Seismic data processing

Abstract

Multicomponent seismic data 1a and fibre optic seismic data 1b is provided for a region. The data is input 2 into the neural network to train it to predict multicomponent seismic data from fibre optic

Description

The present invention relates to the field of seismic data processing, for example for permanent reservoir monitoring and / or monitoring below the sea bed. It is known to perform permanent reservoir monitoring by acquiring multicomponent ocean bottom seismic data with multicomponent sensors. However, such sensors are expensive. According to a first aspect, there is provided a method of training a neural network to predict seismic data, e.g. conventional seismic data, from fibre optic seismic data, the method comprising: providing first multicomponent seismic data and first fibre optic seismic data for a region; and inputting the first multicomponent seismic data and the first fibre optic seismic data into the neural network and training the neural network to predict seismic data from fibre optic seismic data based on the inputted first multicomponent seismic data and first fibre optic seismic data. As such, a neural network may be provided that can predict seismic data from fibre optic seismic data. The predicted seismic data (that can be provided by the trained neural network) is preferably multicomponent seismic data. However, in some cases single component (e.g. pressure) seismic data may be predicted. This may depend on the seismic data used to train the neural network for example, and / or, on (later) seismic data input into the trained neural network. Fibre optic seismic data can be much cheaper to obtain than multicomponent seismic data. However, fibre optic seismic data typically only contains seismic strain data in one dimension (measured along the length of a fibre optic cable). As such, it is not always as useful as multicomponent seismic data. By providing a neural network that can predict seismic data from fibre optic seismic data, this means that, e.g. in the future, (more useful) multicomponent seismic data may be predicted from (cheaper) fibre optic seismic data, without the requirement that multicomponent sensors be provided. In some cases, for example, one or more multicomponent sensors may be provided in a region in order to measure multicomponent seismic data with which to train a neural network (e.g. as described above). However, they could then be recovered and potentially reused elsewhere. In other cases, for example where there is / are already one or more multicomponent seismic sensors provided, e.g. for permanent reservoir monitoring, a trained neural network could be used to provide predicted (e.g. multicomponent) seismic data from fibre optic seismic data in situations such as when one or more multicomponent seismic sensors fails. In some cases, a relatively coarsely sampled set of (e.g. conventional) seismic data plus fibre optic data may be used to train a neural network. Then, the neural network trained with such data may be used to predict more finely sampled seismic data along a fibre optic cable. Multicomponent seismic data may mean seismic data with two or more components. For example, such data may comprise one or more variables that may be measured in two or more dimensions / directions and / or it could comprise pressure data as well as an additional measurement. Multicomponent seismic data may comprise particle velocity and / or particle acceleration data measured in one or more dimensions / directions. However, such data may not necessarily be measured directly (and this may depend on the type of multicomponent sensor used to measure the data, such as whether a hydrophone, seismometer (geophone) and / or an accelerometer is used). Rather, in some cases other data may be measured and then processed (e.g. passed through one or more algorithms) in order to determine such velocity and / or acceleration data from the measured multicomponent data. For example, if a hydrophone is used as a multicomponent sensor, it would measure fluid pressure. A seismometer is a type of geophone that is especially designed for seismology (low frequencies) or passive monitoring (very sensitive). If a seismometer / geophone is used as a multicomponent sensor, it would measure ground motion or particle velocity. The measured data could then be processed to determine, e.g. particle, acceleration data. Accelerometers are typically solid-state sensors. If an accelerometer is used it would measure particle acceleration, which could then be processed to determine particle velocity data. As such, in some cases the first multicomponent seismic data used in the method may be (e.g. raw, unprocessed) measured multicomponent seismic data, e.g. as measured by the multicomponent sensor(s). In other cases, such data may have undergone (initial) processing, e.g. to apply one or more corrections for example to balance source variability and / or to remove noise, before it is used as the first multicomponent seismic data in the method. In other cases, (e.g. raw, unprocessed) measured multicomponent seismic data may have undergone processing such that multicomponent particle velocity and / or acceleration data is derived from the (e.g. raw, unprocessed) measured multicomponent seismic data and then used as the first multicomponent seismic data in the method. The first multicomponent seismic data may comprise pre- and / or post-stack multicomponent seismic data, optionally having undergone some further processing. Multicomponent seismic data may be measured with one or more multicomponent seismic sensors. Such sensors may comprise more than one hydrophone, seismometer (geophone) and / or accelerometer. For example, 3C (three component) seismic data may be measured with a multicomponent seismic sensor comprising three orthogonally oriented geophones or accelerometers. 4C (four component) seismic data may be measured with a multicomponent seismic sensor comprising a hydrophone as well as three orthogonally oriented geophones or accelerometers. Multicomponent seismic sensors may be arranged on a sea bed, on (a component(s) of) an (a subsea) installation, and / or in a well, for example. Multicomponent seismic data may be acquired, for example, by placing measurement systems (receivers) with several sensors on the sea floor in a two-dimensional pattern or along a borehole in a one-dimensional pattern. At a surface, active seismic sources may be used to create seismic waves, which are recorded by the measurement systems. This typically creates a 5D (or 4D) data cube of recorded multicomponent seismic data, where two dimensions are determined by the location of the source, two (or one) are determined by the location of the receiver(s), and one is determined by time. Fibre optic seismic data may mean seismic data measured with (or derived from data measured with) (using) one or more fibre optic cables, for example as part of a distributed acoustic sensing (DAS) system. Such a cable may be used to detect and measure data (only) in one dimension, e.g. along the length of the cable. DAS is based on measurements of the backscatter of laser light in the fibre optic cable. This type of measurement is very sensitive to strain variations along the fibre optic cable. These measurements can be used to estimate strain variations and thus (i.e. hence derive) particle velocities along the fibre optic cable. The seismic data that is detected and measured using a fibre optic cable (e.g. DAS data) may be data from which seismic (inline) strain data may be determined or estimated. In other words, seismic strain data may not be measured directly with a fibre optic cable. However, the data measured using a fibre optic cable may be used to determine or estimate seismic (inline) strain data. For example, data measured using a fibre optic cable may be processed (e.g. by applying one or more normalisation, filtering, resampling, correction and / or noise removal algorithms) and used to determine or estimate seismic (inline) strain data. In some cases, (further) processing (e.g. normalisation, filtering and / or resampling) of the estimated seismic (inline) strain data may be required, e.g. to make it suitable for input into a neural network. Such processing and / or determination or estimation of seismic (inline) strain data from the data measured using a fibre optic cable may be performed by one or more processors connected to the fibre optic cable, e.g. in a DAS system. The one or more processors may be provided in a box or module connected to the cable, for example. As such, the one or more fibre optic cables may be provided as part of a DAS system, e.g. specifically provided for this purpose. Alternatively, the one or more fibre optic cables may be or comprise data or telecommunication cables. One or more interrogator boxes or modules may be connected to such cables. This may enable such data or telecommunication fibre optic cables to be used to detect and measure seismic data. The one or more fibre optic cables may be arranged on a sea bed, either on the surface of the seabed or in a trench, and / or, in some cases, in a well, and / or on (a component(s) of) an (a subsea) installation. The one or more fibre optic cables (e.g. one fibre optic cable with an interrogator) may cover a distance of around 150 km, for example. The one or more fibre optic cables may be arranged in a line(s), grid, or in a coil. Ideally, the one or more fibre optic cables are arranged such that they cover a two-dimensional area, e.g. of the sea bed. However, in some cases, a single cable in a straight line (i.e. covering just one dimension) may be sufficient. Preferably, the multicomponent sensors (or at least one or more of the multicomponent sensors) are collocated with the fibre optic cable(s). For example, the multicomponent sensors, or at least one or more of the multicomponent sensors, may be located at a position or positions along or (immediately) adjacent to the fibre optic cable(s). As such, the (first) multicomponent seismic data and the first fibre optic seismic data for may be for (or correspond to) the same or at least partially overlapping regions. In some cases, the first multicomponent seismic data and / or the first fibre optic seismic data may be transformed spatially (e.g. if the multicomponent sensors and fibre optic cable(s) are not (or not completely) collocated), such that the first multicomponent seismic data and the first fibre optic seismic data are spatially aligned. This may create a dataset of labels or a labelled dataset comprising the (transformed) first multicomponent seismic data and / or the first fibre optic seismic data. Before inputting the first multicomponent seismic data and the first fibre optic seismic data into the neural network, the first multicomponent seismic data and the first fibre optic seismic data are preferably each split into two-dimensional slices of data. For example, a two-dimensional slice of multicomponent seismic data may comprise a varying time and varying of one source and / or receiver parameter. Varying of a source and / or receiver parameter may comprise varying a parameter describing or indicating the location(s) of source or receiver positions. The best or preferred way of splitting the first multicomponent seismic data and the first fibre optic seismic data into two-dimensional slices of data may depend on the application (i.e. the purpose for which the eventual data or trained neural network is intended, and / or the method of data acquisition, for example. Alternatively, in some cases it may not be necessary to split the first multicomponent seismic data and the first fibre optic seismic data into two-dimensional slices. For example, the first multicomponent seismic data and the first fibre optic seismic data may not be sliced at all or they may be split into subsets with more than two dimensions. The neural network may be or comprise a direct conversion neural network or an indirect conversion neural network. For example, if the neural network comprises a direct conversion neural network, this could comprise a convolutional neural network (CNN) or a vision transformer (ViT). With a CNN, this type of neural network may be configured to extract feature maps from the fibre optic data. From this latent space, a seismic reflectivity image may be reconstructed, preferably with convolution weights trained to minimize a loss function that measures a distance between a generated and an actual seismic reflectivity image. With a ViT, the input fibre optic data may be divided into fixed-size patches, each of which may be treated as a separate token. These patches may then be linearly embedded and fed into a transformer architecture, which processes the patches to produce a final feature map. Compared to traditional CNNs, which use convolutional layers to extract spatial features from the image, ViT may use attention mechanisms to model interpatch relationships and capture global context information. One advantage of using a ViT is that it may capture long-range dependencies and global context information, which may be useful in style transfer tasks where the overall style of the image needs to be preserved. If the neural network comprises an indirect conversion neural network, this could comprise a generative adversarial neural network (GAN). GANs are a type of neural network architecture that is composed of two sub-networks: a generator and a discriminator. In the present invention, the generator may create images that mimic actual seismic data conditioned by the input from the input data, while the discriminator may attempt to determine whether the result is generated or from the training set, effectively evaluating the quality of these images and providing feedback to the generator. The two networks may be trained in parallel, e.g. in an adversarial fashion. Preferably, the neural network is trained using the (labelled) first multicomponent seismic data and first fibre optic seismic data. This may comprise backpropagating any loss(es) and / or adjusting weights of the network(s) to minimize a difference between predicted and known seismic data, e.g. between a predicted seismic reflectivity provided from the neural network and a known seismic reflectivity provided from the first multicomponent seismic data. The method may further comprise inputting second multicomponent seismic data and second fibre optic seismic data for the region into the neural network; and checking predictions of seismic data from fibre optic seismic data by the neural network based on the inputted second multicomponent seismic data and second fibre optic seismic data. As such, the performance of the neural network may be evaluated, e.g. during training, for example with multicomponent and fibre optic data reserved for that purpose. By “second multicomponent seismic data” and “second fibre optic seismic data” is simply meant seismic and fibre optic seismic data that is not the exact same data as the first multicomponent seismic data and first fibre optic seismic data. The second multicomponent seismic data and second fibre optic seismic data could, for example, and preferably do, come from the same multicomponent and fibre optic data sets (e.g. the same vintages) as the first multicomponent seismic data and first fibre optic seismic data, respectively. The second multicomponent seismic data and second fibre optic seismic data do not necessarily need to be (and preferably are not) data (or based on / derived from data) that are recorded at a different time to the first multicomponent seismic data and first fibre optic seismic data. In some cases, a multicomponent seismic data set (e.g. a vintage of multicomponent seismic data) may be provided. This may consist of multicomponent seismic data measured in a single survey, i.e. at substantially the same time. This data set (or vintage) may be split into two or more subsets to provide (e.g. at least) first and second multicomponent seismic data sets. In some cases, the first and / or second multicomponent seismic data sets (each) comprise less than half, less than a quarter, or around 10% of the (main) multicomponent seismic data set from which it (they) is (are) derived. However, this may depend on the variation within the (main) multicomponent seismic data set and it may be adapted by controlling the norm(s) used in the validation (checking predictions). Similarly, in some cases, a fibre optic seismic data set (e.g. a vintage of fibre optic seismic data) may be provided. This may consist of fibre optic seismic data measured in a single survey, i.e. at substantially the same time. This data set (or vintage) may be split into two or more subsets to provide (e.g. at least) first and second fibre optic seismic data sets. As such, a single multicomponent seismic data set may be provided (and e.g. split) for both training the neural network and then checking the performance of the neural network (e.g. its accuracy at predicting seismic data from fibre optic seismic data). In addition or alternatively, a single fibre optic seismic data set may be provided (and e.g. split) for both training the neural network and then checking the performance of the neural network (e.g. its accuracy at predicting seismic data from fibre optic seismic data). For example, second fibre optic data may be input into a trained neural network and the neural network may predict (corresponding) (e.g. multicomponent) seismic data based on the input second fibre optic data. This predicted (e.g. multicomponent) seismic data based on the second fibre optic data may then be compared to the second multicomponent seismic data to determine how accurate the neural network’s predictions are, and / or to assess the performance of the neural network. How accurate the predictions need or are desired to be may vary according to the case and the data available. According to a further aspect, there is provided a method of predicting seismic data from fibre optic seismic data for a region. The method may comprise: providing a neural network for predicting seismic data from fibre optic seismic data for the region, wherein the neural network is trained according to the method described herein (with any of its optional or preferred features); and inputting third fibre optic seismic data for the region into the neural network and running the neural network to predict seismic data from the third fibre optic seismic data for the region. As such, (preferably multicomponent) seismic data may be predicted from (third) fibre optic seismic data using a neural network trained according to the method described herein (with any of its optional or preferred features). Although “third” fibre optic data is referred to here, this is merely to distinguish it from the “first” and (optional) “second” fibre optic data referred to above. As the second fibre optic data is optional, it need not necessarily be provided and thus in some cases only “first” and “third” fibre optic data may be provided. However, it is preferred that in most cases “second” fibre optic data is provided and used for checking / assessing the accuracy of the trained neural network. The predicted seismic data could have one or more components. For example, in some cases, just single component (e.g. pressure) data may be predicted. However, in other cases, multicomponent seismic data may be predicted. The third fibre optic seismic data may have been measured at a time or times later than a time(s) or period at / during which the first (and optional second) fibre optic seismic data was measured, e.g. when multicomponent seismic data is no longer available, at least in part. The third fibre optic seismic data may correspond to a different geographic region to that of the first (and optional second) fibre optic data. The first and / or second multicomponent seismic data may be pre-processed, e.g. before performing steps described above, for example to apply one or more corrections and / or to remove or reduce any noise in the data. The first, second and / or third fibre optic seismic data may be pre-processed, e.g. before performing steps described above, for example to apply one or more corrections and / or to remove or reduce any noise in the data. The predicted seismic data may be processed such that it can be used in further analysis of interpretation. Such processing may depend on the particular application. According to a further aspect, a method of evaluating or monitoring the status of a region is provided. The method may comprise: providing predicted seismic data for the region, the predicted seismic data being predicted according to the method described herein (with any of its optional or preferred features); and analysing the predicted seismic data to evaluate or monitor the status of the region. A result or results of analysing the predicted seismic data to evaluate or monitor the status of the region may comprise, for example, a (e.g. three-dimensional) map or image of a subsurface region and its structure, data about any changes in reservoir volume, area, boundary positions, pressure, and / or fluid composition(s) etc., geological composition data about the status of any foundations in the region, e.g. if any cracking or movement has occurred. Such a method may further comprise displaying at least one result of analysing the predicted multicomponent seismic data graphically, e.g. on one or more screens. The method may further comprise finding and / or monitoring, based on at least one result of analysing the predicted multicomponent seismic data: hydrocarbons in the region; foundations in the region; locations or possible locations of carbon storage in the region; an overburden in the region. The method may further comprise making a decision, based on at least one result of analysing the predicted multicomponent seismic data, about: whether or where to drill for hydrocarbons in the region; whether or from where to produce hydrocarbons in the region; the status of any foundations in the region and whether any repair, maintenance or reinforcement is needed or desirable; whether and where to perform carbon storage in the region; whether / where / when / how strongly an earthquake may be likely to occur; whether / where there is cracking in an overburden; and / or how to monitor a carbon storage unit and / or overburden. The method may then further comprise: drilling for hydrocarbons in the region; producing hydrocarbons from the region; injecting one or more fluids such as water or CO2 into the region, e.g. to facilitate (further) hydrocarbon production, repairing, maintaining and / or reinforcing any foundations in the region; and / or performing carbon storage in the region, evacuating a region and / or addressing any cracking in the overburden. According to a further aspect, there is provided a computer program product comprising computer-readable instructions that, when run on one or more processors or a computer, cause the one or more processors or computer to perform any of the methods described herein (e.g. for training a neural network and / or predicting seismic data from fibre optic seismic data) with any of their optional or preferred features. According to a further aspect, there is provided a system for acquiring seismic data for use in training a neural network. The system comprises: one or more multicomponent seismic sensors; one or more fibre optic cables; and one or more seismic sources. The one or more multicomponent seismic sensors and the one or more fibre optic cables are arranged to detect seismic signals emitted by the one or more seismic sources to thereby obtain multicomponent seismic data and fibre optic seismic data, respectively The system may further comprise at least one processing means for processing the multicomponent seismic data and fibre optic seismic data to train the neural network to predict seismic data from fibre optic seismic data. The at least one processing means may be arranged to perform the method as described herein (e.g. for training a neural network), with any of its optional or preferred features. According to a further aspect, there is provided a system for providing predicted seismic data. The system comprises: one or more fibre optic cables; one or more seismic sources; and one or more processing means. The one or more fibre optic cables are arranged to detect seismic signals emitted by the one or more seismic sources to thereby obtain fibre optic seismic data; and the one or more processing means are arranged to predict seismic data from the fibre optic seismic data by inputting the fibre optic seismic data into a neural network. The neural network may have been trained according to the method for training a neural network described herein, with any of its optional or preferred features. Preferred embodiments of the invention will now be described with reference to the accompanying figures, in which: Fig. 1 is a flow diagram illustrating a method of processing seismic data; and Fig. 2 is a schematic diagram illustrating a system for acquiring seismic data. As illustrated in Fig. 1, a method for processing seismic data for permanent reservoir monitoring comprises the following steps: 1a - providing first multicomponent seismic data for a region; 1b - providing first seismic data for the region measured with one or more fibre optic cables (this will be referred to as fibre optic seismic data); 2 - inputting the first multicomponent seismic data and the first fibre optic seismic data into a neural network and training the neural network to predict multicomponent seismic data from fibre optic seismic data based on the inputted first multicomponent seismic data and first fibre optic seismic data; 3 - inputting second multicomponent seismic data and second fibre optic seismic data for the region into the neural network; 4 - checking the neural network’s predictions of multicomponent seismic data from fibre optic seismic data based on the inputted second multicomponent seismic data and second fibre optic seismic data; 5 - inputting third fibre optic seismic data for the region into the neural network and running the neural network to predict multicomponent seismic data from the third fibre optic seismic data for the region; 6 - analysing the predicted multicomponent seismic data from the third fibre optic data to monitor the status of a reservoir; 7 - displaying the result(s) of step 6 graphically. These steps will now be described in more detail. At step 1a, first multicomponent seismic data for a region is provided. Multicomponent seismic data is measured with multicomponent seismic sensors arranged on the sea bed, in a well, or on an installation, for example. In some cases, the sensors are provided as part of a permanent reservoir monitoring system. The number of multicomponent sensors can vary quite a bit and may depend on factors (if applicable) such as size of reservoir and depth. Deeper locations may use fewer and / or more sparsely spread out sensors. For example, between 4 and 50 sensors per km2 may be provided. However, in some cases, the number may be smaller or greater than this. The multicomponent sensors or “nodes” may continuously or periodically (depending on the application, for example) detect and measure seismic data with two or more components. In addition, at step 1b, first seismic data for the region measured with one or more fibre optic cables is also provided. The one or more fibre optic cables are provided as part of a distributed acoustic sensing system and detect and measure seismic inline strain data (only) in one dimension, along the length of the cable. Alternatively, the one or more fibre optic cables are data or telecommunication cables to which an interrogator box has been connected. The one or more fibre optic cables are arranged on the sea bed, either on the surface of the seabed or in a trench, and / or, in some cases, in a well. The one or more fibre optic cables can cover a distance of around 150 km, for example. They may be arranged in a line(s), grid, or in a coil. Ideally, the one or more fibre optic cables are arranged such that they cover a two-dimensional area of the sea bed. However, in some cases, a single cable in a straight line (i.e. covering just one dimension) may be sufficient. The seismic data (multicomponent and / or fibre optic) is measured by emitting a seismic signal (wave) from a seismic source (e.g. an airgun). The seismic source is typically located on a boat, for example. The seismic signal is emitted towards the sea bed and reflects off different layers present in the subsurface. These reflections are then detected and measured by the multicomponent sensors and / or fibre optic cables to provide multicomponent seismic data and / or fibre optic seismic data, respectively. Here, the term “providing” seismic data could refer to measuring such seismic data or it could, and in most cases does, refer to just providing such (previously-measured or “vintage”) seismic data from another source, e.g. from a database or memory. At step 2, the first multicomponent seismic data and the first fibre optic seismic data are input into a neural network. The neural network could be a generative adversarial neural network, for example. The neural network is then trained to predict multicomponent seismic data from fibre optic seismic data based on the inputted first multicomponent seismic data and first fibre optic seismic data. In other words, previously-measured multicomponent and fibre optic seismic data can be used to teach a neural network to predict multicomponent seismic data from (future) fibre optic seismic data. At step 3, second multicomponent seismic data and second fibre optic seismic data for the region are input into the neural network. This is for checking the neural network’s predictions of multicomponent seismic data from fibre optic seismic data at step 4 below. These seismic data (the second multicomponent seismic data and second fibre optic seismic data for the region) are data that were recorded at the same time as the first multicomponent seismic data and first fibre optic seismic data. For example, prior to steps 1a and 1b, a (previously recorded) multicomponent seismic data set (vintage) could be split into (at least) the first multicomponent seismic data and second multicomponent seismic data. In addition, a (previously recorded) fibre optic data set (vintage) could be split into (at least) the first fibre optic seismic data and second fibre optic seismic data. Exactly how a (previously recorded) multicomponent seismic data set (vintage) is split into (at least) the first multicomponent seismic data and second multicomponent seismic data, e.g. the relative sizes of the first and second multicomponent seismic data sets, may be case dependent and depend, for example, on what it is desired to measure. As such, in some cases the first multicomponent seismic data set may be larger than the second and in other cases it may be smaller. In other cases, the first and second multicomponent seismic data sets may be the same size. At step 4, the neural network’s predictions of multicomponent seismic data from fibre optic seismic data are checked based on the inputted second multicomponent seismic data and second fibre optic seismic data. Specifically, the second fibre optic data is input into the trained neural network and the neural network predicts (corresponding) multicomponent seismic data based on the input second fibre optic data. The predicted multicomponent seismic data based on the second fibre optic data is then compared to the second multicomponent seismic data to determine how accurate the neural network’s predictions are. At this stage, if the accuracy of the predicted data is sufficient, a trained neural network for predicting multicomponent seismic data from fibre optic data has been created. It can then be used in the following steps to predict multicomponent seismic data from (future) fibre optic data. This can be useful in cases where multicomponent seismic data is not available, e.g. because a multicomponent sensor (node) has failed or simply because multicomponent sensors are not (or no longer) provided in a region. It may also / alternatively be useful as a relatively coarsely sampled set of (e.g. conventional) seismic data plus fibre optic data may be used to train a neural network, but the neural network trained with such data may be used to predict more finely sampled seismic data along a fibre optic cable. As such, at step 5, third fibre optic seismic data for the region is input into the neural network. The third fibre optic seismic data is (ideally) measured using the same fibre optic cable(s) and same, or same type of (e.g. vibrator, air gun etc.), seismic source as the first and second fibre optic seismic data. For example, the seismic source used to generate the third fibre optic data should ideally create the same kind of wave (e.g. a shear wave) as that created by the seismic source used to generate the first (and second) fibre optic seismic data, such that the wave interacts with the subsurface in substantially the same way. However, the third fibre optic seismic data is measured at a later time. The neural network is then run to predict multicomponent seismic data from the inputted third fibre optic data for the region. The predicted multicomponent seismic data from the third fibre optic data is outputted from the neural network. At step 6, the predicted multicomponent seismic data from the third fibre optic data is analysed, e.g. using known or previously established methods for analysing multicomponent seismic data, e.g. to monitor the status of the reservoir. At step 7, the results of the analysis (e.g. the status of the reservoir and its surroundings) may be displayed graphically, e.g. on a screen. Based on the status of the reservoir and its surroundings, a decision may be made for example with respect to future production and / or drilling in the reservoir. Next, (e.g. further) hydrocarbons may be produced from the reservoir and / or (e.g. further) drilling into the reservoir may be performed. Although the above description refers to the case of permanent reservoir monitoring, the method may also be used in other scenarios, such as monitoring below the sea bed beneath a wind farm or pipeline, monitoring below the sea bed used for carbon storage to detect CO2 leakage, passive earthquake monitoring, and / or monitoring any cracking in an overburden. In such cases, fibre optic cables may already be provided, e.g. to a number of wind turbines, for data or telecommunication. Such cables could also be used to measure fibre optic data for use in the present invention. In such cases, the output (at steps 6 and 7) may relate to the status of the seabed. Based on that output, a decision could be made about how safe any foundations are and whether any adjustments may be needed, for example. A decision could also be made about locations for carbon capture and storage, whether an earthquake is likely, where an earthquake is likely, how likely an earthquake may be to occur, how strong an earthquake may be, and / or whether and / or where there is any cracking in an overburden, for example. Fig. 2 is a schematic diagram illustrating a system for acquiring seismic data. This figure illustrates the scenario of acquiring seismic data for monitoring a hydrocarbon reservoir or CO2 storage unit 15, or for monitoring the overburden 16, e.g. for leakage detection from the CO2 storage. A fibre optic cable 12 is provided on the sea bed 10. An interrogator box (not shown) is connected to the fibre optic cable 12 to enable it to be able to measure seismic data. A number of nodes 13 (sensors) for measuring seismic data are provided at locations along the fibre optic cable 12. A vessel 14 on the sea surface 11 emits a seismic signal towards the sea bed 10, for example with an air gun (not shown). The seismic signal reflects off layers in the region 16 and from the reservoir or storage unit 15 under the sea bed and the reflected signals are detected and measured with the fibre optic cable 12 and the nodes 13. Fibre optic and seismic data collected in this way is then used to train a neural network (as described above) such that the neural network can predict corresponding multicomponent seismic data from fibre optic seismic data. Once a neural network has been trained in this way, if desired and / or practical, the nodes 13 may be removed and potentially used elsewhere. The remaining fibre optic cable 12 can then be used to collected further fibre optic data which can be input into the trained neural network such that the neural network can predict corresponding multicomponent seismic data from further fibre optic seismic data. The predicted multicomponent seismic data is then analysed using standard known techniques, for example, to monitor the status of the region 16 under the sea bed 10, a hydrocarbon reservoir 15 in that region 16, and / or a CO2 storage unit 15 in that region 16. As such, a fibre optic cable 12 can be used to obtain predicted multicomponent seismic data without the need for expensive multicomponent seismic sensors (nodes). 10

Claims

10 07 251520251. A method of training a neural network to predict seismic data from fibre optic seismic data, the method comprising:providing first multicomponent seismic data and first fibre optic seismic data for a region; andinputting the first multicomponent seismic data and the first fibre optic seismic data into the neural network and training the neural network to predict seismic data from fibre optic seismic data based on the inputted first multicomponent seismic data and first fibre optic seismic data.

2. A method as claimed in claim 1, wherein training the neural network to predict seismic data comprises training the neural network to predict multicomponent seismic data.

3. A method as claimed in claim 1 or 2, wherein the neural network is a generative adversarial neural network.

4. A method as claimed in claim 1, 2 or 3, further comprising: inputting second multicomponent seismic data and second fibre optic seismic data for the region into the neural network; and checking predictions of seismic data from fibre optic seismic data by the neural network based on the inputted second multicomponent seismic data and second fibre optic seismic data.

5. A method as claimed in claim 4, wherein:a multicomponent seismic data set is provided, the multicomponent seismic data set being split into at least the first multicomponent seismic data and the second multicomponent seismic data; and / or a fibre optic seismic data set is provided, the fibre optic seismic data set being split into at least the first fibre optic seismic data and the second fibre optic seismic data.

6. A method of predicting seismic data from fibre optic seismic data for a region, the method comprising:providing a neural network for predicting seismic data from fibre optic seismic data for the region, wherein the neural network is trained according to the method of any preceding claim; and10 07 25inputting third fibre optic seismic data for the region into the neural network and running the neural network to predict seismic data from the third fibre optic seismic data for the region.

7. A method as claimed in claim 6, wherein running the neural network to5 predict seismic data from the third fibre optic seismic data for the regioncomprises running the neural network to predict multicomponent seismic data from the third fibre optic seismic data for the region.

8. A method as claimed in claim 6 or 7, wherein the third fibre optic seismic data is measured at a time later than a time at which the first (and optional10 second) fibre optic seismic data was measured.

9. A method as claimed in any preceding claim, wherein any fibre optic seismic data is seismic data that has been measured with:one or more fibre optic cables arranged on a sea bed, in a well and / or on a subsea installation.15 10. A method as claimed in claim 9, wherein the one or more fibre optic cablesare data or telecommunication cables.

11. A method as claimed in claim 9 or 10, wherein the one or more fibre optic cables are arranged to cover a two-dimensional area.

12. A method of evaluating or monitoring the status of a region, the method20 comprising:providing predicted seismic data for the region, the predicted seismic data being predicted according to the method of any of claims 6-11;andanalysing the predicted seismic data to evaluate or monitor the25 status of the region.

13. A method as claimed in claim 12, further comprising displaying at least one result of analysing the predicted seismic data graphically.

14. A method as claimed in claim 12 or 13, further comprising making a decision, based on at least one result of analysing the predicted seismic30 data, about:whether or where to drill for hydrocarbons in the region;whether or from where to produce hydrocarbons in the region;the status of any foundations in the region and whether any repair, maintenance or reinforcement is needed;10 07 25whether and where to perform carbon capture and storage in the region;whether / where / when / how strongly an earthquake may be likely to occur; and / or5 whether / where there is cracking in an overburden.

15. A method as claimed in claim 14, further comprising: drilling for hydrocarbons in the region; producing hydrocarbons from the region; injecting one or more fluids into the region;10 repairing, maintaining and / or reinforcing any foundations in theregion;performing carbon capture and storage in the region;evacuating a region; and / or addressing any cracking in an overburden.15 16. A computer program product comprising computer-readable instructionsthat, when run on one or more processors or a computer, cause the one or more processors or computer to perform the method of any of claims 1-14.

17. A system for acquiring seismic data for use in training a neural network, the system comprising:20 one or more multicomponent seismic sensors;one or more fibre optic cables;one or more seismic sources; and at least one processing means; wherein: the one or more multicomponent seismic sensors and the one or25 more fibre optic cables are arranged to detect seismic signals emitted by theone or more seismic sources to thereby obtain multicomponent seismic data and fibre optic seismic data, respectively; andthe processing means is for the multicomponent seismic data and fibre optic seismic data to train the neural network to predict seismic data 30 from fibre optic seismic data.

18. A system as claimed in claim 17, wherein the at least one processing means is arranged to perform the method of any of claims 1-14.

19. A system for providing predicted seismic data, the system comprising: one or more fibre optic cables;35 one or more seismic sources; andone or more processing means; wherein the one or more fibre optic cables are arranged to detect seismic signals emitted by the one or more seismic sources to thereby obtain fibre optic seismic data; and 5 the one or more processing means are arranged to predict seismic data from the fibre optic seismic data by inputting the fibre optic seismic data into a neural network, the neural network having been trained according to the method of any of claims 1-5.10LO CXI

Images