Power consumption prediction method based on multi-task interpolation echo state network

Abstract

The invention provides a power consumption prediction method based on a multi-task interpolation echo state network. The power consumption prediction method comprises the following steps: acquiring to-be-detected sewage data; obtaining to-be-input data according to the to-be-detected sewage data by using a synthetic time sequence; the to-be-input data are input into a multi-task interpolation echo state network model, power consumption predicted values are obtained, the power consumption predicted values are a pumping energy consumption value and an aeration energy consumption value, and the multi-task interpolation echo state network model comprises an input layer, a storage pool and an output layer. The problem that in the prior art, pumping energy consumption and aeration energy consumption in the sewage treatment process are difficult to accurately and efficiently predict is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of power consumption prediction, and in particular to a power consumption prediction method based on a multi-task interpolation echo state network. Background Art

[0002] my country attaches great importance to ecological environmental protection and energy conservation and carbon reduction. Urban sewage treatment plants have achieved extensive development, but they are facing the problem of high energy consumption and high operating costs. The power consumption of sewage treatment plants is mainly composed of aeration energy consumption and pumping energy consumption. Pumping energy consumption is mainly caused by the water inlet lift pump and the sludge return pump. Aeration energy consumption mainly refers to the power consumed in the process of the blower introducing dissolved oxygen into the aeration tank. Accurate prediction of pumping energy consumption and aeration energy consumption can enable sewage treatment plants to take corresponding measures in a targeted manner for the pumping link or aeration link, thereby reducing energy consumption, and thus provide a basis for decision-making to optimize and control the sewage treatment process, and promote the efficient and stable operation of sewage treatment plants and the sustainable development of ecological protection. At present, the energy consumption of sewage treatment plants is affected by different biological and environmental factors, making the establishment of prediction models complicated and challenging. It is difficult to achieve accurate and efficient prediction of pumping energy consumption and aeration energy consumption for sewage treatment processes. Summary of the invention

[0003] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a power consumption prediction method based on a multi-task interpolation echo state network. The present invention solves the problem in the prior art that it is difficult to accurately and efficiently predict the pumping energy consumption and aeration energy consumption of the sewage treatment process.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A method for predicting power consumption based on a multi-task interpolation echo state network, comprising:

[0006] Obtaining the sewage data to be tested;

[0007] Using the synthetic time series, the data to be input is obtained according to the sewage data to be tested;

[0008] The data to be input is input into a multi-task interpolation echo state network model to obtain a power consumption prediction value, wherein the power consumption prediction value is a pumping energy consumption value and an aeration energy consumption value, and the model of the multi-task interpolation echo state network includes an input layer, a reserve pool and an output layer.

[0009] Preferably, the sewage data to be tested includes:

[0010] Outlet COD concentration, effluent total phosphorus concentration, inlet pH, effluent pH, and DO concentration.

[0011] Preferably, the step of obtaining the sewage data to be tested includes:

[0012] Get labeled and unlabeled data;

[0013] The interpolation method based on the distance coefficient uses unlabeled data to obtain virtual interpolation points;

[0014] The sewage data to be tested is obtained according to the virtual interpolation points and the actual interpolation points corresponding to the labeled data.

[0015] Preferably, the distance coefficient-based interpolation method uses unlabeled data to obtain virtual interpolation points, including:

[0016] Determine the label data at the first moment and the label data at the second moment;

[0017] Obtaining an average value of unlabeled data between the first moment and the second moment according to the label data at the first moment and the label data at the second moment;

[0018] A virtual interpolation point is determined according to the average value of the unlabeled data.

[0019] Preferably, the method of using a synthetic time series to obtain the data to be input according to the sewage data to be tested includes:

[0020] Obtaining a labeled time series according to actual interpolation points corresponding to the labeled data;

[0021] Obtaining a virtual time series according to the virtual interpolation points;

[0022] Obtaining a synthetic time series according to the marked time series and the virtual time series;

[0023] The data to be input is obtained according to the synthetic time series.

[0024] Preferably, the network structure of the model of the multi-task interpolation echo state network is 5-(k1N+k2N)-2, where N is an initialized positive integer, usually 10, and k j ≥1 and is a positive integer, indicating the number of sub-reserve pools owned by each ESN module, k j N represents the reserve tank size for the pumping energy consumption of task 1 and the aeration energy consumption of task 2, where j = 1, 2.

[0025] Preferably, it also includes:

[0026] Obtaining input data and power consumption forecast values;

[0027] The Laida criterion is used to detect abnormal values ​​of the input data and the power consumption forecast value, and if abnormal values ​​exist, the abnormal values ​​are replaced.

[0028] Preferably, the expression of the model of the multi-task interpolation echo state network is:

[0029]

[0030] Among them, s is the original data, is the normalized data, s min and max are the minimum and maximum values ​​of the input data before normalization.

[0031] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0032] The present invention provides a method for predicting power consumption based on a multi-task interpolation echo state network, comprising: obtaining sewage data to be tested; obtaining data to be input according to the sewage data to be tested by using a synthetic time series; inputting the data to be input into a multi-task interpolation echo state network model to obtain a power consumption prediction value, wherein the power consumption prediction value is a pumping energy consumption value and an aeration energy consumption value, and the model of the multi-task interpolation echo state network comprises an input layer, a reserve pool and an output layer. The present invention realizes simultaneous and accurate prediction of pumping energy consumption and aeration energy consumption by preprocessing multi-rate sampling data and paralleling reserve pools, which is conducive to ensuring the optimized operation and energy saving and consumption reduction of sewage treatment plants. The present invention aims at the multi-rate phenomenon existing in the data collected by the sewage treatment plant, designs an interpolation method based on a distance coefficient to construct virtual points, and enriches the data samples; secondly, in view of the problem that the correlation of different energy consumption parameter prediction tasks is ignored, a multi-task learning mechanism is established, and relevant information is shared through parallel reserve pools. Then, an adaptive dynamic adjustment algorithm of the reserve pool is designed to adapt to changes in the external working environment. The established prediction model can simultaneously predict the two main energy-consuming links, pumping energy consumption and aeration energy consumption, alleviating the data waste caused by the multi-rate phenomenon and utilizing the useful information ignored by single-task learning. It can also adjust the size of the reserve pool according to dynamically changing working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0034] Figure 1 A flow chart of a method for predicting power consumption based on a multi-task interpolation echo state network according to the present invention;

[0035] Figure 2 This is a test result diagram of the prediction method for the pumping energy consumption task of the present invention;

[0036] Figure 3 This is a test result diagram of the prediction method for aeration tasks of the present invention;

[0037] Figure 4 It is a test error diagram of the prediction method of the pumping energy consumption task of the present invention;

[0038] Figure 5 It is a test error diagram of the prediction method of the aeration energy consumption task of the present invention;

[0039] Figure 6 Detailed schematic diagram of the prediction method for aeration energy consumption task of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] like Figure 1 As shown, the present invention provides a method for predicting power consumption based on a multi-task interpolation echo state network, comprising:

[0043] Step 100: Obtaining the sewage data to be tested;

[0044] Step 200: using the synthetic time series to obtain the input data according to the sewage data to be tested;

[0045] Step 300: Input the data to be input into the multi-task interpolation echo state network model to obtain the power consumption prediction value, wherein the power consumption prediction value is the pumping energy consumption value and the aeration energy consumption value, and the model of the multi-task interpolation echo state network includes an input layer, a reserve pool and an output layer.

[0046] Furthermore, the sewage data to be tested includes:

[0047] Outlet COD concentration, effluent total phosphorus concentration, inlet pH, effluent pH, and DO concentration.

[0048] Furthermore, it also includes:

[0049] Obtaining input data and power consumption forecast values;

[0050] The Laida criterion is used to detect abnormal values ​​of the input data and the power consumption forecast value, and if abnormal values ​​exist, the abnormal values ​​are replaced.

[0051] Specifically, Figure 6 As shown in the figure, the variables with strong correlation with pumping energy consumption and aeration energy consumption in the data collected by the sewage treatment plant are selected as input, including effluent COD concentration, effluent total phosphorus concentration, inlet pH, effluent pH, and DO concentration. Pumping energy consumption and aeration energy consumption are selected as output variables. In order to eliminate the influence of factors such as sensor calibration and failure, the Laida criterion is used to replace the detected abnormal values ​​with the normal values ​​of the previous moment. The specific formula is as follows:

[0052] s o ={s∣s>(μ+3σ)∪s<(μ-3σ)} (1)

[0053] where s o Indicates outliers, s indicates original data, μ and σ indicate the mean and standard deviation of the data, respectively. In order to remove the impact of different variable dimensions on the prediction, the following formula is used to normalize the input and output variables to the interval [0,1].

[0054]

[0055] in represents the normalized data, s min and max Respectively represent the minimum and maximum values ​​of the batch of data before normalization. The processed input data is recorded as u(t) = [u1(t), u2(t), u3(t), u4(t), u5(t)], and the output data is recorded as [y1(t), y2(t)]. Among them, u1(t), u2(t), u3(t), u4(t), u5(t) represent the effluent COD concentration, effluent total phosphorus concentration, influent pH, effluent pH, and DO concentration after pretreatment. y1(t) and y2(t) represent the pumping energy consumption and aeration energy consumption after pretreatment.

[0056] Furthermore, the obtaining of the sewage data to be tested includes:

[0057] Get labeled and unlabeled data;

[0058] The interpolation method based on the distance coefficient uses unlabeled data to obtain virtual interpolation points;

[0059] The sewage data to be tested is obtained according to the virtual interpolation points and the actual interpolation points corresponding to the labeled data.

[0060] Furthermore, the distance coefficient-based interpolation method uses unlabeled data to obtain virtual interpolation points, including:

[0061] Determine the label data at the first moment and the label data at the second moment;

[0062] Obtaining an average value of unlabeled data between the first moment and the second moment according to the label data at the first moment and the label data at the second moment;

[0063] A virtual interpolation point is determined according to the average value of the unlabeled data.

[0064] Specifically, an interpolation method for multi-rate sampling data based on distance coefficients. The data collected from the sewage treatment plant has the characteristics of multi-rate. The multi-rate phenomenon refers to the fact that easily measurable variables such as pH, DO concentration and other input variables are updated every τ minutes, while target variables such as pumping energy consumption and aeration energy consumption are updated every Kτ (K>1) minutes. This difference in update frequency leads to part of the data being labeled and part of the data being unlabeled, that is, there are only values ​​of input variables but no corresponding values ​​of output variables. The mathematical formula is as follows, where u (i) represents input data, i represents the type of sampling rate, Represents the input data sampled at a high rate, and its sampling time is rτ, that is, the time can be τ, 2τ, …, LKτ. Represents the input data sampled at a low rate, and its sampling time is rKτ, that is, Kτ, 2Kτ, ..., LKτ can be obtained. (rKτ) It represents the output data of low-rate sampling, and its sampling time is rKτ, that is, Kτ, 2Kτ, …, LKτ can be obtained, where L represents the sample length, and K represents how many times the high sampling rate is the low sampling rate.

[0065]

[0066] y={y (rKτ)} r＝1,2,…,L (4)

[0067] Therefore, the present invention designs an interpolation method based on distance coefficient, which can use unlabeled data to generate new virtual interpolation points and increase the number of samples in the model. The input data of the newly inserted auxiliary prediction point is the average value of the unlabeled data between each two adjacent labeled data, that is, between time jKτ and (j+1)Kτ. time.

[0068] Furthermore, the method of using the synthetic time series to obtain the input data according to the sewage data to be tested includes:

[0069] Obtaining a labeled time series according to actual interpolation points corresponding to the labeled data;

[0070] Obtaining a virtual time series according to the virtual interpolation points;

[0071] Obtaining a synthetic time series according to the marked time series and the virtual time series;

[0072] The data to be input is obtained according to the synthetic time series.

[0073] Specifically,

[0074] In order to solve Calculate input data and historical labeled data Euclidean distance Solve the weight according to the distance coefficient The formula is as follows. Among them, Indicates virtual input data With labeled time series The Euclidean distance of Represents the weight of the virtual output data.

[0075]

[0076] Virtual output data can be obtained as follows.

[0077]

[0078] The same method is also used to generate. At this time, a virtual sequence S2 is obtained, which can be expressed as follows.

[0079]

[0080] The synthetic time series S consists of the marked time series S1 and the virtual time series S2, which can be expressed as follows, where Represents the input data of the synthetic time series S and output data The sampling time is From Kτ, …get LKτ.

[0081]

[0082] Furthermore, the network structure of the multi-task interpolation echo state network model is 5-(k1N+k2N)-2, where N is an initialized positive integer, usually 10. j ≥1 and is a positive integer, indicating the number of sub-reserve pools owned by each ESN module, k j N represents the reserve tank size for the pumping energy consumption of task 1 and the aeration energy consumption of task 2, where j = 1, 2.

[0083] The multi-task interpolation echo state network is divided into three parts, namely the input layer, the reserve pool and the output layer. The input layer has five input variables: effluent COD concentration, effluent total phosphorus concentration, inlet pH, effluent pH, and DO concentration; the reserve pool is composed of two ESN modules in parallel, each ESN module contains several sub-reserve pools with the same nodes, has multi-task learning capabilities, and captures and memorizes input sequence information; the output layer is the predicted pumping energy consumption and aeration energy consumption, and only uses the real-time values ​​to evaluate the prediction accuracy. Therefore, the neural network model has a total of 5 input variables and 2 output variables, and the network structure of the multi-task interpolation echo state network is determined to be 5-(k1N+k2N)-2, where N is an initialized positive integer, usually 10. k j ≥1(j=1,2) and is a positive integer, indicating the number of sub-reservation pools owned by each ESN module, k j N(j=1,2) represents the reserve tank size for the pumping energy consumption of task 1 and the aeration energy consumption of task 2 respectively.

[0084] For the dth ESN module, assume that Sub-reserve pools, each sub-reserve pool has N nodes. k (k=1,2,…,k d ) sub-reservoir, the input weight matrix is The sub-reserve pool weight matrix is Based on the singular value decomposition technique, using the predefined diagonal matrix and two randomly constructed orthogonal matrices and generate and are the output weights of task 1 pumping energy consumption and task 2 aeration energy consumption, respectively. At time step t, the internal state matrix And sub-reservoir output It can be calculated by the following formula

[0085]

[0086] Where f(·) is the activation function, usually the hyperbolic tangent function tanh(), and u(t) is the input signal. From the perspective of the dth ESN module, its input weight is Reserve Pool Weight It can be expressed as Reserve pool status X d (t+1) and the corresponding output O d (t+1) is updated as follows. Among them, is the dth module kThe output weight of each sub-reserve pool, is the dth module k The status of each sub-reserve pool.

[0087]

[0088] Let X(t) = [X1(t); X2(t)], the network output of task p (p = 1, 2) is O p (t) is expressed as

[0089]

[0090] in is the output weight of the reservoir of output node p. Formula (16) can be restated as

[0091]

[0092] Solved by gradient descent method, as shown in formula (18). p (t) is the modeling error of training, E p (t) is the loss function, is the corresponding supervisory signal matrix, y1 and y2 are the teacher signals in the training phase, i.e., the pumping energy consumption and aeration energy consumption to be predicted.

[0093]

[0094] e p (t) = Y p (t)-O p (t) (19)

[0095]

[0096] Where η is the learning rate, is the corresponding gradient. L1 is the length of the training sample.

[0097] Among them, formulas 12-21 together constitute a multi-task interpolation echo state network model.

[0098] Furthermore, the network is trained:

[0099] Initialize two ESN modules of the two tasks of the network. Each module has a sub-reservoir with N=10 nodes. Randomly generate the input weight matrix and internal connection matrix of each sub-reservoir.

[0100] Enter a training data sample To ESN module 1 and module 2, use the multi-rate data interpolation method to process the data and obtain and Collection Status renew Calculate the training modeling error e p (t).

[0101] Calculate the comprehensive modeling error E(t) at each time step, for the dth in the dth module k In order to evaluate its learning ability for each task, the contribution As follows. Among them, represents the dth module of the pth task k The output weight of each sub-reserve pool, Indicates the dth module k The status of each sub-reserve pool.

[0102]

[0103] In order to measure the prediction ability of the dth module for task p, the cumulative contribution is introduced as follows.

[0104]

[0105] In order to obtain a compact network structure, the design principle of a dynamic structure adjustment strategy is given.

[0106] Growth mechanism: To compensate for large modeling errors, new sub-reserves are added. The network growth condition is expressed as follows.

[0107] E(t)>E a and k d N≤N max (25)

[0108] Where E a and N max is a predefined positive value, E a Represents the maximum modeling error in practical applications, determined according to the actual task, N max Represents the maximum size of a single module pool, usually set to 200. Then, select the module α with the smallest cumulative contribution to task 1 and task 2 and the sub-pool α with the largest contribution h .

[0109]

[0110] Next, the new sub-reservoir is initialized and inserted into module α. Assigning input weights to the new neurons Internal weight ΔW new (t) and output weights As follows, at this time k d =k d +1.

[0111]

[0112] Pruning mechanism: In order to obtain a concise network, a node pruning algorithm is designed by eliminating sub-reservoir pools with smaller contributions. First, sort the sub-reservoir pools from high to low according to their contribution. Then, select the sub-reservoir pool with the smallest contribution to Task 1 and Task 2 and name it β h In ESN module β, the sub-reserve with the second smallest contribution is denoted as β h' If the following conditions are met,

[0113]

[0114] Where γ>0 is the deletion threshold, and γ=0.05 is set. h and β h' The weights are adjusted as follows.

[0115]

[0116] Stable strategy: Once the following conditions are met, the network structure remains unchanged under a fixed reserve pool.

[0117]

[0118] Repeat steps to update Update every time The number of training times i = i + 1, until the maximum number of training times i = i is reached max =L1, then stop training.

[0119] After training, the output of the neural network is denormalized into the predicted estimated values ​​of pumping energy consumption and aeration energy consumption using the following formula:

[0120]

[0121] Going a step further, test the neural network:

[0122] After the test sample data is normalized and outliers are removed according to formulas (1)-(2), the interpolated synthetic sequence is obtained according to formulas (5)-(11) as the input of the trained multi-task interpolation echo state network. The output of the multi-task interpolation echo state network is denormalized according to formula (31) to obtain the estimated measurement values ​​of pumping energy consumption and aeration energy consumption.

[0123] More specifically, the water quality and energy consumption analysis data of a sewage treatment plant in Beijing in 2023 were used to obtain a total of 1,000 groups of data samples, including 500 groups of labeled data samples and 500 groups of unlabeled data samples. Labeled samples refer to the corresponding output data when the input data is updated, and unlabeled data refers to only input data without corresponding output data. The first 700 groups of data are selected as training samples, and the last 300 groups of data are selected as test samples. The first 700 groups of data contain 350 groups of labeled samples, and the last 300 groups of data contain 150 groups of labeled samples. Only labeled samples are used to calculate the test error. The main steps are as follows:

[0124] Step 1: Input variable selection and data preprocessing of soft sensor model.

[0125] Select effluent COD concentration, effluent total phosphorus concentration, inlet pH, effluent pH, and DO concentration as input variables, and select pumping energy consumption and aeration energy consumption as output variables. Use formula (1) to remove outliers, and use formula (2) to normalize the input and output variables to the [0,1] interval.

[0126] Step 2: Initialize the multi-task interpolation echo state network structure and parameters.

[0127] First, determine the multi-task interpolation echo state network structure as 5-(k1N+k2N)-2; second, initialize the input weight of each sub-reservoir and internal weights The value is controlled between [-0.5, 0.5], the maximum singular value Finally, the network updates its state according to formula (12) and (14), and calculates its output according to formula (16).

[0128] Step 3: Train the neural network.

[0129] Step 3.1: Initialize two ESN modules for two tasks of the network. Each module has one sub-reservoir with N=10 nodes. Randomly generate the input weight matrix and internal connection matrix of each sub-reservoir.

[0130] Step 3.2: Input a training data sample To ESN module 1 and module 2, use the multi-rate data interpolation method to process the data and obtain and The state of each sub-reserve pool is collected through formula (12) and formula (14) respectively and the status X of each module d (t+1), updated by formula (18) The training modeling error e is calculated by formula (19): p (t).

[0131] Step 3.3: Calculate the comprehensive modeling error E(t) for each time step by using formula (22), and calculate the contribution of each sub-reservoir and the cumulative contribution of each module by using formulas (23) and (24). If E(t) satisfies formula (25), then add a new sub-reservoir under the specified module according to formulas (26) and (27). If formula (28) is satisfied, the sub-reserve pool β is adjusted according to formula (29) h and β h' The weight of .

[0132] Step 3.4: Repeat steps 3.2-3.3 to update Update every time The number of training times i = i + 1, until the maximum number of training times i = i is reached max =L1, then stop training.

[0133] Step 3.5: After training, the output of the neural network is denormalized into the predicted estimated values ​​of pumping energy consumption and aeration energy consumption using formula (31).

[0134] Step 4: Test the neural network.

[0135] The trained network is used to test the pumping energy consumption and aeration energy consumption. The test results of pumping energy consumption are as follows: Figure 2 As shown, Figure 2 X-axis: time step, Figure 2 Y axis: pumping energy consumption value, unit is kWh; the solid line is the expected pumping energy consumption value, and the dotted line is the test pumping energy consumption output value; the test results of aeration energy consumption are as follows Figure 3 As shown, Figure 3 X-axis: time step, Figure 3 Y axis: aeration energy consumption value, unit is kWh; the solid line is the expected aeration energy consumption value, and the dotted line is the test aeration energy consumption output value; the error between the actual output of pumping energy consumption and the test output is as follows Figure 4 As shown, Figure 4 X-axis: time step, Figure 4 Y axis: pumping energy consumption error, in kWh; the error between the actual output and the test output of aeration energy consumption is as follows Figure 5 As shown, Figure 5 X-axis: time step, Figure 5 Y-axis: aeration energy consumption error, in kWh; The experimental results show the effectiveness of the prediction model for pumping energy consumption and aeration energy consumption based on the multi-task interpolation echo state network.

[0136] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0137] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for predicting power consumption based on a multi-task interpolation echo state network, characterized in that: include: Obtaining the sewage data to be tested; Using the synthetic time series, the data to be input is obtained according to the sewage data to be tested; The data to be input is input into a multi-task interpolation echo state network model to obtain a power consumption prediction value, wherein the power consumption prediction value is a pumping energy consumption value and an aeration energy consumption value, and the model of the multi-task interpolation echo state network includes an input layer, a reserve pool and an output layer.

2. A method for predicting power consumption based on a multi-task interpolation echo state network according to claim 1, characterized in that: The sewage data to be tested include: Outlet COD concentration, effluent total phosphorus concentration, inlet pH, effluent pH, and DO concentration.

3. The method for predicting power consumption based on multi-task interpolation echo state network according to claim 1, characterized in that: The step of obtaining the sewage data to be tested includes: Get labeled and unlabeled data; The interpolation method based on the distance coefficient uses unlabeled data to obtain virtual interpolation points; The sewage data to be tested is obtained according to the virtual interpolation points and the actual interpolation points corresponding to the labeled data.

4. The method for predicting power consumption based on multi-task interpolation echo state network according to claim 3, characterized in that: The distance coefficient-based interpolation method uses unlabeled data to obtain virtual interpolation points, including: Determine the label data at the first moment and the label data at the second moment; Obtaining an average value of unlabeled data between the first moment and the second moment according to the label data at the first moment and the label data at the second moment; A virtual interpolation point is determined according to the average value of the unlabeled data.

5. The method for predicting power consumption based on multi-task interpolation echo state network according to claim 3, characterized in that: The method of using the synthetic time series to obtain the input data according to the sewage data to be tested includes: Obtaining a labeled time series according to actual interpolation points corresponding to the labeled data; Obtaining a virtual time series according to the virtual interpolation points; Obtaining a synthetic time series according to the marked time series and the virtual time series; The data to be input is obtained according to the synthetic time series.

6. The method for predicting power consumption based on multi-task interpolation echo state network according to claim 1, characterized in that: The network structure of the multi-task interpolation echo state network model is 5-(k1N+k2N)-2, where N is an initialized positive integer, usually 10. j ≥1 and is a positive integer, indicating the number of sub-reserve pools owned by each ESN module, k j N represents the reserve tank size for the pumping energy consumption of task 1 and the aeration energy consumption of task 2, where j = 1, 2.

7. The method for predicting power consumption based on multi-task interpolation echo state network according to claim 1, characterized in that: Also includes: Obtaining input data and power consumption forecast values; The Laida criterion is used to detect abnormal values ​​of the input data and the power consumption forecast value, and if abnormal values ​​exist, the abnormal values ​​are replaced.

8. The method for predicting power consumption based on multi-task interpolation echo state network according to claim 1, characterized in that: The expression of the model of the multi-task interpolation echo state network is: Among them, s is the original data, is the normalized data, s min and max are the minimum and maximum values ​​of the input data before normalization.

Images