the correlations between variables across all segments. The time series graph G contains not only the temporal correlations between timestamps and spatial correlations in each timestamp, but also the correlations between different variables at different timestamps, which enables modeling of comprehensive spatiotemporal dependencies in multivariate data of offshore platform loads.

[0075] Since the time series graph G is constructed based on the similarity of the variables between each segment of time series data, the time distance of the variables of each segment of time series data is not considered at this time; the similarity between variables with a longer timestamp distance should be weaker; therefore, in an embodiment of the present invention, it is proposed that each time series feature in the time series graph is connected by a decreasing matrix, and the decreasing matrix combines the timestamp distances between variables of different segments, aiming to improve the accuracy of the graph G; specifically:

[0076] Figure 2 represents an adjacency matrix containing three segments of time series data, each segment contains 4 variables, and the dimension of the adjacency matrix is In the adjacency matrix, each row represents the connection (similarity) between a variable and other variables in all segments; taking the first row as an example, it represents the first variable z in segment (T-1) T-1，1 The first four columns represent its connection with variables in the same segment; since these variables appear at the same time, they should show stronger correlation than in other segments; the next four columns represent z T-1，1 Connections to variables in segment T. Since these variables are in different segments, they are connected to z T-1，1 The correlation should be weakening; measured by the decreasing rate δ. The last four columns represent z T-1，1 The connection with the variable of the (T+1) segment. As the time distance increases, the correlation will naturally decrease further, with δ 2 Therefore, the decreasing matrix is ​​proposed Each element c tr,ij =δ t-r ,like Figure 3 This matrix is ​​used to enhance the correlation between variables in different segments. tr,ij =e tr,ij ·c tr,ij,This approach ensures that variables that are close in time show stronger correlations than variables that are far away in time.

[0077] In the above step S5, in order to use the above-constructed time series graph G to capture the spatiotemporal dependency in the multivariate time series data of the offshore platform, the embodiment of the present invention proposes a mobile pooling graph neural network, inputs the above-obtained time series graph into the mobile pooling graph neural network, and outputs the offshore oilfield power load prediction value;

[0078] The above moving pooling graph neural network includes a moving window layer for capturing local spatiotemporal dependencies and a time pooling layer for extracting high-level features. First, a sliding window of a specific size M is selected, which moves with a stride s. Figure 4 This process is visually demonstrated in the figure, where a connected graph with four segments is shown, each containing four tables. In this example, a window of size 2 is moved with a stride of 1, resulting in three windows. Each window contains two segments, each with four variables. A graph neural network is then used in each window. The graph neural network consists of two phases (propagation and update phase), in which information from neighboring nodes is propagated to the central node. Given the central node of the w-th layer window at layer l It has a set of neighboring nodes in the M segments of the same window The correlation between the central node and its neighbor nodes is After the propagation stage, the propagation characteristics are obtained Then, the update phase uses a nonlinear function to update the propagation features, namely In general, the graph neural network propagates variable information according to the correlation between all variables in M ​​segments, making it possible to fully capture the comprehensive spatiotemporal dependencies within the window to update variable features. The update phase introduces a nonlinear function to update the propagation characteristics of the variables, further enhancing the ability to learn effective representations;

[0079] After obtaining the updated propagation features, the time pooling operation is used to extract the high-level features of each window; specifically: given M segments of time series data, the updated propagation features of the i-th variable Use the average pooling strategy to perform temporal pooling and obtain the sensor features of the w-th window Subsequently, the output variable, i.e., the offshore oilfield power load forecast, is obtained by stacking across all windows.

[0080] The embodiment of the present invention also includes:

[0081] S6. Display the forecast value of offshore oil field power load through visualization tools to facilitate decision-making by platform managers.

[0082] Embodiment 2:

[0083] The embodiment of the present invention provides an ultra-short-term prediction system for offshore oilfield power load, applying the above method, the system comprises: a data acquisition module, a data processing module, a time sequence diagram construction module and a prediction module; wherein:

[0084] Data acquisition module, used to obtain relevant data of offshore oil fields in real time;

[0085] The data processing module is used to pre-process the offshore oil field related data; and divide the pre-processed offshore oil field related data into multiple time series data according to a preset sliding window;

[0086] A timing diagram construction module is used to construct a timing diagram based on multiple segments of timing data;

[0087] The prediction module is used to input the time series graph into the trained graph neural network and output the predicted value of offshore oil field power load.

[0088] The system also includes a display module; the display module is used to display the offshore oil field power load forecast value through a visualization tool.

[0089] In summary, the present invention provides an ultra-short-term prediction method and system for offshore oilfield power load, which can comprehensively consider various factors affecting offshore platform power load by combining load data and multivariate time series data. This method not only focuses on historical load data, but also integrates multiple variables such as environmental conditions, platform production activities, and equipment operating status, thereby improving the comprehensiveness and accuracy of the prediction.

[0090] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0091] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for ultra-short-term prediction of offshore oilfield power load, characterized in that: The steps include: S1. Obtain relevant data of offshore oil fields in real time; S2. Preprocessing the offshore oil field related data; S3, dividing the pre-processed offshore oilfield related data into multiple time series data segments according to a preset sliding window; S4, constructing a time series diagram based on the multiple time series data; S5. Input the time series diagram into the mobile pooling graph neural network, and output the predicted value of the offshore oil field power load.

2. The method for ultra-short-term prediction of offshore oilfield power load according to claim 1, characterized in that: The offshore oil field related data includes environmental data, production activity data and equipment status data.

3. The method for ultra-short-term prediction of offshore oilfield power load according to claim 1, characterized in that: The preprocessing includes cleaning outliers, normalizing and filling missing data.

4. The method for ultra-short-term prediction of offshore oilfield power load according to claim 1, characterized in that: The step S3 specifically includes: S31, encoding each segment of time series data to obtain variable node features; S32, based on the variable node features, adding relative position information to the variables between each segment of time series data by position embedding, to obtain updated variable node features; S33. Construct a time series diagram based on the updated variable node features.

5. The method for ultra-short-term prediction of offshore oilfield power load according to claim 1, characterized in that: The various timing features in the timing diagram are connected through a decreasing matrix; the decreasing matrix is ​​expressed as: Where C represents a decreasing matrix; t represents the tth time series data; r represents the rth time series data; c tr,ij Represents the element in the i-th row and j-th column in the decreasing matrix C; D represents the dimension of the time series data; K represents a total of K segments of time series data; i represents the i-th dimension; j represents the j-th dimension.

6. The method for ultra-short-term prediction of offshore oilfield power load according to claim 1, characterized in that: The mobile pooling graph neural network includes a mobile window layer and a time pooling layer; The moving window layer processes the time sequence diagram to obtain propagation characteristics of the neighboring node information in the process of being propagated to the central node; Updating the propagation characteristics to obtain updated propagation characteristics; Processing the update propagation features through the time pooling layer to obtain sensor features corresponding to each window; The sensor features corresponding to each window are stacked to output the offshore oilfield power load forecast value.

7. The method for ultra-short-term prediction of offshore oilfield power load according to claim 1, characterized in that: Also includes: S6. Displaying the offshore oil field power load forecast value through a visualization tool.

8. An ultra-short-term prediction system for offshore oilfield power load, characterized in that: Applying the method described in any one of claims 1 to 7 above, the system comprises: a data acquisition module, a data processing module, a timing diagram construction module and a prediction module; The data acquisition module is used to obtain relevant data of offshore oil fields in real time; The data processing module is used to pre-process the offshore oilfield related data; and divide the pre-processed offshore oilfield related data into multiple time series data segments according to a preset sliding window; The timing diagram construction module is used to construct a timing diagram based on the multiple segments of timing data; The prediction module is used to input the time series diagram into a trained graph neural network and output a predicted value of offshore oil field power load.

9. The offshore oilfield power load ultra-short-term prediction system according to claim 8, characterized in that: Also includes: Display module; The display module is used to display the offshore oil field power load forecast value through a visualization tool.

Images