Method for evaluating operation condition of multistage centrifugal water injection pump

Abstract

The invention belongs to the technical field of fluid machinery, and particularly relates to a method for evaluating the operation condition of a multistage centrifugal water injection pump. In order to solve the problem that in the prior art, the prediction precision of the operation condition of the multi-stage centrifugal pump is low, operation condition factor data of the multi-stage centrifugal water injection pump are obtained, wheel level factor data of the multi-stage centrifugal water injection pump are calculated, and the operation condition of the multi-stage centrifugal water injection pump is predicted according to various factor data influencing the operation condition of the water injection pump and result data. According to the method, the working condition early warning of the pump at a certain moment can be predicted through the method, the pump can be located on a certain stage of impeller, the problem can be found and solved before the impeller breaks down through maintenance and replacement in advance, the working condition of the water injection pump can be predicted, and the service life of the water injection pump can be prolonged. The maintenance cost and the failure rate are reduced, and the reliability and the stability of the multi-stage centrifugal pump are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of fluid machinery, and in particular relates to a method for evaluating the operating conditions of a multi-stage centrifugal water injection pump. Background Art

[0002] Multistage centrifugal pump is a common industrial application pump, mainly used to transport high-pressure fluid media, such as oil, natural gas, water, etc. Its main feature is the multi-impeller structure, which means that multiple impellers are installed inside the same pump body, and each impeller can increase a certain head and flow, thereby achieving a higher total head and total flow.

[0003] Traditional centrifugal pump models usually only consider the design and evaluation of a single impeller. Due to the special nature of the multi-impeller structure of a multi-stage centrifugal pump, although the impellers of each level are independent, they will also be affected by the previous and next impellers during operation. If a certain level of impeller of a multi-stage centrifugal pump is damaged, the entire pump will not necessarily stop operating immediately. However, the faulty impeller will reduce the performance of the pump, affect the volume flow and total head of the pump, and will also deteriorate the operating points of the remaining impellers, ultimately causing the pump to stop operating. Once the impeller of the pump fails, it is necessary to shut down the pump and disassemble and evaluate each level of the impeller during the maintenance process, which makes the maintenance cost high and the maintenance workload huge. The existing technology only predicts the status of the multi-stage centrifugal pump, and cannot effectively prevent the failure of the multi-stage centrifugal pump, resulting in low prediction accuracy. Summary of the invention

[0004] The purpose of the present invention is to provide a method for evaluating the operating condition of a multi-stage centrifugal water injection pump, so as to solve the problem that the prior art cannot predict the operating condition of each stage of the impeller of the multi-stage centrifugal pump, resulting in low prediction accuracy.

[0005] In order to solve the above technical problems, the present invention provides a method for evaluating the operating conditions of a multi-stage centrifugal water injection pump, comprising the following steps:

[0006] 1) Obtaining the operating condition factor data of the water injection pump and preprocessing it to obtain the sequence data of the impellers at each level; calculating the wheel level factor data between two adjacent impellers according to the sequence data of the impellers at each level, wherein the wheel level factor data represents the data of the operating state of the two adjacent impellers that affect each other; using the calculated wheel level factor data and the processed sequence data of the impellers at each level and the corresponding operating condition result data as training data;

[0007] 2) Construct a prediction model for predicting the operating conditions of impellers at each stage of the water injection pump based on the turbine factor data and sequence data of the water injection pump, train the prediction model using the training data to obtain a trained prediction model, and then use the trained prediction model to predict the water injection pump to be predicted, and obtain the operating conditions of each stage of impeller of the predicted water injection pump at a certain moment in the future.

[0008] The beneficial effects are as follows: in order to solve the problem of low accuracy in predicting the operating conditions of multi-stage centrifugal pumps in the prior art, the present invention obtains the operating data of the multi-stage centrifugal water injection pump and calculates its wheel stage factor data. According to the data of various factors and result data affecting the operating conditions of the water injection pump, a model for predicting the operating conditions of the multi-stage water injection pump is constructed, and finally the water injection pump to be predicted is predicted through the constructed prediction model. The present invention can predict the operating condition warning of the pump at a certain moment and locate it on any stage of the impeller through this method. Through early maintenance and replacement, the problem can be discovered and solved before the impeller fails, thereby reducing maintenance costs and failure rates and improving the reliability and stability of the multi-stage centrifugal pump.

[0009] Furthermore, the calculation of the impeller stage factor includes: the power of the interaction between two adjacent impeller stages, the difference percentage of the interaction between the efficiency of two adjacent impeller stages, and the height difference of the interaction between the heads of two adjacent impeller stages. The specific calculation formula is as follows:

[0010] The power calculation formula for the interaction between two adjacent impellers is:

[0011]

[0012] The calculation formula for the difference percentage of the interaction between the efficiency of two adjacent impeller stages is:

[0013]

[0014] The calculation formula for the height difference of the alternating lift of two adjacent impeller stages is:

[0015] H i =H n-1 -H n

[0016] Among them, P i is the power exchanged between two adjacent impeller stages; η i H is the difference percentage of the interaction between the efficiency of two adjacent impellers; i H is the height difference between the heads of two adjacent impellers; n is the head of the nth stage impeller; H n-1 is the head of the n-1th stage impeller; η n is the efficiency of the nth stage impeller; η n-1 is the efficiency of the n-1th stage impeller.

[0017] The beneficial effect is that taking into account multiple key factors and the structural reflection of the multi-stage impeller on the overall state of the multi-stage centrifugal water injection pump is conducive to improving the accuracy of the prediction.

[0018] Furthermore, the factor data for obtaining the operating conditions of the water injection pump include data collected by sensors: at least one of the output voltage, output current, given pressure, output frequency, motor speed, air cooler front temperature, air cooler middle temperature, air cooler rear temperature, front bearing temperature, rear bearing temperature, lubricating oil outlet temperature, motor electronics front temperature, motor electronics middle temperature, motor electronics rear temperature, cooling outlet temperature, cooling water inlet pressure, lubricating oil inlet pressure, pump lubricating oil inlet pressure, pump lubrication temperature, motor lubrication temperature, bearing front temperature, bearing rear temperature, pump inlet temperature, pump outlet temperature, liquid level, pump inlet pressure, pump outlet pressure, manifold pressure, pump dilution station pressure and motor dilution station pressure.

[0019] The beneficial effects are as follows: historical data of the operating conditions of the impellers at each stage of the water injection pump are collected. Since the historical data have an impact on the operating status of the impellers at each stage, a model established based on the data that have an impact on the operating conditions of the impellers at each stage can accurately predict the operating status of the impellers at each stage of the water injection pump, so as to find and solve the problem before the impeller fails, reduce maintenance costs and failure rates, and improve the reliability and stability of the multi-stage centrifugal pump.

[0020] Furthermore, the factors for obtaining the operating conditions of the water injection pump also include water quality survey data: PH index, S 2- content, suspended solid content, oil content in sewage, Fe 3+ At least one of the content, iron content, the number of bacterial SRB, the number of bacterial TGB and the number of bacterial IB.

[0021] The beneficial effect is that since the operating status of the impellers at each level of the water injection pump is also affected by the water quality, the operating status of the impellers at each level of the water injection pump is judged according to the above water quality data, making the judgment result more accurate.

[0022] Furthermore, the prediction model for predicting the operating conditions of the water injection pump adopts a multi-head self-attention mechanism model.

[0023] The beneficial effect is that the weight matrix generated by the attention mechanism can well display the hidden relationship between different interactions, thereby alleviating the poor interpretability problem brought by recurrent neural networks.

[0024] Furthermore, the preprocessing includes cleaning the acquired operating condition factor data to remove invalid information.

[0025] Its beneficial effects are: reducing the calculation of invalid information, saving calculation time and improving prediction efficiency.

[0026] Furthermore, the preprocessing also includes sorting the cleaned impeller sequence data according to certain data of the impeller.

[0027] Furthermore, the preprocessing also includes completing the remaining impeller sequence data based on the longest impeller sequence data in the sorted impeller sequence data.

[0028] The beneficial effect is that it is convenient to calculate the wheel stage factor data between two adjacent impeller stages.

[0029] Furthermore, the impeller sequence data is sorted according to the power size of the impeller. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a model framework diagram of the present invention;

[0031] Figure 2-1 It is a performance curve diagram of the first-stage impeller of the MD550-50 multistage pump of the present invention;

[0032] Figure 2-2 It is a performance curve diagram of the second stage impeller of the MD550-50 multistage pump of the present invention;

[0033] Figure 2-3 It is a performance curve diagram of the third-stage impeller of the MD550-50 multistage pump of the present invention;

[0034] Figure 3 It is the training model evaluation diagram of the present invention. DETAILED DESCRIPTION

[0035] The basic concept of the present invention is to obtain the operating condition factor data of the multi-stage centrifugal water injection pump, and calculate the wheel stage factor data of the two adjacent impellers, and build a model for predicting the operating condition of the multi-stage water injection pump based on the various factor data and result data that affect the operating condition of the water injection pump, and finally predict the water injection pump to be predicted by the constructed prediction model, thereby improving the accuracy and generalization ability of the prediction. Based on this concept, a method for evaluating the operating condition of a multi-stage centrifugal water injection pump of the present invention can be realized.

[0036] The present invention will be described in detail below with reference to the accompanying drawings and method embodiments.

[0037] Method Example:

[0038] The present invention provides a method for evaluating the operating conditions of a multi-stage centrifugal water injection pump, and the specific implementation steps include:

[0039] Step 1: Collect data representing the operating conditions of the water injection pump, including sensor data for representing the dynamic operating conditions of the multi-stage centrifugal water injection pump, historical record data sets for representing the historical operating conditions, and a census data table for representing the water quality of the water distribution room, and clean the collected data to remove invalid information;

[0040] The sensor collected data include: the output voltage, output current, given pressure, output frequency, motor speed, air cooler front temperature, air cooler middle temperature, air cooler rear temperature, front bearing temperature, rear bearing temperature, lubricating oil outlet temperature, motor electronic front temperature, motor electronic middle temperature, motor electronic rear temperature, cooling outlet temperature, cooling water inlet pressure, lubricating oil inlet pressure, pump lubricating oil inlet pressure, pump lubrication temperature, motor lubrication temperature, bearing front temperature, bearing rear temperature, pump inlet temperature, pump outlet temperature, liquid level, pump inlet pressure, pump outlet pressure, manifold pressure, pump diluted oil station pressure and motor diluted oil station pressure; water quality survey data include: PH index, S 2- , suspended solids, oil content in sewage, iron content ∑Fe, iron content Fe 3+ , any one of bacteria SRB, bacteria TGB and bacteria IB; the historical collection data is consistent with the data contained in the sensor collection data; the specific parameter values ​​are shown in Tables 1 and 2 below:

[0041] Table 1

[0042]

[0043]

[0044] Table 2

[0045]

[0046] Step 2: Group the sequence data obtained in step 1 by impeller ID, sort the sequence data of impellers with different powers, and determine the order of the impellers. Since the lengths of the sequence data of different impeller powers are different, the longest sequence data is used as the basis, and the remaining sequence data are filled with data without influence.

[0047] Step 3, calculating the impeller stage factors of the sequence data, including the power of the interaction between two adjacent impeller stages, the difference percentage of the interaction between the efficiency of two adjacent impeller stages, and the height difference of the interaction between the heads of two adjacent impeller stages;

[0048] Taking the MD550-50 multistage pump as an example, the performance curve of its first-stage impeller is as follows: Figure 2-1 As shown, the performance curve of the second stage impeller is as follows Figure 2-2 , the performance curve of the third-stage impeller is as follows Figure 2-3 , the performance parameters are shown in Table 3 below:

[0049] Table 3

[0050]

[0051] Calculate the interaction power P between two adjacent impellers i The calculation formula is:

[0052]

[0053] The difference between the efficiency interactions of two adjacent impeller stages η u Percentage calculation formula:

[0054]

[0055] The height difference H between the heads of two adjacent impellers u Calculation formula:

[0056] H i =H n-1 -H n

[0057] First-stage impeller: head H1 = 180m, efficiency η1 = 0.83, flow rate Q1 = 125.0L / s;

[0058] Secondary impeller: head H2 = 150m, efficiency η2 = 0.84, flow rate Q2 = 152.7L / s;

[0059] Three-stage impeller: head H3 = 135m, efficiency η3 = 0.80, flow rate Q3 = 163.9L / s;

[0060] In summary, the interaction power between two adjacent impeller stages is 2.22kW and 2.05kW respectively, the difference percentage of efficiency interaction between two adjacent impeller stages is 1.20% and -4.76% respectively, and the height difference of head interaction between two adjacent impeller stages is 30m and 15m respectively.

[0061] Step 4: Combine the collected water injection pump operating condition data and the calculated impeller stage factor data as training data to build a model for predicting the operating condition of a multi-stage water injection pump. Use the built model to predict the water injection pump to be predicted. Finally, each stage of the impeller will get a number of 0-1, and use this value to represent the operating state of a certain stage of the impeller at a certain moment. The specific process is as follows:

[0062] 1) Combine the round-level factor data in step 3 (time series data set at time t) to obtain the sequence The impeller long sequence data record is used as the embedding vector of the model. The impeller long sequence data includes the impeller ID and the impeller power. In the embedding and positioning layer, the impeller long sequence record is dimensionality-encoded.

[0063] The encoding consists of three multi-head attention modules and a feed-forward neural network, such as Figure 1 As shown in Figure 2, the input of the multi-head attention module is Q, K, V, and its calculation formula is as follows:

[0064]

[0065]

[0066]

[0067] W q , W k and W v They are respectively the linear mapping parameter matrices of query-key-value, which are learnable parameters. They are generated by interactive embedding with different projection matrices. The obtained attention value is processed by a residual connection and layer normalization as the input of the feedforward neural network. The output of the feedforward neural network is processed by a residual connection and layer normalization as the output of the Encoder.

[0068] 2) Q, K, and V are calculated from the input through different weight matrices. First, Q and K are multiplied and divided by the dimension Get an attention weight matrix. The weight scaling dot product attention operation is to avoid the inner product being too large and the gradient of the Softmax function being too small, even if the training process has a more stable gradient. The formula is as follows:

[0069]

[0070] 3) After the attention weight matrix is ​​further normalized by Softmax, the Sorces value can be interpreted as the similarity of M records in the sequence to other records. By adding the similarity matrix to the mask matrix, the records of subsequent time can be masked to prevent label leakage. The formula is as follows:

[0071] Sorces=Soft·m(h)·M

[0072] 4) Multiply the Sorces of the masked similarity matrix by V to obtain a weighted Attention result, which is equivalent to purifying the input, so that it can maintain a higher level of attention on records with high relevance and enhance its expressiveness. The formula is as follows:

[0073] Attention=Scores·V

[0074] 5) The multi-head attention mechanism is used to enable the model to pay attention to information from different representation subspaces, enhance generalization ability and expand the ability to pay attention to different positions. W is a learnable parameter, and its formula is as follows:

[0075] MultiHeadAttention=Concat(Head1,…ead n )

[0076] 6) The weight matrix generated by the attention layer is passed through the prediction layer activated by the Sigmoid function to obtain the final prediction result. The formula is as follows:

[0077] R=Sigmoid⊙

[0078] Step 5: According to Figure 3 The Sigmoid function training model result of MultiHeadAttention shows that at 5233s, the influence value of the second-stage impeller is 0.782, which is a large influence value. The influence value of the wheel stage set is also increased by the influence value of the second-stage impeller. The second-stage impeller can be replaced in advance.

[0079] The present invention obtains the operating data of a multi-stage centrifugal water injection pump and calculates its wheel stage factor data, thereby combining various data that affect the operating conditions of the water injection pump to construct a model for predicting the operating conditions of the multi-stage water injection pump. Finally, the water injection pump to be predicted is predicted through the constructed prediction model. The present invention can predict the operating condition warning of the pump at a certain moment and locate it on a certain stage of impeller through this method. Through early maintenance and replacement, the problem can be discovered and solved before the impeller fails, thereby reducing maintenance costs and failure rates and improving the reliability and stability of the multi-stage centrifugal pump.

Claims

1. A method for evaluating the operating conditions of a multi-stage centrifugal water injection pump, characterized in that: The steps include: 1) Obtaining the operating condition factor data of the water injection pump and preprocessing it to obtain the sequence data of the impellers at each level; calculating the wheel level factor data between two adjacent impellers according to the sequence data of the impellers at each level, wherein the wheel level factor data represents the data of the operating state of the two adjacent impellers that affect each other; using the calculated wheel level factor data and the processed sequence data of the impellers at each level and the corresponding operating condition result data as training data; 2) Construct a prediction model for predicting the operating conditions of impellers at each stage of the water injection pump based on the turbine factor data and sequence data of the water injection pump, train the prediction model using the training data to obtain a trained prediction model, and then use the trained prediction model to predict the water injection pump to be predicted, and obtain the operating conditions of each stage of impeller of the predicted water injection pump at a certain moment in the future.

2. The method for evaluating the operating condition of a multi-stage centrifugal water injection pump according to claim 1, characterized in that: The calculation of the impeller stage factor includes: the power of the interaction between two adjacent impeller stages, the difference percentage of the interaction between the efficiency of two adjacent impeller stages, and the height difference of the interaction between the heads of two adjacent impeller stages. The specific calculation formula is as follows: The power calculation formula for the interaction between two adjacent impellers is: The calculation formula for the difference percentage of the interaction between the efficiency of two adjacent impeller stages is: The calculation formula for the height difference of the alternating lift of two adjacent impeller stages is: H i =H n-1 -H n Among them, P i is the power exchanged between two adjacent impeller stages; η i H is the difference percentage of the interaction between the efficiency of two adjacent impellers; i H is the height difference between the heads of two adjacent impellers; n is the head of the nth stage impeller; H n-1 is the head of the n-1th stage impeller; η n is the efficiency of the nth stage impeller; η n-1 is the efficiency of the n-1th stage impeller.

3. The method for evaluating the operating condition of a multi-stage centrifugal water injection pump according to claim 1, characterized in that: The factor data for obtaining the operating conditions of the water injection pump include data collected by sensors: at least one of the output voltage, output current, given pressure, output frequency, motor speed, air cooler front temperature, air cooler middle temperature, air cooler rear temperature, front bearing temperature, rear bearing temperature, lubricating oil outlet temperature, motor electronic front temperature, motor electronic middle temperature, motor electronic rear temperature, cooling outlet temperature, cooling water inlet pressure, lubricating oil inlet pressure, pump lubricating oil inlet pressure, pump lubrication temperature, motor lubrication temperature, bearing front temperature, bearing rear temperature, pump inlet temperature, pump outlet temperature, liquid level, pump inlet pressure, pump outlet pressure, manifold pressure, pump diluted oil station pressure and motor diluted oil station pressure.

4. The method for evaluating the operating condition of a multi-stage centrifugal water injection pump according to claim 3, characterized in that: The factors for obtaining the operating conditions of the water injection pump also include water quality survey data: PH index, S 2- content, suspended solid content, oil content in sewage, Fe 3+ At least one of the content, iron content, the number of bacterial SRB, the number of bacterial TGB and the number of bacterial IB.

5. The method for evaluating the operating condition of a multi-stage centrifugal water injection pump according to claim 1, characterized in that: The prediction model for predicting the operating conditions of water injection pumps adopts a multi-head self-attention mechanism model.

6. The method for evaluating the operating condition of a multi-stage centrifugal water injection pump according to claim 1, characterized in that: Preprocessing includes cleaning the acquired operating condition factor data to remove invalid information.

7. The method for evaluating the operating condition of a multi-stage centrifugal water injection pump according to claim 6, characterized in that: The preprocessing also includes sorting the cleaned impeller sequence data according to certain data of the impeller.

8. The method for evaluating the operating condition of a multi-stage centrifugal water injection pump according to claim 7, characterized in that: The preprocessing also includes filling the remaining impeller sequence data based on the longest impeller sequence data in the sorted impeller sequence data.

9. The method for evaluating the operating condition of a multi-stage centrifugal water injection pump according to claim 7, characterized in that: The impeller sequence data is sorted by the power size of the impeller.

Images