An offshore oilfield water injection system energy saving optimization method and system

Abstract

The present application relates to the field of offshore oilfield water injection system operation control, and provides an offshore oilfield water injection system energy-saving optimization method and system. The method comprises the following steps: inputting fluid simulation working condition samples into an agent prediction model for training, and constructing a multi-dimensional reliable applicable domain; inputting control variables of a target working condition into the multi-dimensional reliable applicable domain for reliability verification, and inputting the control variables into the agent prediction model for prediction calculation to obtain a benchmark wellhead injection pressure set and a benchmark pump set power; taking the pump set outlet pressure as a starting point, performing monotonic step-by-step descending processing on the pump set outlet pressure, and performing hierarchical compensation adjustment on the branch valve opening to obtain a multi-layer pump valve collaborative candidate operation scheme set; performing intersection screening on the candidate schemes to obtain an effective candidate operation scheme set; and comprehensively evaluating and sorting the effective candidate operation scheme set to obtain an optimal pump set outlet pressure and an optimal branch valve opening setting value. The present application improves the control precision of pump valve collaborative optimization.

Description

Technical Field

[0001] This invention relates to the field of operation and control technology for offshore oilfield water injection systems, and in particular to an energy-saving optimization method and system for offshore oilfield water injection systems. Background Technology

[0002] Offshore oilfield water injection systems utilize injection pump sets to pressurize water and distribute it to each injection well via trunk pipelines and branch regulating valves. Within this system, a significant nonlinear coupling relationship exists between pump set outlet pressure, branch valve opening, single-well water distribution, and wellhead injection pressure. The setting of the pump set outlet pressure directly affects the overall system energy consumption and the pressure distribution status of each injection well. To reduce the computational overhead of online optimization, some methods employ surrogate prediction models trained on simulated operating data to establish a mapping relationship between control variables and wellhead injection pressure and pump power, thereby reducing the number of calls to the mechanistic simulation model.

[0003] However, existing optimization methods for water injection systems based on surrogate prediction models typically only use the range of values ​​of a single variable as the boundary of model use, lacking credibility constraints on the distribution structure of the input conditions at the training sample level. This results in a high risk of prediction distortion for surrogate models under conditions outside the coverage of the training samples, and the generated optimization schemes lack engineering credibility, thereby affecting the safety and stability of on-site control execution. Summary of the Invention

[0004] This invention provides an energy-saving optimization method and system for offshore oilfield water injection systems to overcome the shortcomings of existing technologies.

[0005] This invention provides an energy-saving optimization method for offshore oilfield water injection systems, comprising: S1: Input the fluid simulation condition samples into the surrogate prediction model for training to obtain the surrogate prediction model, and construct a multi-dimensional reliable application domain for the surrogate prediction model based on the fluid simulation condition samples. S2: Input the control variables of the target working condition into the multi-dimensional reliable application domain for reliability verification, and input the target working condition input vector that has passed the verification into the proxy prediction model for prediction calculation to obtain the benchmark wellhead injection pressure set and benchmark pump group power. S3: Starting from the pump group outlet pressure in the target working condition input vector, the pump group outlet pressure is monotonically and gradually decreased according to a preset step size. Under each pump pressure level, the branch valve opening is graded and compensated based on the wellhead injection pressure margin of multiple injection wells to obtain a set of multi-layer pump valve collaborative candidate operation schemes. S4: Simultaneously input multiple candidate schemes from the multi-layer pump-valve collaborative candidate operation scheme set into the equipment physical operation boundary constraints and the multi-dimensional reliable applicable domain constraints for intersection screening, and obtain a set of effective candidate operation schemes by intersection screening; S5: Based on pressure safety constraints and energy-saving constraints, the set of effective candidate operation schemes is comprehensively evaluated and ranked to obtain the optimal pump set outlet pressure and the optimal branch valve opening setting value.

[0006] According to the present invention, an energy-saving optimization method for an offshore oilfield water injection system is provided, wherein step S1 further includes: S11: Input the fluid simulation working condition sample into the surrogate prediction model for training to obtain the surrogate prediction model describing the nonlinear mapping relationship between the control variables of the water injection system and the wellhead injection pressure and pump power; S12: Extract the maximum and minimum values ​​of multiple univariate parameters in the fluid simulation working condition sample to obtain the set of univariate parameter value boundaries; S13: Using the fluid simulation working condition samples as a reference set, calculate the distance metric between multiple target input vectors and training samples in the reference set to obtain the local neighborhood consistency index. S14: Based on the distribution state of the fluid simulation working condition samples, the sample distribution density in each input space region is statistically analyzed to obtain a sample distribution density distribution map; S15: The multidimensional reliable application domain is determined by the set of univariate parameter value boundaries, the local neighborhood consistency index, and the sample distribution density distribution map.

[0007] According to the present invention, an energy-saving optimization method for offshore oilfield water injection system is provided. In step S1, the surrogate prediction model is based on a pre-trained sequence model. The surrogate prediction model is obtained by fine-tuning the base model using a parameter-efficient fine-tuning method.

[0008] According to the energy-saving optimization method for offshore oilfield water injection systems provided by the present invention, the reliability verification performed in step S2 includes: S21: Compare each control variable of the target working condition with the set of single-variable parameter value boundaries in the multidimensional trusted application domain one by one. If any parameter exceeds the corresponding boundary range, generate a verification report containing the parameter name, current input value, upper and lower boundary limits and the direction of exceeding the limit, and terminate the subsequent optimization process. S22: For the target operating condition control variables that have passed the boundary set verification of single-variable parameter values, supplement the fixed pipeline parameters, fluid property parameters and equipment structural parameters to obtain the target operating condition input vector.

[0009] According to the energy-saving optimization method for offshore oilfield water injection systems provided by the present invention, the graded compensation adjustment in step S3 includes: S31: The difference between the predicted wellhead injection pressure of each injection well under each pump pressure level and the corresponding minimum allowable injection pressure is used to obtain the wellhead injection pressure margin of each injection well. S32: Based on the calculated wellhead injection pressure margin, prioritize each injection well from smallest to largest to obtain the injection well priority ranking result; S33: Combining the local impact of pump set outlet pressure changes on multiple branch valves, the opening degree of the corresponding branch valves is adjusted sequentially according to the priority ranking of the injection wells to obtain the branch valve opening degree compensation scheme under the current pump pressure level. S34: Combine the current pump pressure level with the branch valve opening compensation scheme to obtain a set of candidate operation schemes for multi-level pump and valve coordination.

[0010] According to the energy-saving optimization method for offshore oilfield water injection systems provided by the present invention, the specific method of compensation adjustment in step S33 is as follows: Priority is given to increasing the opening of the branch valves corresponding to water injection wells with smaller pressure margins and higher sensitivity to changes in pump outlet pressure in the priority ranking results. After each valve opening adjustment, the adjusted control variables are re-input into the proxy prediction model for prediction calculation. The prediction calculation results are used to update the wellhead injection pressure and pump power of the current candidate scheme and serve as the input for the next adjustment.

[0011] According to the energy-saving optimization method for offshore oilfield water injection systems provided by the present invention, the generation process of the multi-layer pump-valve collaborative candidate operation scheme set in step S3 adopts a fast heuristic optimization algorithm, which specifically includes: At each pump pressure level, the branch valve opening is locally expanded within the current state neighborhood according to a preset disturbance step size to obtain multiple valve disturbance candidate combinations; for each valve disturbance candidate combination, the proxy prediction model is called in real time to perform prediction calculation, and the prediction results are evaluated and filtered in real time according to the comprehensive evaluation function. During the continuous optimization process, if several consecutive candidate expansions fail to improve the comprehensive evaluation function value, an early stop mechanism is triggered to terminate the candidate expansion process at the current pump pressure level.

[0012] According to the present invention, in the method for energy-saving optimization of offshore oilfield water injection system, in step S5, when comprehensively evaluating and ranking the set of effective candidate operation schemes based on pressure safety constraints and energy-saving constraints, the energy-saving constraints are based on the power of the benchmark pump group, and the pressure safety constraints are based on the set of benchmark wellhead injection pressures.

[0013] According to the energy-saving optimization method for offshore oilfield water injection systems provided by the present invention, the constraint mode for the comprehensive evaluation ranking in step S5 includes: When the target operating condition has a corresponding fluid simulation reference result, the verification mode is entered. The verification mode specifically involves: calculating the relative deviation of the predicted wellhead injection pressure of multiple candidate schemes in the effective candidate operating scheme set relative to the fluid simulation reference result, and eliminating candidate schemes whose relative deviation is greater than a first preset threshold. When there is no corresponding fluid simulation reference result for the target operating condition, the deployment mode is entered. The deployment mode specifically involves: calculating the average relative change of the predicted wellhead injection pressure of multiple candidate schemes in the effective candidate operating scheme set relative to the benchmark wellhead injection pressure set, and eliminating candidate schemes whose average relative change is greater than a second preset threshold.

[0014] The present invention also provides an energy-saving optimization system for an offshore oilfield water injection system, used to execute an energy-saving optimization method for an offshore oilfield water injection system as described in any of the above claims, comprising: Training module: used to input fluid simulation condition samples into the surrogate prediction model for training, obtain the surrogate prediction model, and construct a multi-dimensional reliable application domain for the surrogate prediction model based on the fluid simulation condition samples; Calculation module: used to input the control variables of the target working condition into the multi-dimensional reliable application domain for reliability verification, input the target working condition input vector that has passed the verification into the proxy prediction model for prediction calculation, and obtain the benchmark wellhead injection pressure set and benchmark pump group power; Adjustment module: It is used to take the pump group outlet pressure in the target working condition input vector as the starting point, and perform monotonically decreasing process on the pump group outlet pressure according to the preset step size. Under each pump pressure level, the branch valve opening is adjusted in a graded compensation manner based on the wellhead injection pressure margin of multiple water injection wells to obtain a set of multi-layer pump valve collaborative candidate operation schemes. The filtering module is used to simultaneously input multiple candidate schemes from the multi-layer pump-valve collaborative candidate operation scheme set into the physical operation boundary constraints of the equipment and the multi-dimensional reliable applicability domain constraints for intersection filtering, and obtain a set of effective candidate operation schemes by intersection filtering; Output module: Used to comprehensively evaluate and rank the set of effective candidate operation schemes based on pressure safety constraints and energy-saving constraints, and obtain the optimal pump set outlet pressure and the optimal branch valve opening setting value.

[0015] This invention provides an energy-saving optimization method and system for offshore oilfield water injection systems. By constructing a multi-dimensional reliable applicability domain encompassing univariate parameter value boundaries, local neighborhood consistency, and sample distribution density, the reliable applicability boundary of the surrogate prediction model is expanded from a traditional single-range constraint to a structured multi-level constraint space. This simultaneously limits the extrapolation risk of the model outside the training distribution in both the target operating condition verification and candidate scheme generation stages, effectively reducing the probability of instability in field execution due to model prediction distortion. Secondly, this invention uses a step-by-step decrease in pump outlet pressure as the main path, combined with a branch valve graded compensation adjustment mechanism based on wellhead injection pressure margin. This establishes a synergistic relationship between pump pressure regulation and valve control, avoiding the injection instability problem caused by insufficient pressure margin in local injection wells under traditional decoupled control methods. The overall pressure coordination capability of the multi-well system is significantly improved. Furthermore, this invention also uses the intersection of equipment physical operation boundary constraints and multi-dimensional reliable applicability domain constraints for screening, ensuring that candidate operating schemes are guaranteed in both model prediction reliability and engineering executability, fundamentally reducing the probability of unexecutable schemes entering the evaluation and ranking stage. Furthermore, this invention effectively compresses the candidate search space by combining a fast heuristic optimization algorithm with a finite trial strategy and a continuous no-improvement early stop mechanism, significantly improving the computational efficiency of online optimization and meeting the real-time control response requirements of offshore oilfield water injection systems. The invention also introduces a dual-mode stability evaluation mechanism, enabling it to output optimal pump outlet pressure and branch valve opening setpoints with pressure stability guarantees under both simulated reference results and no-reference results, thus broadening the method's engineering applicability. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This invention provides a schematic flowchart of an energy-saving optimization method for offshore oilfield water injection systems. Figure 2 This invention provides a schematic diagram of an energy-saving optimization system structure for an offshore oilfield water injection system. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, embodiments of this invention, and should not be construed as limiting the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention. In the description of this invention, it should be understood that the terminology used is for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0019] The embodiments of the present invention are described below with reference to the figures.

[0020] like Figure 1 As shown, the present invention provides an energy-saving optimization method for offshore oilfield water injection systems, comprising: S1: Input the fluid simulation condition samples into the surrogate prediction model for training to obtain the surrogate prediction model, and construct a multi-dimensional reliable application domain for the surrogate prediction model based on the fluid simulation condition samples.

[0021] This invention first establishes a fluid simulation model corresponding to the topology of an offshore oilfield water injection system. This simulation model is built upon the actual pipeline network structure, pipeline length, diameter, wall thickness, fluid properties, pump performance parameters, valve parameters, and inlet boundary conditions. For the established fluid simulation model, this invention generates operating condition sample data by changing the water distribution of individual wells, the pump set outlet pressure, and the branch valve opening. The variable inputs for the simulation operating conditions include the water distribution of six individual wells, the boosting parameters of multiple injection pumps, and the openings of multiple branch valves. The total water distribution of the system is obtained by summing the water distribution of each individual well, and the pump set outlet pressure is determined by the pump inlet reference pressure and the pump boosting parameters. After each set of variable inputs is combined with fixed pipeline network parameters, the simulation model is run to obtain the injection pressure at the wellhead of multiple wells, the outlet pressure of each pump, the power of each pump, the total power of the pump set, and the pressure and flow data of selected nodes. Finally, the operating conditions that converge and meet the basic operating conditions are retained as valid samples, forming a fluid simulation operating condition sample set.

[0022] Step S1 further includes: S11: Input the fluid simulation working condition sample into the surrogate prediction model for training to obtain a surrogate prediction model that describes the nonlinear mapping relationship between the control variables of the water injection system and the wellhead injection pressure and pump power; wherein, the surrogate prediction model is based on the pre-trained sequence model, and the surrogate prediction model is obtained by fine-tuning the basic model using an efficient parameter fine-tuning method.

[0023] After obtaining the fluid simulation working condition sample set, in step S11, this invention converts each valid working condition into a structured sequence sample with a fixed field order. The structured sequence includes, in sequence, an input identifier, an input field name and corresponding engineering value, an output identifier, an output field name and corresponding engineering value. Each field is encoded in the form of "field name = normalized value", and adjacent fields are separated by a preset separator. The normalization process is based on a linear mapping of the maximum and minimum values ​​of each field in the training sample set, mapping the original engineering quantities to a unified numerical range. The normalization parameters are saved along with the sample set.

[0024] Subsequently, this invention constructs a surrogate prediction model based on a pre-trained sequence model. The pre-trained sequence model is a neural network model that has undergone parameter training on a large-scale general text corpus and possesses sequence understanding and generation capabilities. Internally, it uses a multi-layer self-attention mechanism to weighted aggregate features at each position in the input sequence, thereby capturing the dependencies between different positions within the sequence. Based on this, this invention employs LoRA for efficient parameter fine-tuning. Specifically, the core processing of LoRA involves paralleling a pair of low-rank decomposition matrices next to each target weight matrix of the base model. During training, only the parameters of this pair of low-rank matrices are updated, while the original weights of the base model remain frozen. Specifically, for a target weight matrix in the base model, LoRA introduces two low-rank matrices whose ranks are much smaller than the dimensions of the original matrix. During training, the gradient only flows through these two low-rank matrices, thus significantly reducing the number of parameters that need to be updated. During fine-tuning, this invention organizes structured sequence samples according to the fixed field order and encoding format of the training phase and inputs them into the model. After receiving the input field sequence, the model generates the corresponding output field sequence at the output end, namely the normalized encoding of the predicted injection pressure values ​​at each wellhead and the predicted total power values ​​of the pump group. After training, this invention saves the basic model identifier, LoRA adaptation parameters, field order, sequence template, normalization parameters, and fixed parameter template. During inference, the input structured sequence is organized according to the same field order and encoding format as in the training phase. After parsing the model output, it undergoes inverse normalization to obtain the wellhead injection pressure engineering quantity and the total pump power engineering quantity, which are the surrogate prediction model's prediction outputs for the current input conditions. Furthermore, the validation set does not participate in fine-tuning but is used to calculate the average relative error, maximum relative error, and error distribution of the wellhead injection pressure and total pump power, serving as an evaluation basis for the model's prediction performance.

[0025] S12: Extract the maximum and minimum values ​​of multiple univariate parameters in the fluid simulation working condition sample to obtain the set of univariate parameter value boundaries.

[0026] After the surrogate prediction model is trained, in step S12, this invention constructs a multi-dimensional reliable applicability domain based on the fluid simulation working condition sample set. Specifically, firstly, the set of univariate parameter value boundaries is extracted. For each input field in the fluid simulation working condition sample set, this invention iterates through all valid samples and extracts the maximum and minimum values ​​of the field in all samples to form the value boundary of the field. After repeating the above operation for all input fields, the upper and lower boundaries corresponding to each field are summarized to obtain the set of univariate parameter value boundaries. This set records the coverage range of the training samples in each input dimension. Subsequently, whenever a control variable enters the optimization process, it must be compared with this set field by field.

[0027] S13: Using the fluid simulation working condition samples as a reference set, calculate the distance metric between multiple target input vectors and training samples in the reference set to obtain a local neighborhood consistency index.

[0028] In step S13, this invention calculates a local neighborhood consistency index. Specifically, when the distance between a target input vector and its nearest neighbors in the training sample set exceeds a certain range, it indicates that the target condition falls into a sparse region of the training samples, and the prediction reliability of the surrogate prediction model decreases in this region. The specific processing involves using all valid samples in the fluid simulation condition sample set as a reference set. For each target input vector to be verified, the distance metric between it and all training samples in the reference set is calculated. Subsequently, this invention performs distance measurement, calculated in a normalized input space to eliminate the influence of differences in the dimensions of different fields. Specifically, this invention extracts the distance values ​​corresponding to the nearest neighbors of the target input vector in the reference set. These distance values ​​are statistically processed to obtain the local neighborhood consistency index. The larger the index value, the higher the degree to which the target condition deviates from the center of the training sample set, and it will be assigned a higher unreliability risk weight in the subsequent application domain determination.

[0029] S14: Based on the distribution state of the fluid simulation working condition samples, the sample distribution density in each input space region is statistically analyzed to obtain a sample distribution density distribution map.

[0030] After obtaining the local neighborhood consistency index, this invention further statistically analyzes the sample distribution density of the fluid simulation working condition sample set in each region of the input space. Specifically, the invention divides the normalized input space into several sub-regions, counts the number of training samples falling into each sub-region, and calculates the sample density value within that sub-region. Based on this, the invention repeats the above operation for all sub-regions to obtain a sample distribution density map covering the entire input space. Regions with lower distribution density indicate sparse coverage of training samples in that input space location, and the interpolation accuracy of the surrogate prediction model in that region is relatively low.

[0031] S15: The multidimensional reliable application domain is determined by the set of univariate parameter value boundaries, the local neighborhood consistency index, and the sample distribution density distribution map.

[0032] After obtaining the set of univariate parameter value boundaries, the local neighborhood consistency index, and the sample distribution density map, step S15 combines these three to jointly determine the multidimensional reliable applicability domain. Specifically, the constructed multidimensional reliable applicability domain consists of three layers of constraints: the first layer is the hard boundary defined by the set of univariate parameter value boundaries, where any input field exceeding the corresponding upper or lower boundary is directly determined to be outside the domain; the second layer is the neighborhood structure constraint based on the local neighborhood consistency index, where any case where the distance between the target input vector and the nearest sample in the reference set exceeds a preset neighborhood threshold is determined to be outside the domain; the third layer is the density constraint based on the sample distribution density map, where any case where the sample density of the sub-region into which the target input vector falls is lower than a preset density threshold is determined to be outside the domain. For the constructed applicability domain, only when the target input vector simultaneously satisfies the above three layers of constraints is it determined to be within the multidimensional reliable applicability domain, allowing it to be called and calculated by the subsequent proxy prediction model.

[0033] S2: Input the control variables of the target working condition into the multidimensional reliable application domain for reliability verification, and input the target working condition input vector that has passed the verification into the proxy prediction model for prediction calculation to obtain the benchmark wellhead injection pressure set and benchmark pump group power.

[0034] The credibility verification performed in step S2 includes: S21: Compare each control variable of the target working condition with the set of single-variable parameter value boundaries in the multidimensional trusted application domain one by one. If any parameter exceeds the corresponding boundary range, generate a verification report containing the parameter name, current input value, upper and lower boundary limits and the direction of exceeding the limit, and terminate the subsequent optimization process.

[0035] Furthermore, the target operating conditions are provided by the front-end interface, database, or industrial data interface. The user inputs the required water distribution volume for each injection well, the current pump set outlet pressure, and the current opening degree of each branch valve. After receiving the above input, this invention first calculates the target total water distribution volume based on the sum of the water distribution volumes of each single well, and generates pump boosting parameters by combining the pump inlet reference pressure, forming a set of control variables for the target operating conditions to be verified.

[0036] Subsequently, the present invention compares the value of each field in the set of control variables with the upper and lower boundaries of the corresponding field in the set of univariate parameter value boundaries one by one. The comparison process is performed sequentially by field: for each field, it is determined whether its current input value falls within the closed interval enclosed by the lower and upper limits of the corresponding training data.

[0037] If all fields pass the comparison, the target working condition control variable set passes the single variable parameter value boundary verification and proceeds to step S22; if the current input value of any field is lower than the lower limit of the training data or higher than the upper limit of the training data, the present invention generates a verification report. The verification report records the parameter name that failed the verification, the current input value, the lower limit of the training data, the upper limit of the training data, and the direction of exceeding the limit, and terminates the current optimization process.

[0038] S22: For the target operating condition control variables that have passed the boundary set verification of single-variable parameter values, supplement the fixed pipeline parameters, fluid property parameters and equipment structural parameters to obtain the target operating condition input vector.

[0039] In step S22, the target operating condition control variable, which passed the boundary set verification of the single-variable parameter values ​​in step S21, is concatenated and completed with fixed pipeline parameters, fluid property parameters, and equipment structural parameters. The fixed pipeline parameters include pipeline physical parameters that remain unchanged during optimization, such as pipeline length, diameter, and wall thickness; fluid property parameters include fluid characteristic parameters such as the density and viscosity of the injected water; and equipment structural parameters include inherent equipment parameters such as pump performance curve parameters and valve flow coefficients. These fixed parameters were determined and saved during the offline modeling stage and are directly read during target operating condition processing. Subsequently, the target operating condition control variable and the above three types of fixed parameters are concatenated according to the same field order as in the training stage to obtain a complete target operating condition input vector. This input vector is then converted into a structured sequence according to the same encoding format as in the training stage and input into the surrogate prediction model for prediction calculation, obtaining the set of predicted injection pressure values ​​for each injection wellhead and the predicted total power value of the pump group, i.e., the set of benchmark wellhead injection pressure and benchmark pump group power.

[0040] S3: Starting from the pump group outlet pressure in the target working condition input vector, the pump group outlet pressure is monotonically decreased step by step according to a preset step size. Under each pump pressure level, the branch valve opening is adjusted in stages based on the wellhead injection pressure margin of multiple injection wells to obtain a set of multi-layer pump and valve collaborative candidate operation schemes.

[0041] After completing the target operating condition verification and benchmark prediction in step S2, this invention uses the pump set outlet pressure in the target operating condition input vector as the search starting point and performs a monotonically decreasing process on the pump set outlet pressure according to a preset step size. Each decrease in step size forms a new pump pressure level. Under the premise that the water demand of each injection well remains unchanged, each pump pressure level corresponds to a new set of control variable states. Under these states, this invention sequentially performs wellhead injection pressure margin calculation, injection well priority ranking, and branch valve opening grade compensation adjustment, generating candidate operating schemes layer by layer, and finally summarizing to obtain a multi-layer pump-valve coordinated candidate operating scheme set.

[0042] The hierarchical compensation adjustment in step S3 includes: S31: The difference between the predicted wellhead injection pressure of each injection well under each pump pressure level and the corresponding minimum allowable injection pressure is used to obtain the wellhead injection pressure margin of each injection well.

[0043] At each pump pressure level, this invention first encodes the control variables corresponding to the current pump pressure level (including the current pump set outlet pressure, the water distribution of each injection well, and the current branch valve opening) into a structured sequence according to the same field order as in the training phase. This sequence is then input into the surrogate prediction model for prediction calculation, yielding the predicted wellhead injection pressure values ​​for each of the six injection wells at the current level. Subsequently, for each injection well, this invention subtracts its predicted wellhead injection pressure value from the corresponding minimum allowable injection pressure. The minimum allowable injection pressure is the lower limit of pressure required to meet the injection task of that well, determined by the water injection system engineering parameters and pre-stored. The value obtained after subtraction is the wellhead injection pressure margin of that well, representing the remaining space between the current predicted injection pressure and the minimum allowable pressure. After performing the above subtraction operation on each of the six injection wells, this invention finally obtains a set of wellhead injection pressure margins for each injection well, containing six margin values.

[0044] S32: Based on the calculated wellhead injection pressure margin, prioritize each injection well from smallest to largest to obtain the injection well priority ranking result.

[0045] After obtaining the set of injection pressure margins at the wellheads of each injection well, this invention sorts the six injection wells in ascending order of their injection pressure margin values. The sorting result directly reflects the urgency of the current pressure status of each injection well: the injection well with the smallest margin value is closest to the minimum allowable injection pressure, has the highest pressure risk, and is ranked first in the priority sequence; the injection well with the largest margin value has the most sufficient pressure margin and is ranked last in the priority sequence. After sorting, the priority ranking result of the injection wells is obtained. This result is an ordered list recording the order of the six injection wells from highest to lowest pressure risk and their corresponding branch valve numbers.

[0046] S33: Based on the local impact of pump set outlet pressure changes on multiple branch valves, the opening degree of the corresponding branch valves is adjusted sequentially according to the priority ranking of the injection wells to obtain the branch valve opening degree compensation scheme under the current pump pressure level.

[0047] The specific method of compensation adjustment in step S33 is as follows: priority is given to increasing the opening of the branch valves corresponding to the injection wells with small pressure margins and high sensitivity to changes in pump outlet pressure in the priority ranking results of the injection wells; after each valve opening adjustment, the adjusted control variables are re-input into the proxy prediction model for prediction calculation, and the prediction calculation results are used to update the wellhead injection pressure and pump power of the current candidate scheme, and used as the input for the next adjustment.

[0048] After obtaining the priority ranking of injection wells, in step S33, this invention combines the local impact of pump set outlet pressure changes on each branch valve to perform graded compensation adjustments on the branch valve openings. The local impact of pump set outlet pressure changes on each branch valve refers to the change in wellhead injection pressure in the pipeline where each branch valve is located when the pump pressure drops. This impact is determined by the pipeline network topology and the resistance characteristics of each branch pipeline, and is reflected in the prediction results of the proxy prediction model under the current operating conditions.

[0049] This invention prioritizes the branch valves corresponding to the injection wells with the smallest pressure margin and the highest sensitivity to changes in pump outlet pressure, based on the injection well priority ranking. For each branch valve, the invention increases the valve opening by a preset disturbance step size, generating a set of valve disturbance candidate combinations. Then, the invention re-encodes the complete control variables containing the adjusted branch valve openings into a structured sequence according to field order, inputs it into a proxy prediction model for prediction calculation, and obtains the predicted wellhead injection pressure and pump power of each injection well after adjustment. The invention replaces the corresponding values ​​in the current candidate schemes with the prediction calculation results as the input state for the next valve adjustment. Subsequently, the above opening adjustment and prediction calculation process is repeated for the branch valves corresponding to the next injection well according to the injection well priority ranking, until all branch valves requiring compensation under the current pump pressure level are adjusted. After all branch valves are adjusted, the invention finally obtains a branch valve opening compensation scheme for the current pump pressure level, which records the adjusted opening values ​​of each of the six branch valves.

[0050] After each valve opening adjustment and prediction calculation is completed, this invention immediately performs constraint judgment on the current candidate scheme. The judgment includes: whether the predicted wellhead injection pressure of each injection well meets the pressure safety constraint, whether each control variable is within the physical operating boundary of the equipment, and whether the current combination of control variables is within the multidimensional reliable application domain. Candidate schemes that only meet all of the above constraints enter the next round of adjustment iteration, while candidate schemes that do not meet the constraints are directly eliminated.

[0051] S34: Combine the current pump pressure level with the branch valve opening compensation scheme to obtain a set of candidate operation schemes for multi-level pump and valve coordination.

[0052] After the candidate expansion process for the current pump pressure level is completed, this invention combines the current pump pressure level value with the branch valve opening compensation schemes retained after real-time evaluation and screening at that level to form a candidate operation scheme record for the current level. This record includes the current pump set outlet pressure value, the adjusted opening values ​​of each of the six branch valves, the predicted injection pressure values ​​at the wellheads of each injection well corresponding to the proxy prediction model, and the predicted pump set power values. This invention repeats the above processing steps S31 to S33 and the candidate scheme recording operation for each level during the pump pressure drop process, sequentially merging the candidate operation schemes generated at each level into the candidate scheme record list, ultimately obtaining a multi-layer pump-valve collaborative candidate operation scheme set covering all pump pressure levels.

[0053] In step S3, the generation process of the multi-layer pump-valve collaborative candidate operation scheme set adopts a fast heuristic optimization algorithm. Specifically, the fast heuristic optimization algorithm is as follows: at each pump pressure level, the branch valve opening is locally expanded within the current state neighborhood according to a preset disturbance step size to obtain multiple valve disturbance candidate combinations; for each valve disturbance candidate combination, the proxy prediction model is called in real time to perform prediction calculation, and the prediction results are evaluated and screened in real time according to the comprehensive evaluation function; during the continuous optimization process, when several consecutive candidate expansions fail to improve the comprehensive evaluation function value, an early stop mechanism is triggered to terminate the candidate expansion process at the current pump pressure level.

[0054] Furthermore, the aforementioned branch valve graded compensation adjustment process employs a fast heuristic optimization algorithm. At each pump pressure level, this invention performs local candidate expansion of the branch valve opening within the current state's neighborhood according to a preset disturbance step size. Specifically, for the current valve opening state, a preset positive or negative disturbance step size is applied to each branch valve, generating multiple sets of valve disturbance candidate combinations. Each set of valve disturbance candidate combinations corresponds to a complete set of control variable states, including the pump group outlet pressure at the current pump pressure level and the disturbed opening values ​​of each branch valve.

[0055] For each candidate combination of valve disturbances, this invention encodes it into a structured sequence and inputs it into a surrogate prediction model for prediction calculation, obtaining the predicted injection pressure and pump power of each injection well corresponding to that candidate combination. Subsequently, this invention substitutes the prediction results into a comprehensive evaluation function for real-time evaluation.

[0056] Specifically, the comprehensive evaluation function takes the comprehensive index of pressure deviation of each injection well and the power consumption of the pump set as inputs, and outputs a comprehensive evaluation value after weighting and superimposing according to preset weight coefficients. The comprehensive index of pressure deviation is calculated by the degree of deviation of the predicted injection pressure of each candidate wellhead from the minimum allowable injection pressure; the greater the deviation, the worse the comprehensive evaluation value. After completing the prediction calculation and comprehensive evaluation for each group of valve disturbance candidate combinations, this invention compares the current comprehensive evaluation value with the recorded current best comprehensive evaluation value. If the current value is better than the recorded best value, the best value and the corresponding candidate scheme are updated.

[0057] During the continuous optimization process, this invention tracks the changes in the comprehensive evaluation function value within consecutive candidate expansion cycles. When several consecutive candidate expansions fail to improve the comprehensive evaluation function value relative to the current optimal value, this invention triggers an early stop mechanism, terminating the candidate expansion process at the current pump pressure level and ceasing to generate new valve disturbance candidate combinations for the current level. The specific number of consecutive cycles in the early stop mechanism is determined by a preset early stop threshold, which is pre-configured in the system parameters.

[0058] S4: Simultaneously input multiple candidate schemes from the multi-layer pump-valve collaborative candidate operation scheme set into the physical operation boundary constraints of the equipment and the multi-dimensional reliable applicable domain constraints for intersection screening, and obtain a set of effective candidate operation schemes by intersection screening.

[0059] After obtaining the set of candidate operation schemes for multi-layer pump-valve coordination, in step S4, the present invention performs two types of constraint screening on each candidate scheme simultaneously. The first type is the physical operation boundary constraint of the equipment: checking whether the pump group outlet pressure in the candidate scheme is within the range of the minimum and maximum allowable outlet pressure of the pump group, whether the opening degree of each branch valve is within the range of the minimum and maximum allowable valve opening degree, and whether the adjustment amount of each parameter exceeds the upper limit of the single adjustment range allowed on site. The second type is the multi-dimensional credible application domain constraint: the control variables of the candidate scheme are combined and input into the multi-dimensional credible application domain, and the single variable parameter value boundary set comparison, local neighborhood consistency index judgment, and sample distribution density map density judgment are performed in sequence. Only when all three judgments are passed is it considered to satisfy the application domain constraint. Based on the above two constraints, the present invention only retains candidate schemes that pass both types of constraints, and obtains the set of effective candidate operation schemes by intersection screening.

[0060] S5: Based on pressure safety constraints and energy-saving constraints, the set of effective candidate operation schemes is comprehensively evaluated and ranked to obtain the optimal pump set outlet pressure and the optimal branch valve opening setting value.

[0061] In step S5, when comprehensively evaluating and ranking the set of effective candidate operation schemes based on pressure safety constraints and energy-saving constraints, the energy-saving constraints are based on the power of the benchmark pump group, and the pressure safety constraints are based on the set of benchmark wellhead injection pressures.

[0062] The constraint mode for the comprehensive evaluation ranking in step S5 includes: When the target operating condition has a corresponding fluid simulation reference result, the verification mode is entered. The verification mode specifically involves: calculating the relative deviation of the predicted wellhead injection pressure of multiple candidate schemes in the effective candidate operating scheme set relative to the fluid simulation reference result, and eliminating candidate schemes whose relative deviation is greater than a first preset threshold. When there is no corresponding fluid simulation reference result for the target operating condition, the deployment mode is entered. The deployment mode specifically involves: calculating the average relative change of the predicted wellhead injection pressure of multiple candidate schemes in the effective candidate operating scheme set relative to the benchmark wellhead injection pressure set, and eliminating candidate schemes whose average relative change is greater than a second preset threshold.

[0063] In step S5, the present invention performs a comprehensive evaluation and ranking of each candidate scheme based on pressure safety constraints and energy-saving constraints. Specifically, the energy-saving constraint is handled as follows: the benchmark pump power obtained in step S2 is read, and the predicted pump power value of each candidate scheme in the effective candidate operation scheme set is compared with the benchmark pump power. Candidate schemes with predicted pump power values ​​not lower than the benchmark pump power are eliminated, and only candidate schemes with predicted pump power values ​​lower than the benchmark pump power are retained. The pressure safety constraint is handled as follows: the benchmark wellhead injection pressure set obtained in step S2 is read, and the predicted wellhead injection pressure of each injection well in each candidate scheme is compared with the value of the corresponding well in the benchmark wellhead injection pressure set. Combining the upper limit of the allowable maximum injection pressure and the lower limit of the allowable minimum injection pressure, candidate schemes where the predicted wellhead injection pressure of any injection well exceeds the allowable range are eliminated.

[0064] Based on the above constraint screening, the present invention performs stability evaluation by entering either verification mode or deployment mode depending on whether there is a corresponding fluid simulation reference result for the target working condition.

[0065] In verification mode, a corresponding fluid simulation reference result exists for the target operating condition. This invention predicts the wellhead injection pressure for each injection well in the set of valid candidate operating schemes, and calculates the relative deviation between the predicted wellhead injection pressure and the corresponding wellhead injection pressure value in the fluid simulation reference result for each well. The relative deviation is calculated by dividing the difference between the predicted value and the reference value by the absolute value of the reference value. This invention takes the maximum value of the relative deviation for each well. If the maximum relative deviation exceeds a first preset threshold, the candidate scheme is eliminated.

[0066] In deployment mode, there is no corresponding fluid simulation reference result for the target working condition. This invention predicts the wellhead injection pressure of each injection well for each candidate scheme, calculates the relative change of each well relative to the corresponding well value in the set of reference wellhead injection pressures, and averages the relative changes of all injection wells to obtain the average relative change. If the average relative change exceeds the second preset threshold, the candidate scheme is eliminated.

[0067] After the above-mentioned screening, the present invention sorts the remaining candidate schemes from best to worst according to the comprehensive evaluation function value, and takes the candidate scheme ranked first to output its pump group outlet pressure and the opening value of each branch valve, so as to obtain the optimal pump group outlet pressure and the optimal branch valve opening setting value.

[0068] like Figure 2 As shown, the present invention also provides an energy-saving optimization system for offshore oilfield water injection systems, comprising: Training module 100: used to input fluid simulation condition samples into the surrogate prediction model for training, obtain the surrogate prediction model, and construct a multi-dimensional reliable application domain for the surrogate prediction model based on the fluid simulation condition samples; Calculation module 200: is used to input the control variables of the target working condition into the multi-dimensional reliable application domain for reliability verification, and input the target working condition input vector that has passed the verification into the proxy prediction model for prediction calculation to obtain the set of benchmark wellhead injection pressure and benchmark pump group power; Adjustment module 300: is used to take the pump group outlet pressure in the target working condition input vector as the starting point, perform monotonically decreasing process on the pump group outlet pressure according to a preset step size, and perform graded compensation adjustment on the branch valve opening based on the wellhead injection pressure margin of multiple injection wells under each pump pressure level to obtain a set of multi-layer pump valve collaborative candidate operation schemes. Filtering module 400: is used to simultaneously input multiple candidate schemes from the multi-layer pump-valve collaborative candidate operation scheme set into the equipment physical operation boundary constraints and the multi-dimensional reliable applicability domain constraints for intersection filtering, and obtain a set of effective candidate operation schemes by intersection filtering; Output module 50: is used to comprehensively evaluate and rank the set of effective candidate operation schemes based on pressure safety constraints and energy-saving constraints, so as to obtain the optimal pump set outlet pressure and the optimal branch valve opening setting value.

[0069] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0070] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0071] To verify the effectiveness of the proposed method, typical offshore oilfield water injection system operating conditions were selected for testing, and the results of mechanism simulation were used as a comparison benchmark. Test results show that the average relative error of the surrogate prediction model of this invention on the validation set for each wellhead injection pressure does not exceed 0.30%, the average relative error of the total pump power does not exceed 0.06%, and the maximum relative error is controlled within an acceptable range. The prediction accuracy meets the engineering requirements for online optimization. Regarding energy saving, compared with traditional operating schemes, the optimal pump outlet pressure and branch valve opening settings output by this invention can reduce the total power of the pump group. The expected energy saving rate is calculated as the ratio of the difference between the candidate scheme and the benchmark pump power, and the energy saving effect is significant under typical test conditions. In terms of computational efficiency, the fast heuristic optimization algorithm of this invention, combined with an early stopping mechanism, significantly reduces the number of candidate searches compared to the global exhaustive search method, and the online optimization calculation time meets the response requirements of real-time control. Regarding pressure safety, in the optimal scheme output after double constraint screening, the injection pressure at the wellhead of all water injection wells meets the minimum allowable injection pressure requirement, and no injection instability occurs.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for energy-saving optimization of offshore oilfield water injection systems, characterized in that, include: S1: Input the fluid simulation condition samples into the surrogate prediction model for training to obtain the surrogate prediction model, and construct a multi-dimensional reliable application domain for the surrogate prediction model based on the fluid simulation condition samples. S2: Input the control variables of the target working condition into the multi-dimensional reliable application domain for reliability verification, and input the target working condition input vector that has passed the verification into the proxy prediction model for prediction calculation to obtain the benchmark wellhead injection pressure set and benchmark pump group power. S3: Starting from the pump group outlet pressure in the target working condition input vector, the pump group outlet pressure is monotonically and gradually decreased according to a preset step size. Under each pump pressure level, the branch valve opening is graded and compensated based on the wellhead injection pressure margin of multiple injection wells to obtain a set of multi-layer pump valve collaborative candidate operation schemes. S4: Simultaneously input multiple candidate schemes from the multi-layer pump-valve collaborative candidate operation scheme set into the equipment physical operation boundary constraints and the multi-dimensional reliable applicable domain constraints for intersection screening, and obtain a set of effective candidate operation schemes by intersection screening; S5: Based on pressure safety constraints and energy-saving constraints, the set of effective candidate operation schemes is comprehensively evaluated and ranked to obtain the optimal pump set outlet pressure and the optimal branch valve opening setting value.

2. The energy-saving optimization method for offshore oilfield water injection systems according to claim 1, characterized in that, Step S1 further includes: S11: Input the fluid simulation working condition sample into the surrogate prediction model for training to obtain the surrogate prediction model describing the nonlinear mapping relationship between the control variables of the water injection system and the wellhead injection pressure and pump power; S12: Extract the maximum and minimum values ​​of multiple univariate parameters in the fluid simulation working condition sample to obtain the set of univariate parameter value boundaries; S13: Using the fluid simulation working condition samples as a reference set, calculate the distance metric between multiple target input vectors and training samples in the reference set to obtain the local neighborhood consistency index. S14: Based on the distribution state of the fluid simulation working condition samples, the sample distribution density in each input space region is statistically analyzed to obtain a sample distribution density distribution map; S15: The multidimensional reliable application domain is determined by the set of univariate parameter value boundaries, the local neighborhood consistency index, and the sample distribution density distribution map.

3. The energy-saving optimization method for offshore oilfield water injection systems according to claim 1, characterized in that, The surrogate prediction model in step S1 is based on a pre-trained sequence model, and the surrogate prediction model is obtained by fine-tuning the base model using an efficient parameter fine-tuning method.

4. The energy-saving optimization method for offshore oilfield water injection systems according to claim 1, characterized in that, The credibility verification performed in step S2 includes: S21: Compare each control variable of the target working condition with the set of single-variable parameter value boundaries in the multidimensional trusted application domain one by one. If any parameter exceeds the corresponding boundary range, generate a verification report containing the parameter name, current input value, upper and lower boundary limits and the direction of exceeding the limit, and terminate the subsequent optimization process. S22: For the target operating condition control variables that have passed the boundary set verification of single-variable parameter values, supplement the fixed pipeline parameters, fluid property parameters and equipment structural parameters to obtain the target operating condition input vector.

5. The energy-saving optimization method for an offshore oilfield water injection system according to claim 1, characterized in that, The graded compensation adjustment in step S3 includes: S31: The difference between the predicted wellhead injection pressure of each injection well under each pump pressure level and the corresponding minimum allowable injection pressure is used to obtain the wellhead injection pressure margin of each injection well. S32: Based on the calculated wellhead injection pressure margin, prioritize each injection well from smallest to largest to obtain the injection well priority ranking result; S33: Combining the local impact of pump set outlet pressure changes on multiple branch valves, the opening degree of the corresponding branch valves is adjusted sequentially according to the priority ranking of the injection wells to obtain the branch valve opening degree compensation scheme under the current pump pressure level. S34: Combine the current pump pressure level with the branch valve opening compensation scheme to obtain a set of candidate operation schemes for multi-level pump and valve coordination.

6. The energy-saving optimization method for an offshore oilfield water injection system according to claim 5, characterized in that, The specific method of compensation adjustment in step S33 is as follows: Priority is given to increasing the opening of the branch valves corresponding to water injection wells with smaller pressure margins and higher sensitivity to changes in pump outlet pressure in the priority ranking results. After each valve opening adjustment, the adjusted control variables are re-input into the proxy prediction model for prediction calculation. The prediction calculation results are used to update the wellhead injection pressure and pump power of the current candidate scheme and serve as the input for the next adjustment.

7. The energy-saving optimization method for an offshore oilfield water injection system according to claim 1, characterized in that, The generation process of the multi-layer pump-valve collaborative candidate operation scheme set in step S3 adopts a fast heuristic optimization algorithm, which specifically includes: At each pump pressure level, the branch valve opening is locally expanded within the current state neighborhood according to a preset disturbance step size to obtain multiple valve disturbance candidate combinations; for each valve disturbance candidate combination, the proxy prediction model is called in real time to perform prediction calculation, and the prediction results are evaluated and filtered in real time according to the comprehensive evaluation function. During the continuous optimization process, if several consecutive candidate expansions fail to improve the comprehensive evaluation function value, an early stop mechanism is triggered to terminate the candidate expansion process at the current pump pressure level.

8. The energy-saving optimization method for an offshore oilfield water injection system according to claim 7, characterized in that, In step S5, when comprehensively evaluating and ranking the set of effective candidate operation schemes based on pressure safety constraints and energy-saving constraints, the energy-saving constraints are based on the power of the benchmark pump group, and the pressure safety constraints are based on the set of benchmark wellhead injection pressures.

9. The energy-saving optimization method for an offshore oilfield water injection system according to claim 1, characterized in that, The constraint mode for the comprehensive evaluation ranking in step S5 includes: When the target operating condition has a corresponding fluid simulation reference result, the verification mode is entered. The verification mode specifically involves: calculating the relative deviation of the predicted wellhead injection pressure of multiple candidate schemes in the effective candidate operating scheme set relative to the fluid simulation reference result, and eliminating candidate schemes whose relative deviation is greater than a first preset threshold. When there is no corresponding fluid simulation reference result for the target operating condition, the deployment mode is entered. The deployment mode specifically involves: calculating the average relative change of the predicted wellhead injection pressure of multiple candidate schemes in the effective candidate operating scheme set relative to the benchmark wellhead injection pressure set, and eliminating candidate schemes whose average relative change is greater than a second preset threshold.

10. An energy-saving optimization system for an offshore oilfield water injection system, used to execute the energy-saving optimization method for an offshore oilfield water injection system as described in any one of claims 1-9, characterized in that, include: Training module: used to input fluid simulation condition samples into the surrogate prediction model for training, obtain the surrogate prediction model, and construct a multi-dimensional reliable application domain for the surrogate prediction model based on the fluid simulation condition samples; Calculation module: used to input the control variables of the target working condition into the multi-dimensional reliable application domain for reliability verification, input the target working condition input vector that has passed the verification into the proxy prediction model for prediction calculation, and obtain the benchmark wellhead injection pressure set and benchmark pump group power; Adjustment module: It is used to take the pump group outlet pressure in the target working condition input vector as the starting point, and perform monotonically decreasing process on the pump group outlet pressure according to the preset step size. Under each pump pressure level, the branch valve opening is adjusted in a graded compensation manner based on the wellhead injection pressure margin of multiple water injection wells to obtain a set of multi-layer pump valve collaborative candidate operation schemes. The filtering module is used to simultaneously input multiple candidate schemes from the multi-layer pump-valve collaborative candidate operation scheme set into the physical operation boundary constraints of the equipment and the multi-dimensional reliable applicability domain constraints for intersection filtering, and obtain a set of effective candidate operation schemes by intersection filtering; Output module: Used to comprehensively evaluate and rank the set of effective candidate operation schemes based on pressure safety constraints and energy-saving constraints, and obtain the optimal pump set outlet pressure and the optimal branch valve opening setting value.

Images